Using SSD caching in servers. Adaptec (Microsemi) and LSI (BROADCOM) controllers

« Best practice» For many corporate customers, it is the use of hybrid SSD/HDD storage systems. This solution allows you to take advantage of both types of media - large capacity HDD and high performance SSD in IOPS (input/output operations per second) - but at the same time remains economically attractive.

In a hybrid SSD/HDD storage system, the main capacity is represented by inexpensive hard drives, and a small pool for “hot”, frequently used data is provided by flash memory. In a rationally designed hybrid storage system, with a small number of SSD drives, significant acceleration of operations with the main data storage pool is achieved.

IMPLEMENTATION OF HYBRID STORAGE

In practice, two main acceleration methods are used - data caching and multi-level storage (tiering). Both use the concept of hot data to improve I/O performance, but in reality they are completely different approaches.

With caching, one or more SSD drives serve as a cache for a virtual storage pool where the primary storage is on hard drives. In this case, SSDs do not provide additional capacity - they are an invisible “layer” for applications that increases I/O performance. Information is always transferred to the main storage pool, but “hot” data is also copied to cache memory (on the SSD). Subsequent accesses to this or nearby data use cache memory instead of the main storage pool, resulting in significant performance gains.

With multi-level storage, data is sorted accordingly and placed on the SSD or HDD level (there can be more than two levels): “hot” ones are sent to flash memory, and less frequently used ones are sent to hard drives.

WHAT'S BETTER?

Multi-level storage does not imply data redundancy, so the implementation of RAID in this case becomes more complex - the purchase of additional SSDs is required. The very act of sorting data and distributing it into levels has a negative impact on performance. Such systems must manage data that turns from “hot” to “cold” over time. Due to the lack of redundancy, frequently used data should be moved to the main pool as soon as it becomes less useful. These background processes consume IOPS and impact I/O performance during these moves. Multi-level storage operates most efficiently in cases where the corresponding algorithms are adapted to the requirements and objectives of the customer. Achieving ideal performance requires constant monitoring and adjustment of algorithms.

Unlike complex multi-level storage, caching on SSDs is easier to implement in existing storage systems. Hybrid storage systems with SSD caching do not require additional administration, and the application treats such a system in the same way as any other network storage system, only it works much faster. The implementation of RAID and data protection are similar, and buy for this additional SSDs not required.

Copies of data are placed on the SSD, so there is no need to move it in the background to the main storage pool. There will be no associated costs affecting productivity. Caching on SSDs, however, needs to be adjusted to specific enterprise applications, but the simplicity of the caching system means that the administration of the system will be significantly less complex than that of a comparable storage system with tiered storage.

The costs associated with installing storage systems with tiered storage and its maintenance will be justified only in very large organizations that can afford both the installation of rack-mount SSD modules to organize a dedicated pool of flash memory, and an increase in the staff of system administrators to manage storage systems. For most companies that do not have extremely large storage pools, SSD caching is the preferred option for accelerating storage systems.

RECORDING PROBLEM

Being more powerful, SSD drives have certain limitations on data recording, and this must be kept in mind when choosing a storage acceleration method. Although data stored on flash drives can be read infinite number times, their cells allow a limited number of write cycles. This problem is compounded by the need to delete the entire block even when writing smaller amounts of data. To solve this problem, modern flash memory controllers use distributed writing methods, caching write operations, and background “garbage collection”. However, writing to an SSD remains a more complex operation than reading. Writing to the same cells too often can cause flash memory to degrade quickly.

If in a client system write operations on an SSD can be distributed in such a way that each individual block of media will be overwritten quite rarely, then in a hybrid storage system the SSD level is actively used to store “hot” data of the entire disk pool. With caching and tiered storage, SSD operations will become very intensive, and the benefits of wear-out prevention algorithms will be negated. This means that in both cases (caching and storage tiering), the SSD tier is best used to speed up reads rather than reads and writes.

IMPLEMENTATION OF CACHING ON SSD

In a system with SSD caching, I/O operations are performed in the usual way: first reads and writes are performed on the HDD. If this operation triggers caching, data is also copied from the HDD to the SSD. Then, any subsequent read operation of the same logical block reads it directly from the SSD, increasing overall performance and reducing response time. The SSD layer acts as an invisible I/O accelerator, and if any SSD failure occurs, the data will still be available in the main storage pool protected by RAID.

FILLING THE CACHE MEMORY

The cache, like the main storage capacity, is divided into groups of sectors of equal size. Each group is called a cache block, and each block consists of subblocks. The cache block size can be configured for a specific application, such as a DBMS or Web server.

Reading data from the HDD and writing it to the SSD is called filling the cache memory. This background operation usually occurs after the main read or write operation. Since the purpose of the cache is to store frequently used data, not every I/O operation should fill it, but only one for which the counter threshold value is exceeded. Typically, fill counters are used for reading and writing.

Thus, read and write counters are associated with each block of the main storage capacity. When an application reads data from a cache block, its read counter is incremented. If there is no data in the cache memory and the read counter value is greater than or equal to the read fill value, then a cache fill operation is performed in parallel with the main read operation (the data is cached). If the data is already in the cache, it is read from the SSD, and the filling operation is not performed. If the read counter is less than the threshold, it is incremented and the padding operation is not performed. For the write operation the scenario is the same. It is explained in more detail in the illustrations on the previous spread.

What happens to the contents of the cache after it is “warmed up”? If the SSD has free space, the cache continues to fill with “hot” data. When the SSD capacity is exhausted, the Least Recently Used (LRU) data rewrite algorithm is applied, that is, new “hot” data is written in place of the last data in the cache memory.

If the amount of hot data exceeds the capacity of the SSD, the percentage of data read from the cache memory decreases, and performance decreases accordingly. In addition, the smaller the SSD capacity (and the larger the volume of hot data), the more intense the exchange of “hot” data. As a result, the SSD will wear out faster.

Qsan experts recommend using Intel SSD DC S3500 drives. Thus, a 480 GB SSD has a mean time between failures (MTBF) of 2 million hours. As for performance, the typical latency for these drives is 50 ms, the maximum read latency is 500 ms (99.9% of the time), and the performance at random reading in blocks of 4 KB reaches 75 thousand IOPS, when writing - 11 thousand IOPS. This is a good option for SSD caching.

READ-WRITE CACHING

A read operation when there is no data in the cache occurs as follows:

Data is read from the HDD.
The SSD filling operation is in progress.

Read operation when there is data in cache:

The application issues a request to read data.
Data is read from the SSD.
The requested data is returned to the application.
If the SSD fails, the data is read from the HDD.

Application actions when writing data:

The application issues a request to record data.
Data is written to HDD.
The status of the operation is returned to the application.
The operation of filling the cache memory on the SSD is in progress.

SSD CACHE SETTING

To ensure that your application uses the SSD cache as efficiently as possible, it can be configured. The main parameters are the size of the cache memory block, the filling thresholds for reading and writing.

