Comparison of the speed characteristics of disk arrays firms DotHill and NetApp

In this post, we will compare DotHill 3730 and NetApp 5412 disk arrays. Comparison and testing was conducted to determine the feasibility of changing DotHill data storage systems to NetApp.

One of the IT-GRAD customers for many years used DotHill 3730 disk arrays to store all necessary, but it was time to change them. Not that they were bad, just for a few years of operation accumulated "resentment", changed the requirements and wanted a new one.

The requirements for the storage system under test are as follows:

• 20TB user space;
• reliability of storage;
• read and write speed with a block size of 512 bytes of at least 200 megabytes per second.

NetApp storage systems perfectly meet the above requirements, but for greater persuasiveness, testing was organized.
')
To test the disk array, we assembled a stand:

• server (SuperMicro 6027R-TRF / E5-2670 - 2 pcs / 64 GB RAM / QLogic ISP2532 - 2 pcs.) - 1 pc.
• RAID array NetAPP E5412 - 1 pc.
• RAID array DotHill 3730 - 1 pc.

The disk array NetAPP (NetAPP) was tested first, it used one of the two controllers, connected to the server through four FC ports. The DotHill (DotHill) disk array on each controller has 2 ports, used everything to equalize the theoretical bandwidth. Each of the disk arrays was tested separately from the other, connected directly to the server. On each of the disk arrays, the disks were combined into one LUN (RAID-6) and left 1 disk in a hot spare, the LUN was divided into 4 equal parts.

To obtain characteristics of disk arrays, we used the Flexible I / O tester (fio) test package, which performs operations on the specified file, device, or several files or devices at once. Working directly with the device, in our case with four partitions stored on a disk array, allows you to exclude the file system from consideration, as well as reduce the impact on the obtained characteristics of various caches. The following modes are available for testing:

• reading;
• record;
• random reading;
• random entry;
• simultaneous read and write;
• simultaneous random read and random write.

For each of the modes, you can set your own data block size and the number of simultaneous streams. We were mainly interested in the speed of reading and writing, the number of operations per second (IOPs) and other parameters were also recorded. The ability to test multiple devices at the same time, specifying their settings for each, was very useful for simulating the real load.

During the creation of the stand, DotHill was configured via the built-in WEB interface. In the same interface you can get information about its condition. There is little information, but there is a necessary minimum.

NetApp has no WEB interface. To configure it, installation of additional SANTrisity 10.83 software was required. The interface was intuitive, setting up problems did not cause, and access to the characteristics and condition of the piece of iron was just awesome. For the sake of interest, we compared the test results and the data on speeds and IOPs, which were provided by the software in real time. The data matched with decent accuracy.

We decided to start by checking how all the ports and partitions are the same, whether the controller has a pet. Having tortured both disk arrays for a couple of days, there was no serious difference, the spread of characteristics did not exceed 1% of the average value for the test.

Then we decided to check what one section out of four is capable of, running a test suite in all available modes, with data block sizes of 4, 16, 32, 128, 512, 1024, 2048 and 4096 kilobytes. With the number of streams 1, 16, 128 and 1024 for each block size, data lasted just over a week. Of course with the preservation of all protocols.

As it should be, only one section was chased, we took three at once. Three sections were not chosen randomly; in this mode, the operation of a disk array as part of a computing complex is simulated, and the fourth section should store “golden copies”. The test suite did not change much, they refused only to simulate streaming recording (they excluded the size of blocks 1024, 2048 and 4096).

In order not to frighten the public with a huge stream of numbers, which we have collected, here we will give only graphs with the main characteristics.

In the analysis of the data revealed a very interesting point.

During the test, 3 sections in the mode of combined reading and writing with a data block size of 4 kilobytes the difference in performance was eighteen times in favor of NetAPP, both for reading and for writing. Apparently, this difference has arisen due to NetAPP's advanced brains, well optimized to work in this mode.

For the rest of the modes, the E5412 also circumvents the DotHill. Averaging all the data obtained from the tests of the three sections, it turned out that NetApp is about 4 times faster than DotHill.

In terms of administration capabilities, SANTricity provides all the necessary administrative tools. The administration interface from DotHill is inferior to SANTricity in features, especially in terms of collecting performance statistics.

As for the price

The “corporate” price of NetAPP E-5412 at the time of testing is more than 1.6 times the price of the similar DotHill 3730. If the price of NetApp was approximately 3,600,000 rubles, then Dothill cost approximately 2,250,000 rubles. Over the past year since testing, which is about half a year, the new NetApp E-2700 model for people with modest requests has been released. Its recommended value is 1.12 times more than the cost of the DotHill 3730. The E-2700 was built according to the same principles as the E-5400 model, but has a lower controller bandwidth.

We also measured the consumed electrical power during the tests. Got NetApp - 280W, DotHill - 227W.

So, the conclusion is obvious: changing the DotHill equipment to NetApp is fully justified. In terms of price-quality ratio, NetApp is the best-in-class solution.