I/O behaviour on MSA20

[JBartlett]

I am working with a customer who has an MSA20 (12 x 750GB SATA) in a single RAID 5 config. The MSA20 is hanging off of an SA6404 in an HP DL380 G5. Using Backup Exec CPS, we noticed 2 distinct "states" that the I/O produces (using Windows 2003 perfmon):

1200 disk transfers/sec
85ms write latency
120ms read latency
17 MB/sec throughput

600 disk transfers/sec
10ms write latency
8ms read latency
72MB/sec throughput

The CPS server is doing the same job in both situations. What is the "nature" of the I/O that would cause such odd numbers? Why does only 17MB/sec cause huge delays and transfers, while 72MB/sec is fine? We haven't seen the first situation lately, it seems to have settled down a bit. Thanks for any insight.

[Eric de Lange (MSE)]

Hello J,

The states are likely because of the backup workloads. This is expected behavior in my view.

When you see lots of transfers and low throughput, Backup Exec may be backing up user shares or a source that contains smaller files. 1k to 32k or so. This is when the disk heads have to do a lot of 'seeking' which costs time.

When the amount of tranfers goes down and the throughput goes up is usually when large files are being backed up (large being multiple megabytes per file and higher)

To do some comprehensive tests, try testing with HP Library and Tape Tools. LTT allows you to build a 'known' data set on your filesystem containing known filesizes that you can use to do perfromance testing.

Note that 750GB disks are built for 'capacity' and not neccesarily for 'performance'.

Hope this brings some clarification.

-Eric