Block size. A large cache block size is suitable for applications that frequently access adjacent (physically located) data. This is called high locality of calls. Increasing the block size also speeds up the filling of the cache memory on the SSD - the “warming up” of the cache is accelerated, after which applications with high locality of access will demonstrate very high performance. However, increasing the block size generates excess I/O traffic and increases response time, especially for cache-missing data.

A smaller block size is good for applications with less localized data, that is, when data is accessed primarily randomly. The cache memory on an SSD will “warm up” more slowly, but the more blocks there are, the greater the likelihood of the necessary data getting into the cache, especially data with low access locality. With smaller blocks, cache utilization is lower, but the associated losses are also lower, so there is less performance impact on a “miss” when the required data is not in the cache.

Fill threshold value. The cache full threshold is the number of data accesses after which the corresponding block is copied to the SSD cache. With a large value, only frequently used data is cached and data exchange in the cache is reduced, but the “warm-up” time of the cache memory increases and the efficiency of its use increases. With a lower value, the cache memory warms up faster, but it may become overfilled. For most applications, a threshold of 2 is sufficient. Write padding is useful when the data being written is soon read again. This often happens in file systems. Other applications, such as databases, do not have this feature, so it is sometimes better to disable write filling for them altogether.

As you can see, increasing or decreasing each parameter has its positive and negative consequences. It is very important to understand the “locality” of the application. In addition, it is useful to test the system under real loads and see at what parameters it performs best.

EXAMPLE OF SSD-BASED CACHE CONFIGURATION

The test simulated a typical I/O situation (random read 90% + write 10%) to determine the gain that comes from using the SSD cache. During testing, the AegisSAN Q500 system was used in the following configuration:

HDD: Seagate Constellation ES, ST1000NM0011, 1 TB, SATA 6 Gb/s (x8);
SSD: Intel SSD DC 3500, SSDSC2BB480G4, 480 GB, SATA 6 Gb/s (x5);
RAID group: RAID 5;
I/O type: Database Service (8 KB);
I/O mode: 8 KB blocks.

The “warm-up” time is calculated using the following formula:

T = (C × P) / (I × S × D),

where T is the “warm-up” time, I is the average performance in IOPS of one HDD during random reading, S is the size of the I/O block, D is the number of HDDs, C is the total capacity of all SSDs, P is the threshold value for filling the cache memory during reading or recordings. In practice, the cache may take longer to warm up.

For this configuration it will be:

T = (2 TB × 2) / (244 × 8 KB × 8) = 275,036.33 sec = 76.40 hours.

Without SSD caching, average performance was 962 IOPS. When caching was enabled, it increased to 1942 IOPS, that is, the improvement after “warming up” the cache turned out to be twofold - 102%. According to the calculation formula, the warm-up time is 76.4 hours; in the test, after 75 hours, IOPS performance reached its maximum value and remained stable after that.

CONCLUSION

The concept of accelerating hybrid storage systems implements the idea of increasing the performance of the entire system through fast access to “hot” data. Taking into account hardware and administration costs, it can be argued that in general, data caching on an SSD is a best way Gain the benefits of high performance when using flash storage systems without sacrificing data storage reliability.

Bartek Mitnik- Sales Director of Qsan Technology in the EMEA region.

Traditional ways to speed up a PC include upgrading or overclocking the processor and video card, as well as expanding the amount of RAM. At the same time, an equally important part of the computer—the disk subsystem—is often left unattended. Its speed affects the performance of a PC no less than a powerful CPU or a couple of extra gigabytes of RAM - after all, if the hard drive “slows down”, all super-fast components will be forced to wait patiently for it, and with them the user.

Until recently, there were actually three ways to speed up the disk subsystem: replacing the HDD with a faster model, building a RAID array, or switching to an SSD, and each of these approaches has its drawbacks. With the exit Intel chipset Z68, the processor giant offered PC users another way - intermediate caching of data that the system is actively working with on a small SSD. The technology is called Smart Response. By the way, it was not for nothing that we clarified that Intel proposed this technology specifically for PCs: in fact, SSD caching was proposed back in 2009 by Adaptec for high-level server heavy-load RAID arrays (Adaptec MaxIQ), and then similar solutions were presented by others enterprise storage market players. What is typical is that just as competitors followed the pioneer in the corporate segment, the same thing happened in the consumer segment, and today we will look at one of the analogues of Intel Smart Response using an example solid state drive OCZ Synapse Cache. The advantage of such hybrid systems over hard drives is obvious: frequently used data is transferred to a radically faster SSD. And relative to independent solid-state drives, this model of use is more profitable due to the fact that you do not have to sacrifice capacity - after all, the cost of a gigabyte for SSDs and HDDs still differs by an order of magnitude.

Test participants

The “benchmark” for assessing the performance of a traditional hard drive will be Western Digital VelociRaptor WD1500HLHX.

WD VelociRaptor

This is the youngest 150 GB model from the latest generation of “raptors”, characterized by support for SATA 6 Gb/s and a 32 MB buffer. Like the entire WD family of “predators”, the key feature of this disk— spindle speed 10,000 rpm and form factor 2.5" (although the HDD is physically installed on a large 3.5-inch radiator). Due to the higher rotation speed and smaller size plates achieve an increase in linear speed and, in particular, a decrease in access time compared to traditional 7200 rpm models, not to mention the slower “green” series. The result is the fastest SATA drive available on the market for PCs and workstations.

The second participant in testing will be a RAID-0 array of two VelociRaptors - let's see what dividends are brought by simply purchasing a second drive to an existing one and assembling the array on a chipset controller.

The third device in the test is an OCZ Vertex 3 Max IOPS SSD with a capacity of 120 GB.

Today, this is, in fact, the fastest solid-state drive among devices in the 2.5" form factor (we will not take into account marginal devices with PCI Express x4 and HSDL interfaces). The SSD is based on the top modification of the second generation SandForce controller - SF- 2281, uses 25nm NAND memory from Micron. Claimed performance is 550 MB/s for linear read, 500 MB/s for write, access time is 0.1 ms. Maximum performance when writing 4 KB blocks with random. addressing - up to 85,000 IOPS.

The fourth and fifth participants in testing will be hybrid Intel configurations Smart Response from a single WD VelociRaptor in tandem with OCZ Vertex 3 Max IOPS. They will differ only in the caching operating modes. What is Intel Smart Response? As we mentioned above, its essence comes down to caching actively used data from hard drives(which, no matter how fast and perfect they are, are many times inferior to solid-state ones in a number of parameters). System in background analyzes which OS and user software files are accessed most frequently and moves them to the SSD drive. Unfortunately, Intel marketers do not provide the opportunity to use this option for all users of the company's platform - Smart Response is only available on the Z68 chipset. To work as part of such hybrid arrays, the company offers its own SSD Intel 311 (Larson Creek), optimized specifically for these purposes (it is based on SLC chips, which cost an order of magnitude more than MLC, but also “live” much longer). Fortunately, at least there are no restrictions here, so we use the regular OCZ Vertex 3.

