Impressions of HP 6400-series RAID Controllers

[David Sullivan_7]

Has anyone had much experience with the 6402 and 6404 RAID controllers (good/bad)? Some specific questions:

o Would you say they are better performers than software RAID on lower-end systems like the rp3xxx and rp4xxx with external arrays of 8-20 disks? (i.e. they reduce CPU load, I/O wait times, etc.)

o Have you seen demonstrable performance improvements with disk-intensive applications?

TIA!

[Andrew Young_2 · Accepted solution]

Hi David.

The only benchmarking we have done on the 6402 and 6404 controllers has been on DL380 G3's and G4's. We have used a RAID160 controller on an L1500 as well, running HP-UX, but we didn't have anything to compare except to the internal disks. More about that later.

Apart from battery issues which were firmware related, we have only had one problem, which we think was driver related on RedHat ES3, where under one specific circumstance, resizing Oracle 10G tablespaces, would cause a kernel panic, requiring a reboot.

Firstly on the DL380's with our benchmarking we only compared non-cache controllers to the cacheing controllers so I'm not sure how much benefit this will be. Certian file operations performed better than others depending on how often we were able to utilise the read and wite cache. Write cache in particular sped up performance.

Performance varied by OS, with Windows 2000 being the worst, Windows 2003 being better and RedHat ES3 doing the best. Smaller files performed worse than large files but that is to be expected, the problem was the order of magnitude drop in performance on squential reads of small files.

We benchmarked using a single RAID5 volume of 12 disks spread over 2 MSA-30 enclosures with 2 seperate channels. These disks were 146 GB 10K Ultra320 parts.

On the L1500 (rp5450 I think) we got better performance using the RAID controller instead of software mirroring which we used on the internal disks. The problem is the internal disks are Ultra Fast and Wide while the card runs at Ultra160 speeds. Factoring in the difference between read speeds and write speeds we figured we got an improvement of about 6% in write performance. In part this is due to the hardware vs software RAID, but also due to the write cacheing. Since we were doing random access on large database files we didn't think we were getting much benefit from the read cache.

The internal disks were 146GB 10K Ultra Wide parts. The disks in the MSA30 enclosure were 12 x72Gb 15K Ultra320 parts divided into 2 RAID 5 volumes of 6 disks each, each with their own channel. Not however that the card we were using was based on the older 5Si RAID card, this was scaled back to Ultra160 speed.


HTH

Andrew Y