RAID 5 on MSA30

[Sentosa]

Dear All,

We are using DS20E with MSA30 (14 x 146.8GB),
Two set RAID 5 were formed with same size.

However, we felt the RAID-5 disk is much slower than single disk drive without RAID.

Could any know any parameter i can check & tune up to make it faster?

Thanks,
Sentosa

[Karl Rohwedder]

Sentosa,

I have no experience with the MSA30, but in general RAID-5 needs a good amount of write cache to deliver good write performance. How is the cache configured?

regards kalle

[Uwe Zessin]

The MSA30 is just a disk drive enclosure (somewhat similar to the BA356, which I am sure you are familiar with, except that it houses SCSI disk drives in a universal carrier, not SBBs
http://h18006.www1.hp.com/storage/disk_storage/msa_diskarrays/drive_enclosures/ma30/index.html
).

Is the MSA30 connected to a PCI RAID controller (size of cache? cache enabled? chunk size?) or are you running host-based RAID?

What I/O patterns are you using for testing?

[Hoff]

RAID-5 has a very nasty sequence within its failure mode, where applications will see massively degraded I/O throughput (on the order of 20% of normal throughput remains available), and where the RAIDset is vulnerable to a double disk failure, and a double failure is only recoverable from off-line data archives; a roll-in and recovery from the most recent disk BACKUP.

The I/O hammering that transpires during the expended RAID-5 recovery can itself trigger a second disk failure. Real-world data shows that secondary failures here are surprisingly common within the RAID-5 recovery window.

Given the continually-decreasing price of disk spindles, I now tend to prefer RAID-10. The incremental savings in disks between RAID-5 and RAID-10 aren't locally viewed as being worth (to my data) the costs of the failures. If I'm going to the effort of running RAID, I want my data on-line.

http://64.223.189.234/node/410
http://64.223.189.234/node/188

Stephen Hoffman
HoffmanLabs LLC

[Guenther Froehlin]

"However, we felt the RAID-5 disk is much slower than single disk drive without RAID."

Depends what you do.

For reads the RAID-5 disk should be faster if you have a multi-stream read application (i.e. multiple concurrent read I/Os). The 'striping' feature of RAID-5 can use more drives and hence improves overall read performance. It has no effect if you use a single stream application (one I/O at-a-time).

Writes need more work (overhead). Before writing data the old parity blocks have to be read in memory. The parity data is updated and written back and then the new data is written back. You typically have 3 times more I/O than on a simple disk. With enough and smart controller cache you can mostly avoid the two extra I/Os.

Is this Host Based RAID?

/Guenther

[John Gillings]

>Could any know any parameter i can check &
>tune up to make it faster?

Yes, tune the "5" into a "1". Get rid of RAID-5 and use RAID-1 or (better) host based shadowing.

RAID-5 is a technology past its used by date. It was a great idea when disks were small and expensive, but now they're large and cheap, the tradeoffs and economics are all wrong.

Using RAID-5 saves you some hardware costs, but you pay for it in performance. Since the hardware is much cheaper than time, you're better off just shadowing everything. Yes, it's slightly more expensive in hardware, but it's MUCH better in terms of data protection and performance.

[Sentosa]

Thanks