Setting up Intel Smart Response

The procedure for setting up Intel Smart Response is quite simple, although it is not without pitfalls. The first difficulty that a user of an already assembled and working system who wants to speed up their HDD may encounter is the need to switch the controller to RAID mode. Naturally, without some tricks it will not be possible to do this painlessly - the OS will stop loading. The problem is solved either by replacing the drivers with standard ones from Microsoft and editing the registry, or by “injecting” RAID drivers through the Windows 7 installer or Acronis True Image Plus Pack.

The second difficulty is that after the procedures described above, the control utility Intel Rapid Storage still does not display the ability to organize Smart Response. The problem can be solved by reinstalling the drivers (and will probably be fixed in a new version of the package in the future).

Creating a Hybrid Intel Smart Response Array

Status of the created array

So, after installing an SSD into the system, the Accelerate tab appears in the Intel Rapid Storage Control Center, in which you can select how much SSD we want to allocate for caching (13.6 GB or the maximum possible 64 GB), and in what mode it will be operate Smart Response - Enhanced or Maximum. They differ in the nature of caching: the improved one implies buffering only those data for which active read requests are made (executable files, libraries, etc.), and the maximum one also caches write operations. Accordingly, work with all kinds of temporary files, containers (for example, a scratch file Adobe Photoshop or the Lightroom catalog), but in the event of a power outage or SSD failure, data will inevitably be lost, because. physically, until active access to them ceases, they will not be transferred to the HDD.

If you plan to configure Smart Response from scratch, and then install it on a hybrid OS array, then the procedure can be performed in the configuration menu of the disk controller, which is displayed immediately after POST.

The remaining part of the SSD is available to the user

Note that the unused Smart technology The Response part of the SSD remains accessible to the user—for example, software can be installed on it.

Finally, the sixth participant is OCZ Synapse Cache with a capacity of 120 GB.

It actually differs from its brother under the Vertex brand (as well as from the Agility series) only in the firmware.

OCZ Synapse Cache

The basis of this drive is still the SandForce SF-2281, but the firmware of this model is primarily focused on durable operation. For this, the degree of over-provisioning (reserving cells for replacement stock in case of their gradual failure) is as much as 50%.

Drive board

In fact, the 120 GB model has only 60 GB available for work, and the younger version with a capacity of 60 GB has only 30. Obviously, there is no point in using Synapse Cache as a regular SSD.

Back cover

Changes to the firmware were made for a reason. Synapse Cache is designed to work with the OCZ-licensed Dataplex utility from the American company NVELO. Like the Intel Rapid Storage Driver, this utility analyzes all disk operations occurring on the computer on the fly and transfers “hot” data to the SSD in the background. However, there are also differences: firstly, after installing it, the SSD completely disappears from the system and becomes inaccessible to the user. Secondly, Dataplex does not require operation in RAID mode and, therefore, is compatible with motherboards whose HDD controllers do not support this technology. The main “advantage” of this solution is full compatibility with all chipsets, and not just with Intel Z68.

Unfortunately, there are some limitations: Dataplex only works on Windows 7 and does not currently support hard drives with a capacity of over 2 TB (which is planned to be fixed by the end of the year). In addition, it only caches access to the system HDD, so if you want to install software or games on another hard drive, they will not be “accelerated”.

The peculiarity of the technology is that it always caches both reading and writing data. It does not have a safe intermediate mode like Enhanced in Smart Response. Naturally, this makes us fear for the safety of user data, but that is why OCZ Synapse Cache has 50% of the reserve area, and not 6.25%, like Vertex 3.

The beauty of Dataplex is the incredible simplicity of its setup: you just need to connect the SSD, download the utility from the OCZ website (by registering first), install it by entering the code located on the instructions for the drive and its case, and reboot the PC. All.

That's all the configuration utility

Amazingly, there is no need to perform any further manipulations; the system does not have any settings, and they are not needed. In the Start menu there is only the Dataplex status check utility, which cheerfully reports that caching is active.

Well, let's see what turns out to be better.
Testing methodology

Testing was carried out on a test bench with the following configuration:

motherboard: Sapphire Pure Platinum Z68 (Intel Z68 Express);
CPU: Intel Core i3-2100;
RAM: Kingston KVR1333D3N9 (2x2 GB, DDR3-1333);
video card: Palit GeForce GTX 480;
drives: WD VelociRaptor WD1500HLHX x2, OCZ Vertex 3 Max IOPS 120 GB, OCZ Synapse Cache 120 GB;
monitor: LG W3000H;
power supply: Huntkey X7-900 (900 W);
operating system: Microsoft Windows 7 Ultimate x64, Intel RST Driver 10.8.0.1003.

The following set of test applications was used:

CrystalDiskMark 3.0.1 x64 - synthetic assessment of the linear speed of the drive, speeds in multi-threaded mode with a queue depth of 64 requests, with random access in 4 KB blocks, as well as access time;
AS SSD Benchmark 1.6.4237.30508 - synthetic assessment of the linear speed of the drive, speeds in multi-threaded mode with a queue depth of 64 requests, with random access in 4 KB blocks, as well as access time;
HD Tune 5.0—removing a linear reading diagram from drives;
Futuremark PCMark Vantage HDD Suite - a set of test traces that emulate the user's work in the most popular types of applications;
Futuremark PCMark 7 System Storage - similar to PCMark Vantage, is a set of test tracks aimed at evaluating a PC's system storage;
Retouch Artists Photoshop Benchmark - an automated set of filters for Adobe Photoshop, designed to evaluate PC performance;
DriverHeaven Photoshop Benchmark - a similar set of filters for Adobe Photoshop to the previous one;
PPBM5 - benchmark for Adobe Premiere CS5, which is a project of three different video renderers, one of which is critical to hard drive performance.

Also evaluated:

OS startup time using the BootRacer utility (records the time between the start of the OS kernel and the full loading of all services and programs in startup);
OS startup time with Microsoft Word, Excel and PowerPoint 2010 placed in startup, opening text document 4.2 MB (4208 pages), 50.6 MB table (65187 rows) and 72 MB presentation (69 slides), respectively;
launch time of the Crysis 2 test scene in DirectX 11 mode with High Resolution Texture Pack (from pressing Start in the Adrenaline Crysis 2 Benchmark Tool to the start of the scene);
S.T.A.L.K.E.R test scenes launch time Call of Pripyat Benchmark (sum of times from the appearance of the game's splash screen to the start of the test scene).

All tests were run 5 times to allow all caching algorithms to achieve maximum performance.

CrystalDiskMark

WD VelociRaptor WD1500HLHX

2x WD VelociRaptor RAID-0

The first synthetic test, as expected, immediately gives preference to SSD drives, and this is not surprising: devices based on the most powerful second-generation SandForce controllers can boast not only minimal access time (which is the main trump card of solid-state drives), but also huge linear speeds. As a result, OCZ Vertex 3 is far ahead of WD VelociRaptor and RAID-0 based on it. However, there are also interesting results: firstly, it is noticeable that Intel Smart Response has quite significant overhead. In particular, in Enhanced mode we see excellent performance gains in read mode, but write performance is even lower than a single hard drive with linear access. Switching to the Maximized mode has an even greater effect: the system loses another 40 MB/s when reading, but the write speed naturally increases incomparably, especially on small blocks. True, in linear access Smart Response cannot compare even with the RAID-0 of the Raptors, let alone with a single OCZ Vertex 3. However, you need to understand that in in this case recording occurs not to the SSD itself, but “through” it to the hard drive, and the observed increase is the average value obtained due to sharp bursts in speed when writing to a solid-state drive.

