May 14, 2013 at 15:19LSI practice
Even the most avid skeptics already recognize that SSD technologies have a number of undeniable advantages over conventional hard drives and allow you to get significantly higher I / O performance (and in some cases I / O speeds don't happen much). At the same time, SSDs are not yet ready for widespread implementation instead of traditional hard drives for a number of reasons: from price to reliability. What to do? Hybrid solutions come to the rescue, which combine traditional disks with SSDs, allowing you to get (albeit with certain reservations) the advantages of both solutions.
If you don’t go into details, the principle of operation of hybrid systems is quite simple (and if you go deep, you can dive so much that you won’t return in a year) and it is the same for all systems, starting with Seagate Momentus XT and Apple Fusion Drive, ending with expensive and complex solutions for large data storage systems and data centers, which will be discussed today.
For the main data storage, traditional hard disks are used, which are necessarily integrated into RAID, and the SSD is used to cache the most frequently used data, which should be accessed most often. The caching system is taken care of, and most often the cache partition in the system is generally not available as a separate disk.
In large and “serious” systems, I / O optimization issues are even more acute than for desktop computers. The fact that for the user is a second delay in launching a heavy program, in the case of a loaded server can result in many thousands of dollars of loss if the disk subsystem becomes a “bottleneck” that slows down all work.
Speaking about large storage systems, one cannot help but recall the LSI company, which, being one of the largest suppliers for storage systems, could not stay away. The LSI product portfolio includes a suite of solutions to accelerate the performance of disk systems in the Nytro family.
The line of these products is divided into three groups.
Nytro WarpDrive Application Acceleration Card
SSD disk, made in the form of PCI-express cards, which can be installed in most modern servers, providing a high-speed disk for the needs of the system.
Nytro XD Application Acceleration Storage Solution A
complex of WarpDrive cards and specialized software that integrates into the system and uses SSDs to cache critical data. This approach allows you to accelerate any block storage system, regardless of the connection method (iSCSI, DAS, SAS, FC or any other). All you need to accelerate is a block device.
Nytro MegaRAID Application Acceleration Card
These cards are essentially RAID controllers built on top-of-the-line LSI RAID-on-chip solutions, with the addition of high-speed NAND Flash. Special algorithms automatically determine the data to be cached, so this solution works independently of the operating system and is completely transparent to it.
To demonstrate in practice how to speed up work with I / O, on April 11, LSI held a seminar or workshop devoted to this topic (details about this event are given below).
Since corporate IT requires a reasonable and balanced approach, before implementing any technology or solution, you need to make sure that it will solve the existing problem. To help sort this out, LSI developed the Nytro Predictor program, which works with two main server operating systems - Windows and Linux. Predictor uses standard statistics collection tools for each system: xperf for Windows and blktrace for Linux, working in the background, completely transparent for user tasks.
To understand Predictor, you need to work for some time in the background, collecting data while the computer performs typical tasks (it is better to give time with a margin). After that, you can get acquainted with the results of his work, and on the basis of this it is already decided what to do.
Now let's go back to the technical seminar and look at all this in practice.
For experiments, we used the test bench of the following configuration:
SuperMicro server on the X8ST3-F platform, OS Windows 2008 Server R2
630J storage system
Nytro MegaRAID 8110 (200 Gb)
Nytro XD BLP4-400 (400 Gb)
MegaRAID 9271-4i
controller 9266-8i controller with CacheVault and CVM01 9286-8e
Controller
Run Predictor and load simulation. A typical load was an MSSQL query processing a lot of data.
We launched the profiler for data collection and then executed the request. If the profiler worked a little time - the results will be of little use
Having collected the data during the server’s operation in its “typical” mode, we give it to the “predictor”. After analyzing them, he renders his verdict.
You can see in detail how read / write requests were distributed depending on the size of the block, and even in which sectors read / write went.
Here is a very illustrative example of how Nytro Predictor works on a real SAN system. The graphs show very clearly how the cache optimizes I / O as it warms up.
Of course, it is best to look at the results on real tests. Here is an example used during the workshop. To begin with, we measured the performance of the system on RAID0, assembled from 10 Seagate Cheetah 146Gb 15K HDDs. JBOD LSI 620J. Controller: 9266-8i. Platform: SuperMicro X8ST3-F, OS Windows 2008 Server R2
Now, froze after connecting the Nytro MegaRAID 8110, note that the results are more than an order of magnitude better.
