Can somebody interpret my iozone results?

[sofakng]

I'm really confused on how to interpret my iozone results. I'm trying to test out my Windows Storage Server 2008 server (with a Dell PERC 6, using RAID-6).

Here's my spreadsheet (on Google Docs): https://spreadsheets.google.com/ccc?key=0AuSAsm3FdG3pdE5YU1o0UllGUWZyTzBFRkFLbkZRY0E&hl=en

(The third column is the average of the first three columns)

Can somebody tell me how the results look? If I understand correctly the results are in kilobytes per second so if I give by 1024 to get megabytes per second some of the results are nearly 1.0 GB/sec which surely isn't possible on a gigaBIT network.

 

[00Roush]

Your write results are in fact showing higher than 1 GB/sec which is not unusual. It basically has to do with how Iozone times how long it actually takes to complete the writes. Pretty much Iozone is only timing how long it takes to make the write requests to the OS instead of actually timing how long it actually takes to write it to disk. I think using either the -e or the -c command line option helps give results closer to actual.

Your read results look pretty good and similar to what I get between my computers. The problem is my actual transfer speeds are about 90-110 MB/sec. So Iozone is not accurately reporting my throughput. I believe this is due to Iozone being a single threaded program. At least for windows. It can't keep multiple reads or writes in flight at one time like the Windows Vista/Win 7 file copy engine does. For file transfers over a network keeping multiple reads/writes going at one time is the only way to keep a gigabit pipe full due to the latency.

So with Vista and Windows 7 I have been taking the Iozone results with a grain of salt. This is because I don't think Iozone is giving accurate results. I usually test with just a standard drag and drop copy with a very large file of about 17-20 GB. Also I have found Iometer can give good results but it takes a bit to figure out.

00Roush