The second interesting observation: the system with OCZ Synapse Cache in the synthetic test is significantly inferior to both Intel SRT modes. In terms of linear speeds, it is comparable to RAID-0, and when working with small blocks it is up to 50% inferior to Intel SRT. It is difficult to say what explains these results: on the one hand, in synthetic tests, such caching algorithms should interfere as little as possible so as not to wear out the NAND cells, on the other hand, it is unlikely that NVELO managed to develop a more “smart” algorithm than Intel. It is likely that the overhead costs of processing the barrage of requests that CrystalDiskMark and similar utilities generate are simply due to the overhead costs of Dataplex being higher than those of Intel Smart Response.

AS SSD Benchmark

WD VelociRaptor WD1500HLHX

2x WD VelociRaptor RAID-0

WD VelociRaptor + OCZ Vertex 3 Max IOPS (Intel SRT Enhanced)

WD VelociRaptor + OCZ Vertex 3 Max IOPS (Intel SRT Maximized)

WD VelociRaptor + OCZ Synapse Cache

Despite the similarity of the CrystalDiskMark and AS SSD Benchmark utilities, they are based on different testing algorithms, in particular, the latter evaluates the performance of SSDs much more scrupulously, and the total amount of data written to them reaches 3 GB in one pass. As a result, we get a rather interesting picture.

Pay attention to the interesting performance gains when moving from one HDD to RAID-0. In linear read and write mode, it is, as expected, around 80-90%. However, when the array is loaded with small requests in multi-threaded mode, it begins to work more than twice as fast as a single disk! The explanation for this is simple: the Intel Rapid Storage driver logic does an excellent job of caching, and the well-debugged VelociRaptor firmware successfully reorders the request queue. These HDDs are designed specifically to work in such conditions, and it is not surprising that their potential is better revealed in RAID rather than in single mode.

Note that AS SSD, unlike CrystalDiskMark, does not detect a significant drop in read speed in Intel SRT Maximized mode relative to Enhanced mode, although both of them work about 20% slower than a stand-alone SSD. It is also interesting that in the Enhanced mode, the flow of AS SSD read requests is not completely transferred to the caching solid-state drive: in the 4K 64Thrd pattern (access in 4 KB blocks with random addressing in 64 simultaneous threads), in this mode the array demonstrates 18200 IOPS versus 45500 IOPS in Maximized mode.

As for OCZ Synapse Cache, the performance with it remains the same - it is almost twice as slow when reading as Intel SRT, but it copes much better with writing (especially multi-threaded). Most likely, the peculiarities of Dataplex’s operation are again at play here: on the one hand, this algorithm caches read requests less actively, on the other, it copes better with writes.

HD Tune

WD VelociRaptor WD1500HLHX

2x WD VelociRaptor RAID-0

WD VelociRaptor + OCZ Vertex 3 Max IOPS (Intel SRT Enhanced)

WD VelociRaptor + OCZ Vertex 3 Max IOPS (Intel SRT Maximized)

WD VelociRaptor + OCZ Synapse Cache

Finally, examining the read graphs of all six storage subsystem options we're looking at gives a rough idea of why the previous two tests behaved so unusually. As can be seen in the graphs of both Intel SRT modes, when accessing the HDD for reading in linear mode, the driver begins to do something, most likely, actively cache data stored at addresses selected by the test application. As a result, we observe a noticeable “dip” in speed. As soon as the occupied space ends (and in our case, a sharp jump in the graph to the normal level is approximately at the border of the area occupied by the OS and the test package) - everything returns to normal. In addition, the read graph of the system with a single OCZ Vertex 3 also shows that this SSD is very actively conducting service operations on the occupied zone in the background.

The reading graph of a system with OCZ Synapse Cache cannot be explained at all; apparently, the nature of HD Tune calls is simply incomprehensible to Dataplex. In fact, with HD Tune's HDD access format (linear in 1MB blocks), Dataplex even reduces performance relative to the raw speed provided by the hard drive.
Futuremark PCMark Vantage HDD Suite

PCMark Vantage is primarily critical to the drive's read access time, which is why OCZ Vertex 3 occupies a dominant position here. This has a similar effect on the results of a RAID array of two WD Velociraptors: despite the linear read and write speed doubling and the random write speed more than doubling, the results for this configuration are only 400 points higher than for a single “raptor.” For the same reason, the Smart Response Maximized array is only slightly superior to the Enhanced configuration - most of the operations performed by the test package are read-oriented. Note that in the overall standings, OCZ Synapse Cache lags behind Intel SRT by only 10% - significantly less than in synthetic tests.

Interestingly, Dataplex copes with subtests Windows Media Center, Windows Media Player and loading applications better than Intel SRT in Enhanced mode - this is a direct indication of the benefits obtained from caching writes. At the same time on Windows Photo Gallery and Windows Defender, the alternative technology is hopelessly inferior, which is why it ends up behind the Intel solution.

We also note that in PCMark Vantage we encountered strange behavior of OCZ Synapse Cache, or rather Dataplex technology. After the first pass of tests, subsequent ones inevitably showed very low results, and it turned out to be impossible to trace the system: in one pass the array could get 15,000 points, in the second - 7,000, and in the third - 3,000. It was possible to return the indicators to the expected 30,000 only by repeating any of the synthetic tests tests (reboot did not help). Obviously, in this case we have a local software defect, which, quite likely, NVELO will fix in the next release. However, this behavior was not detected in any other test, so this can be considered an isolated case that does not affect the overall result.

Futuremark PCMark 7 System Storage Suite

The redesigned PCMark Vantage 7 System Storage Suite relies even more heavily on access time, but also pays slightly more attention to linear speed when calculating the final result. As a result, the RAID array is already ahead of a single HDD not by 5%, but by as much as 20%. At the same time, the significantly lower linear read speed shown by OCZ Synapse Cache relative to Intel Smart Response does this technology a disservice: it gets 45% lower results than SRT in Maximized mode. If you look at the results in each of the tests, you can see that absolutely everywhere Dataplex is significantly inferior not only to OCZ Vertex 3 Max IOPS, but also to both Intel Smart Response modes, while in PCMark Vantage this technology sometimes outperformed them.

OS startup time

	Run 1	Run 2	Run 3	Run 4	Run 5
WD1500HLHX	28	25	20	20	20
2x WD1500HLHX RAID-0	31	20	17	17	17
OCZ Vertex 3 Max IOPS 120 GB	12	12	9	9	9
	31	14	13	10	10
	24	9	10	9	9
	27	11	11	11	11

Let's move on from specialized tests to assessing what the use of each of the options tested today gives in real life. The first will be to download Windows 7 SP1 64-bit. The measurements were taken during five consecutive reboots.