Here is a simple example of how SSDs work in large storage systems. If you still have questions, I will be glad to answer them.
If you don’t go into details, the principle of operation of hybrid systems is quite simple (and if you go deep, you can dive so much that you won’t return in a year) and it is the same for all systems, starting with Seagate Momentus XT and Apple Fusion Drive, ending with expensive and complex solutions for large data storage systems and data centers, which will be discussed today.
For the main data storage, traditional hard disks are used, which are necessarily integrated into RAID, and the SSD is used to cache the most frequently used data, which should be accessed most often. The caching system is taken care of, and most often the cache partition in the system is generally not available as a separate disk.
In large and “serious” systems, I / O optimization issues are even more acute than for desktop computers. The fact that for the user is a second delay in launching a heavy program, in the case of a loaded server can result in many thousands of dollars of loss if the disk subsystem becomes a “bottleneck” that slows down all work.
Speaking about large storage systems, one cannot help but recall the LSI company, which, being one of the largest suppliers for storage systems, could not stay away. The LSI product portfolio includes a suite of solutions to accelerate the performance of disk systems in the Nytro family.
The line of these products is divided into three groups.
Nytro WarpDrive Application Acceleration Card
SSD disk, made in the form of PCI-express cards, which can be installed in most modern servers, providing a high-speed disk for the needs of the system.
Nytro XD Application Acceleration Storage Solution A
complex of WarpDrive cards and specialized software that integrates into the system and uses SSDs to cache critical data. This approach allows you to accelerate any block storage system, regardless of the connection method (iSCSI, DAS, SAS, FC or any other). All you need to accelerate is a block device.
Nytro MegaRAID Application Acceleration Card
These cards are essentially RAID controllers built on top-of-the-line LSI RAID-on-chip solutions, with the addition of high-speed NAND Flash. Special algorithms automatically determine the data to be cached, so this solution works independently of the operating system and is completely transparent to it.
To demonstrate in practice how to speed up work with I / O, on April 11, LSI held a seminar or workshop devoted to this topic (details about this event are given below).
Since corporate IT requires a reasonable and balanced approach, before implementing any technology or solution, you need to make sure that it will solve the existing problem. To help sort this out, LSI developed the Nytro Predictor program, which works with two main server operating systems - Windows and Linux. Predictor uses standard statistics collection tools for each system: xperf for Windows and blktrace for Linux, working in the background, completely transparent for user tasks.
To understand Predictor, you need to work for some time in the background, collecting data while the computer performs typical tasks (it is better to give time with a margin). After that, you can get acquainted with the results of his work, and on the basis of this it is already decided what to do.
Now let's go back to the technical seminar and look at all this in practice.
For experiments, we used the test bench of the following configuration:
SuperMicro server on the X8ST3-F platform, OS Windows 2008 Server R2
630J storage system
Nytro MegaRAID 8110 (200 Gb)
Nytro XD BLP4-400 (400 Gb)
MegaRAID 9271-4i
controller 9266-8i controller with CacheVault and CVM01 9286-8e
Controller
Run Predictor and load simulation. A typical load was an MSSQL query processing a lot of data.
We launched the profiler for data collection and then executed the request. If the profiler worked a little time - the results will be of little use
Having collected the data during the server’s operation in its “typical” mode, we give it to the “predictor”. After analyzing them, he renders his verdict.
You can see in detail how read / write requests were distributed depending on the size of the block, and even in which sectors read / write went.
Here is a very illustrative example of how Nytro Predictor works on a real SAN system. The graphs show very clearly how the cache optimizes I / O as it warms up.
Of course, it is best to look at the results on real tests. Here is an example used during the workshop. To begin with, we measured the performance of the system on RAID0, assembled from 10 Seagate Cheetah 146Gb 15K HDDs. JBOD LSI 620J. Controller: 9266-8i. Platform: SuperMicro X8ST3-F, OS Windows 2008 Server R2
Now, froze after connecting the Nytro MegaRAID 8110, note that the results are more than an order of magnitude better.
Here is a simple example of how SSDs work in large storage systems. If you still have questions, I will be glad to answer them.