Bechmarking 8 x Intel X25-E RAID-0 on LSI 9260-8i Controller

Post date: Dec 31, 2011 2:02:56 PM

Info

Recently I purchased an LSI 9260-8i controller and 8 Intel X25-E (32GB) based SSDs. The top sequential read speeds were offered by this setup was about 1.6 GiB/sec sequential read with a 1MiB chunk size for the raid. For writes, it varies; however, it is generally between 500MiB/s-1024MiB/s.

Specifications

- LSI 9260-8i (PCI-e 2.0, 5.0GT/s)

- 8 x Intel X25-E 32GB

Setup

1. Windows 7 64-bit

2. RAID-0 (8,16,32,64,128,256,512,1024 chunk benchmarks)

3. Used the default NTFS partition type, no modifcations

Benchmarks

1. Show 1MiB benchmarks.

2. Show 64KiB benchmarks.

3. Show 64KiB vs. 1MiB Crystal Bench Mark test with 1000MiB test file, which is larger than the cache size.

1. The 1MiB chunk size benchmarks

This was the best speed achieved with Everest, when I used a 1MiB chunk size:

(click image for large version)

Crystal disk mark is another popular benchmarking tool; note that when the size is only 100MiB, you see the affects of caching as shown below.

So let's use a bigger chunk size (1GiB), which is more than the cache on the card.

Then, the ATTO benchmark with Direct I/O enabled.

2. The 64KiB chunk size benchmarks

With 64KiB chunk, the read speed is slightly lower:

(click image for large version)

As noted earlier, for test files less than the cache size, Crystal Disk will show inaccurate results, so we will only use a 1GiB test file for this test.

Again, the ATTO benchmark with Direct I/O enabled.

3. The 1024KiB vs. 64KiB benchmarks

Using Crystal Disk and a bigger test file size, this indicates the differences between a 1024KiB and 64KiB chunk size (click for bigger image):

(click image for large version)