As we can see, Microsoft has tried to minimize the impact of a slow disk subsystem in conditions where the user runs the same set of software: already on the second boot, Windows Prefetcher and SuperFetch technologies move the most actively used executable files and libraries to the beginning of the disk (its fastest part) and automatically loading them into RAM at startup, reduces startup time by 12% for a single WD VelociRaptor and 55% (!) for RAID-0. By the third restart they have already reached their maximum efficiency, and the time is reduced even more - by 40% and 82%, respectively!

The transition from HDD to SSD, as expected, greatly reduces boot time - with OCZ Vertex 3 Max IOPS Windows 7 starts in just 12 seconds, and after SuperFetch “throws out” everything unnecessary from the preload - in just 9. And here comes It’s time to be surprised at the performance of hybrid arrays: as we see, the first start of the system turns out to be approximately the same as with an HDD, but the second time the start time is reduced radically. What's interesting is minimum value Intel SRT Maximized and Dataplex systems achieve this already on the second restart, and Enhanced requires three starts.

OS and MS Office startup time

	Run 1	Run 2	Run 3	Run 4	Run 5
WD1500HLHX	60	62	29	23	26
2x WD1500HLHX RAID-0	29	26	28	28	31
OCZ Vertex 3 Max IOPS 120 GB	14	15	12	15	13
WD1500HLHX + OCZ Vertex 3 SR Enhanced	21	16	12	19	12
WD1500HLHX + OCZ Vertex 3 SR Maximized	20	21	15	15	15
WD1500HLHX + OCZ Synapse Cache 128 GB	31	14	16	17	13

Adding “heavy files” of Microsoft Word, Excel and PowerPoint to startup significantly increases the loading time of the OS, and the more obvious the effect of caching becomes. As you can see, on the third restart a single HDD receives a more than twofold increase in performance from SuperFetch and Prefetcher, while these technologies do not affect RAID-0 at all, unlike booting a clean system. Obviously, in the case of two VelociRaptors, the operating system with all the software already fits on the fastest external tracks of the platters, and Microsoft technologies simply cannot make loading faster.

A similar situation is observed with OCZ Vertex 3: all five passes of this test show approximately the same loading time, although there are fluctuations within three seconds. In general, Vertex 3 Max IOPS is twice as fast as RAID-0 and four times as fast as a single WD VelociRaptor.

Compared to the previous three participants, hybrid arrays look especially impressive. Intel SRT Enhanced already at the first launch shows less time than a single HDD (obviously, some OS and software components are duplicated, and transferring them to an SSD already gives a speed increase), and at the third launch it reaches maximum performance identical to OCZ Vertex 3. However, , as with a single SSD, this configuration exhibits fluctuations from pass to pass of up to 7 seconds. A similar situation is observed with Dataplex: an array with OCZ Synapse Cache loads the OS and office suite a couple of seconds slower than Intel SRT, and its performance is also not stable. The only configuration that pleased me with the repeatability of the results was Intel Smart Response Maximized - it completed the third restart in 15 seconds and subsequently did not slow down even once.

PPBM5 (Adobe Premiere Pro CS5) Disk Test

	Run 1	Run 2	Run 3	Run 4	Run 5
WD1500HLHX	142	142	144	143	142
2x WD1500HLHX RAID-0	135	135	134	134	134
OCZ Vertex 3 Max IOPS 120 GB	136	135	133	133	133
WD1500HLHX + OCZ Vertex 3 SR Enhanced	139	135	136	136	136
WD1500HLHX + OCZ Vertex 3 SR Maximized	138	145	141	137	136
	145	135	136	137	143

The disk test from the PPBM5 benchmark is a rendering of a 13 GB AVI movie from a large number source files, which should provide heavy load to the disk subsystem. In practice, we see that it is mainly critical to disk throughput: all configurations that reach about 250 MB/s in linear mode cope with rendering in approximately the same time. Only a single WD VelociRaptor (which is natural) and a hybrid array with OCZ Synapse Cache are behind the leaders, which, as we have already seen from synthetic tests, turns out to be significantly slower than Intel SRT and OCZ Vertex 3 in terms of linear reading speed.

Retouch Artists Photoshop Benchmark (Adobe Photoshop CS5 Extended)

	Run 1	Run 2	Run 3
WD1500HLHX	21,5	21,8	21,2
2x WD1500HLHX RAID-0	19,5	19,7	19,6
OCZ Vertex 3 Max IOPS 120 GB	22,4	20	20,8
WD1500HLHX + OCZ Vertex 3 SR Enhanced	20,7	20,8	20,8
WD1500HLHX + OCZ Vertex 3 SR Maximized	21,2	20,4	20,2
WD1500HLHX + OCZ Synapse Cache 120 GB	20,6	20,2	20,9

This test is a set of filters and operations that are automatically applied to a test image. As you can see from the table, all six configurations cope with it approximately equally quickly, with a gap of about 1.5 seconds. Note that in this case, the number of test passes does not affect the speed in any way (to test it, it was specifically carried out on Intel Smart Response Maximized 10 times - to no avail).

HardwareHeaven Photoshop Benchmark (Adobe Photoshop CS5 Extended)

	Run 1	Run 2	Run 3
WD1500HLHX	200,6	201,2	200,5
2x WD1500HLHX RAID-0	187,9	187,7	188,1
OCZ Vertex 3 Max IOPS 120 GB	198	197,5	198,4
WD1500HLHX + OCZ Vertex 3 SR Enhanced	198,2	197,9	198,2
WD1500HLHX + OCZ Vertex 3 SR Maximized	199,2	198,5	198,3
WD1500HLHX + OCZ Synapse Cache 128 GB	198,8	198,1	198,3

Like the previous test, this set of filters and operations (albeit significantly more complex and resource-intensive) does not receive a performance boost from SSD caching. Of all the participants, only two are worth highlighting: a single WD VelociRaptor turns out to be noticeably slower than all other configurations (although “noticeably” is only 3 seconds), but RAID-0 is unexpectedly much ahead of both hybrid configurations and even SSDs. Considering that it should be inferior to them in all speed parameters, the only logical explanation for this fact is the larger volume used by Photoshop for the scratch file (in all configurations it was given all the free space).

Crysis 2

	Run 1	Run 2	Run 3	Run 4	Run 5
WD1500HLHX	64	62	63	40	39
2x WD1500HLHX RAID-0	52	40	41	40	39
OCZ Vertex 3 Max IOPS 120 GB	45	39	39	42	38
WD1500HLHX + OCZ Vertex 3 SR Enhanced	55	49	48	41	40
WD1500HLHX + OCZ Vertex 3 SR Maximized	57	39	40	40	39
WD1500HLHX + OCZ Synapse Cache 120 GB	67	44	39	40	41

Finally, let's move on to games. Crysis 2 with a set of textures high resolution It takes up 12.5 GB of disk space and takes quite a long time to load. Judging by the minimum results shown by all six configurations in the test, our stand is capable of running the benchmark in about 40 seconds, but there are caveats here.

Firstly, in real conditions, the disk speed affects the loading time of levels and locations, which the player does more than once a minute (unless, of course, he is constantly killed in the same place). Consequently, the advantages of SuperFetch, which we observed on the example of VelociRaptor and RAID-0, most often will not be so noticeable - during the game from level to level, enough data will be read from the disk for the preloader to “contaminate” this cache, and it will not show maximum efficiency . This situation should not happen with hybrid combinations, because The buffer size of the SSD will be sufficient for everything that happens. A particularly significant increase will be observed in the case of duplication of elements between locations: then the first load will take, for example, 30 seconds, and the second may well occur in 10.

Returning to our results, we see that by the second boot, RAID-0, Intel SRT in Maximized mode and, naturally, OCZ Vertex 3 achieve maximum efficiency. OCZ Synapse Cache shows the coveted 40 seconds on the third restart, and Intel SRT Enhanced and a single WD VelociRaptor - at the fourth.

S.T.A.L.K.E.R. Call of Pripyat

	Run 1	Run 2	Run 3	Run 4	Run 5
WD1500HLHX	123	126	121	121	124
2x WD1500HLHX RAID-0	113	97	97	98	97
OCZ Vertex 3 Max IOPS 120 GB	104	98	99	98	99
WD1500HLHX + OCZ Vertex 3 SR Enhanced	118	99	102	101	100
WD1500HLHX + OCZ Vertex 3 SR Maximized	117	99	100	99	101
WD1500HLHX + OCZ Synapse Cache 120 GB	150	99	99	98	100

As with Crysis 2, S.T.A.L.K.E.R. Call of Pripyat does not scale very well from disk subsystem acceleration: the minimum total loading time for all four tests in this benchmark is about 97-98 seconds. However, the effect of caching is noticeable here too, and it is fully achieved already on the second start of all systems, except for a single WD VelociRaptor. Unlike other test participants, this hard drive does not receive a speed boost from Windows caching systems and is inferior to more fast systems about 25 seconds. We would like to emphasize that the hybrid with OCZ Synapse Cache coped with this test no worse than Intel Smart Response.

Conclusions

A fast disk subsystem is no less important than an overclocked processor or a powerful video card. Moreover, it cannot be overclocked - it can only be replaced or supplemented. With the advent of hybrid technologies like Intel Smart Response and NVELO Dataplex, users have a new opportunity to improve PC performance, and, as testing shows, in most cases it is not a compromise at all. Undoubtedly, a single SSD provides higher performance than “hybrids”, but its cost and limited capacity does not allow most users to install whatever they want, regardless of the amount of data. Considering that modern games or professional software can easily take up a dozen or two gigabytes, the most popular SSD capacity of 120 GB is enough for only 8-10 such installations. At the same time, a hybrid array of a high-speed hard drive and a 60 GB SSD will cost about the same amount, but will be incomparably more comfortable to use, albeit a little slower.

Returning to today's testing, we can conclude that Intel Smart Response is currently superior to the developments of other companies in terms of efficiency. NVELO Dataplex, used by OCZ for its SSD Synapse Cache, also copes well with its tasks, but is noticeably inferior to Intel's development. However, judging by the fact that in some cases it still gets ahead, we are not talking about a fundamental flaw, but about a banal imperfection of the software, which, as we know, can be corrected and improved. Considering that NVELO primarily positions Dataplex as a solution for server systems, there is no doubt about the active development of the software part.

And finally, comparing Intel Smart Response and OCZ Synapse Cache, we can only say one thing: there is simply no need to compare them. Smart Response only works on Intel Z68, and organizing this particular array on this chipset will be the best solution. On all other platforms this feature simply does not exist, and there Synapse Cache will be a great way to get the responsiveness of an SSD system without sacrificing HDD capacity.

Testing equipment was provided by the following companies:

Hello admin! I want to buy hard drive with a capacity of 1-2 TB, one computer geek friend advised me to buy an SSHD disk (a hybrid of a hard drive and an SSD solid-state drive), since it works noticeably faster than a regular HDD, but is not as expensive as an SSD solid-state drive. What can you say about such discs?

Hello friends! Very good question. Yes, SSHD hybrid hard drive ( Solid State Hybrid Drive) works 30% faster than a conventional hard drive, and is about the same more expensive. If a regular 1 TB hard drive costs 4,000 rubles, then an SSHD can be bought for 5,400 rubles. Such disks are produced both for regular computers and for laptops.

Firstly, What is a hybrid hard drive?

The technology for the production of hard drives (the only component of a computer that has moving mechanical parts) has long been at a dead end and it is almost impossible to increase the performance of a hard drive through production, as evidenced by the appearance on the market of solid-state drives SSDs and hybrid hard drives SSHD. But if a solid-state drive is a completely non-mechanical storage device based on memory chips, then a hybrid hard drive is, first of all, a regular hard drive with an MLC fast flash memory card (8 GB capacity) soldered onto it, used in the production of solid-state drives. that is, it turns out that SSHD is a hybrid of a regular hard drive and an SSD..

Secondly, why is an SSHD hybrid hard drive faster than a regular hard drive?

Seagate SSHD hybrid drives use self-learning technology - Seagate Adaptive Memory, which examines the operating system installed on the disk from the first seconds of operation, as a result, the most frequently used programs and files are copied to the flash memory of the SSHD disk, such files include, first of all, elements involved in loading the operating system, which means that Windows will be installed from the second or third time boot faster, because Windows will be loaded from flash memory. For example, on my computer loading Windows 8.1 installed on regular HDD, occurs within 35-40 seconds, and on SSHD - 20 seconds, on a regular SSD solid state drive - 15 seconds. The same applies to applications you constantly use; they will launch a little faster. Let's take, for example, a resource-demanding computer modern game, which you constantly play, according to my observations, such a game will load three times faster than on a regular HDD.

Hybrid SSHD hard drive is the golden mean

In general, the ideal drive configuration option is system unit For a typical home user, it looks like this: you buy two drives, the first is an SSD (volume 120-240 GB) for installing the operating system, and the second is a regular HDD for storing files (volume) 2-3 TB, you need about 10,000 rubles for all this . And if you buy one hybrid drive SSHD for 1 TB, it will cost you 5,400 rubles, and SSHD for 2 TB - 7,000 rubles. Of course, everything won’t fly (as is the case with an SSD), but maybe you don’t need such speeds. A hybrid SSHD drive is coming out, this is the golden mean - for little money you get good performance and a large amount of disk space.

Which SSHD to buy

Until recently, SSHD hybrid drives were produced by the company that developed them - Seagate. In total, there are now three Seagate Desktop SSHD models on the market with capacities of 1, 2, 4 TB.

Seagate Desktop SSHD ST1000DX001 1 TB

Seagate Desktop SSHD ST2000DX001 2 TB

Seagate Desktop SSHD ST4000DX001 4 TB

Also, recently Western Digital began to produce SSHD, but they are few on the market, and the model that I came across - WD Blue SSHD, WD40E31X with a capacity of 4 TB, was no different in speed characteristics from a similar model Seagate ST4000DX001 4 TB.

In today’s article, I suggest you consider the Seagate Desktop SSHD ST2000DX001 2 TB model and here’s why. If we take the Seagate Desktop SSHD 1 TB model, then 1 TB of disk space is not enough for modern user computer. If we take the Seagate Desktop SSHD 4 TB model, then on the contrary, not everyone needs a large volume of 4 TB of disk space, and its price is quite high (11,500 rubles), and what is also important is the spindle speed of this drive: 5900 rpm, that is, it is slightly slower than other SSHDs with a capacity of 1 and 2 TB (spindle speed 7200 rpm) and this will certainly affect the performance of the operating system.

So, I persuaded you and we have a model in front of us Seagate Desktop SSHD ST2000DX001 2 TB

Upon closer inspection, the hybrid Seagate drive Desktop SSHD ST2000DX001 2 TB turned out to be ordinary hard drive, only it says SSHD on it.

Disk space - 2 TB

SSD buffer capacity - 8 GB

Cache memory size - 64 MB

Spindle speed - 7200 rpm

On the back of the drive we see a special printed circuit board Adaptive Memory, with 8 GB of fast MLC memory and a “hybrid” controller soldered.

It’s very easy to install the drive into the system unit.

SMART hard drive in the CrystalDiskInfo and Victoria program.

The hybrid drive is new and has been used for 0 hours.

Read and write tests

To make sure that our disk is really good, let's perform several tests on read and write using special programs: CrystalDiskMark 2.0, ATTO Disk Benchmark and SiSoftware Sandra. These utilities will sequentially read and write information to our hybrid disk in small blocks, then show us the result.

CrystalDiskMark 2.0

The simplest and most frequently used program in this regard, you can download it on my Yandex.Disk

The utility is very simple, select only the desired drive letter (in our case E:)

And click AII, the test will begin SSHD disk for performance.

1. Test of sequential reading and writing of large blocks of data;

2. Test of random reading and writing in 512 KB blocks;

3. Test of random reading and writing in 4 KB blocks;

I can say that the result is very worthy, especially the recording in 512 KB and 4 KB blocks.

ATTO Disk Benchmark

Let's test the hybrid disk with another program - ATTO Disk Benchmark.

Select the drive letter of the SSHD hybrid drive and click Start.

Result.

SiSoftware Sandra

A global program capable of diagnosing all computer components and having its own official rating.

As a result, our disk is ahead of 94% of the results. Excellent performance.

Disadvantages of SSHD

In my opinion, the only disadvantage of the SSHD is the small amount of built-in flash memory 8 GB, it would be great if its size increased to 32 GB, then more running programs would be placed in the solid state cache and the performance of Windows would be exactly the same as if it was installed on the SSD.

If you decide to purchase a solid-state SSD drive, there may be several reasons for this:

You are not satisfied with the speed of your HDD.
You need fast operation of Windows and certain types of applications and games.

However, installing an SSD into a computer or laptop and then filling it with information is not enough. It is also necessary to optimize its operation with the operation of your OS.

Let's look at the main methods for optimizing an SSD drive.

AHCI SATA

Technology that allows the TRIM function to be used for various SSDs. It is enabled at the BIOS level of your PC or laptop.

Enable AHCI SATA:

Opening command line using the key combination win + R.
Enter the command: “regedit” (access to the registry).
Go to the following path: HKEY_LOCAL_MACHINE → SYSTEM → CurrentControlSet → Services → storahci.
Change the value of the ErrorControl subkey to 0 (default 3) by calling the context menu and clicking the “Modify” option.
Go to the branch called “StartOverride” and change its value to 0 (default 3).
Restart your PC (laptop), go to BIOS/UEFI (how to enter BIOS, see separately for the model of your laptop or PC motherboard). In the “storage configuration” section, and in the “SATA port” subsection, set AHCI, or in the “SATA RAID/AHCI Mode” section, set AHCI (For different BIOS versions, their sections and subsections).
Check if the function works in Windows. Go to the following path: Control Panel → Device Manager → IDE ATA/ATAPI controllers. The device should appear in the last subsection: “Standard SATA AHCI Controller”.

TRIM function

By default, this feature is enabled on Windows 7 and higher, however, it is better to manually check whether this feature works. The meaning of TRIM is that after deleting files, Windows transmits information to the SSD drive that a certain area of the disk is not used and can be cleared for writing. (V HDD data remain and the recording is made “on top” of the existing one). Over time, if the function is disabled, the drive's performance will drop.

Checking TRIM on Windows:

Launch the command prompt by pressing the key combination win + R.
Enter the command: “fsutil behavior query disabledeletenotify”.
If after entering the message “DisableDeleteNotify = 0” is displayed, then the TRIM function is enabled, if “DisableDeleteNotify = 1”, then TRIM does not function. If TRIM does not work, enter the command: “fsutil behavior set DisableDeleteNotify 0”, then repeat steps 2 and 3.

Defragmentation

This feature helps optimize and speed up HDD operation, but for SSD, it has a detrimental effect. For SSDs, the “automatic defragmentation” feature is disabled by default. To check if it works:

Press the win + R combination.
In the command line window, enter the command: “dfrgui” and click “OK”.
In the window that opens, select your SSD and look at the “Schedule optimization” item. For our SSD it should be disabled.

Indexing

Windows feature that helps you perform quick search files on the disk with large amounts of information, however, increasing the write load on the SSD. To disable it:

Go to the section “This computer”, “My computer”, “Computer” (it’s different for each OS).
Select your SSD and select “Properties” in the context menu.
In the window that opens, uncheck the box next to the option: “Allow the contents of files on this disk to be indexed in addition to file properties.”

Search service

Its function creates a file index, thanks to which finding various files and folders is faster. However, the speed of the SSD is quite enough to abandon it. To disable it you must:

Go to the following address: Control Panel → System and Security → Administrative Tools → Computer Management.
Go to the tab: “Services”.
Find the “Windows search” service and select “Disabled” in the “Startup type” tab.

Hibernation

A mode that allows you to save the contents of RAM on the hard drive, so that the next time you turn it on, information and open applications from the previous session are saved.

When using an SSD, the meaning of this function is lost, since the drive starts up quickly anyway. And “Hibernation”, creating “write-overwrite” cycles, reduces the lifespan of an SSD disk.

Disabling hibernation:

Launch cmd.exe again using the key combination win + R.
Enter the command: “powercfg -h off”.

Write caching

This feature improves the performance of your SSD. When enabled, NCQ writing and reading technology is used. NCQ – accepts multiple requests simultaneously and then arranges their execution order in such a way as to achieve maximum performance.

To connect you need:

Open the command line with the combination win + R
Enter the command: “devmgmt.msc”.
Open « Disk devices", select SSD and select "Properties" in the context menu.
Go to the “Policies” tab.
Check the box next to the option: “Allow recording caching for this device.”

Prefetch and Superfetch

Prefetch– a technology with which frequently used programs are loaded into memory in advance, thereby speeding up their subsequent launch. In this case, a file of the same name is created on disk space.

Superfetch– a technology similar to Prefetch with the difference that the PC predicts which applications will be launched by loading them into memory in advance.

Both features are of no use when using an SSD. Therefore, it is best to turn them off. To do this:

Open the command line using the key combination win + R.
Execute the command: “regedit” (go to the registry).
Follow the path: HKEY_LOCAL_MACHINE → SYSTEM → CurrentControlSet → Control → Session Manager→ Memory Management → PrefetchParameters.
Find several parameters in the registry subkey: “EnablePrefetcher” and “EnableSuperfetch”, set their value to 0 (default 3).

SSD Mini Tweaker utility

All of the above actions can be performed manually, but programmers have created programs called tweakers, the purpose of which is to customize the Windows OS, as well as its individual components, with a few clicks. One such program is SSD Mini Tweaker.

SSD Mini Tweaker– a program, a type of tweaker, that allows you to optimize your SSD without much effort.

Advantages:

Complete Russification.
Works on all OS starting from Windows 7.
Free.
Clear interface.
No installation required.

Other ways

Manipulations such as transferring browser caches, paging files, temporary Windows folders, system backup from an SSD to an HDD (or disabling this feature) are useless, since although they increase the lifespan of the SSD, they limit the potential of its use.

Thus, by performing the simple above-mentioned manipulations with your OS, you can extend the life of your drive, as well as configure it to maximum performance mode.

Introduction

As businesses evolve, workload-intensive applications are often limited by the capabilities of hard disk drives (HDDs). Although HDD capacities have increased dramatically, the speed of random input/output (I/O) operations has not increased at the same pace. However, it is now possible to speed up the processing of read-intensive streams, such as in online transactions (On-Line Transaction Processing - OLTP), in network and file servers, databases, using new technology caching, Infortrend SSD Cache, which uses high speed and low latency SSDs for faster read speeds for frequently needed, vital data. The read speed of SSD is significantly higher compared to HDD and hence SSD Cache can significantly improve random read performance and reduce response time.

Applicability of this document

EonStor DS Family

What is SSD Cache?

A cache is a component that transparently stores data so that subsequent accesses to it can be served more efficiently. It is critical for storage, especially in read-intensive applications. Without enabling SSD Cache, the controller's cache capacity is limited. SSD Cache allows you to use fast SSDs to expand your storage system's cache pool and accumulate frequently accessed data. As SSD Cache capacity increases, the cache hit rate also increases. In other words, more and more “hot” data will be stored in the SSD Cache, future accesses to this data will be served more efficiently and, therefore, read performance will improve.

Why Infortrend SSD Cache?

In many cases where the percentage of reads in a worker thread is significantly higher than writes, and small amounts of data are read repeatedly, SSD Cache can provide the following benefits:

1.Improved reading performance

SSD Cache uses an intelligent algorithm to speed up processing of intensive random read workloads such as OLTP and database access. In such situations, SSD Cache can significantly increase overall read speed. For example, SSD Cache can increase OLTP IOPS by 2.5 times compared to the same system without SSD Cache. At the same time, latency is also reduced and hence the extent of performance improvement depends on the actual application workflows and user behavior.

2.Intelligent software and control algorithm

Intelligent software automatically analyzes data access patterns and recognizes sequential and random read/write operations. Sequential read or write data is not written to the SSD Cache pool, only random read data is accumulated in it to ensure that the SSDs are used most efficiently. More specifically, the firmware automatically moves copies of the most frequently needed data from the controller cache to the SSD Cache pool at the appropriate time. This “hot” data will subsequently be read from the SSD Cache if the system receives a request to read it. The algorithm developed by Infortrend optimizes the cyclic copying of data to SSDs, so that relatively cheap SSDs can be used for this purpose. This solution not only improves read performance, but also extends the life of hard drives by reducing the number of read and write cycles.

3.Simple intuitive user interface

SSD Cache functionality is fully integrated into Infortrend SANWatch and RAIDWatch GUI. They are very easy to set up, manage and maintain. For example, the user can monitor the status of the SSD Cache pool and easily check the remaining lifespan for each SSD.

Infortrend SSD Cache

How Infortrend SSD Cache works

If SSD Cache is enabled and running for a period of time, the intelligent firmware collects statistics and immediately updates the data “temperature” records in the controller cache. Based on these records, the firmware automatically copies small pieces of random, frequently needed data from the controller cache to the SSD Cache pool at appropriate times using a sequential write method to avoid intensive SSD operations and hence increase their lifespan. Bye poolThe SSD Cache Pool will not fill up with “hot” data generated by applications on the host; the method of pre-copying blocks to the SSD using zone prediction speeds up read operations. If the data block size is less than or equal to16 KB, data is copied directly to the SSD pool, even if it is only read once. If the block size is greater than 16 KB, and the program recognizes it as “hot” data (read several times), then it is classified as frequently needed and stored in the SSD pool. Two copies will be stored for this “hot” data - one in the SSD Cache and one on the hard drives.

Typically, upon receiving a read data request, the system checks to see if the corresponding data is available in the controller cache. If the requested data is in the controller cache, the system immediately returns it to the host. If the requested data is not in the controller cache, then the system checks the SSD Cache pool. If the requested data was stored in the SSD Cache based on an assessment of its “temperature”, then the system reads this data directly from the SSD Cache and returns it to the host. Otherwise, the data will be returned from the slower device. Therefore, the more cache hits, the more requests will be served by the SSD Cache, so overall performance and average response time will improve.

What is needed for SSD Cache to work?

1. Requirements software and SANWatch

Software version 512F12 or higher

SANWatch version 3.0.h.14 or higher

2. SSD Cache License

SSD Cache is available under license. Infortrend also provides a 30-day trial license.

3.Relationship between controller cache capacity and maximum size SSD Cache pool:

If the system allows SSD operation Cache, the controller cache will use some space to store hot data, and the size of the hot entries in the controller cache will determine the maximum supported size of the SSD pool. For the starting combination (2 GB per controller), the maximum supported SSD Cache pool size is 150 GB for single controller and 300 GB for dual redundant controller models.

Compared to SSDs, the controller's cache memory is more energy efficient. In addition, not only read data, but also written data can enter the controller cache memory. Therefore, we recommend that EonStor DS users increase the cache memory to 16 GB per controller and purchase suitable cache pool SSDs (to suit your needs and budget) to get the maximum performance benefit.

4. Reset the controller(s) for launch SSD Cash

The last step of starting SSD Cache involves resetting the controller(s). By default, the controller cache does not allocate space for storing “hot” data. Therefore, the controller needs to be reset and initialized to allocate suitable space for hot writes. After resetting the controller and activating the SSD Cache functionit is very easy to operate. There is no need to reset or reboot the system when adding or removing an SSD from the pool. This procedure is performed using an intuitive user interface via SANWatch or RAIDWatch.

5.SSD requirements

Currently, one controller supports up to 4 SSDs. If you want to use the SSD Cache feature, please check whether the SSD model you select is listed on the Infortrend Qualified Vendor List (QVL). Only SSDs from our QVL can be used to improve storage performance as described in this document.

Conclusion

Infortrend SSD Cache is an intelligent solution that dramatically improves storage performance, especially for read-intensive applications, significantly reduces latency and supports large cache pools. It is easy to install, administer and maintain using Infortrend's intuitive user interfaces. We highly recommend using it on systems with heavy workloads and frequently repeated read operations.

You can download the full text of the article with illustrations as a pdf file.