Windows server 2008 R2 - Odd Perf results

[Paco80]

Hello everyone,

I'm building myself a NAS mostly to store all my data in one place. In addition, my girl-friend starting her business as a professional photographer, the disks on her machine is filling up pretty quickly. I mostly want my NAS to be small, low power consumption, upgradable and easy to maintain. Good perf is a bonus but not required.

I went for the following components:

• CHENBRO ES34069 (4 drive slots)
• GIGABYTE GA-D510UD (ATOM D510)
• G.SKILL 2GB DDR2
• 2 x Western Digital Caviar Green WD20EARS 2TB (Software Raid 1)

I'm currently using only 2 of the 4 disk slots, I'll buy more disks when I'll need more space.

I started testing with Unraid, FreeNas, Ubuntu and Windows 2008 R2. I quickly decided not to go with FreeNas (at least for now) for it's lack of support for the new Western Digital Advanced Format. I didn't tweak any of the systems I tested.

I just finished tested Win 2k8 and the reason I'm writing this post is because I have odd results.
I test file transfer by copying regular files such as mp3s and big files (up to 5GB) with Robocopy. I do other tests as well with Iozone.

So far, Win2k8 gave me the best results with robotcopy in average, Ubuntu being close. I have 31/53 Write/Read MB/sec for MP3 and 65/77 Write/Read for big files. This could be limited by my test machine (Windows 7, Motherboard Raid 1 Caviar Black).

However for the iozone test, I get intriguing results, transfer rate seems way to high.
Writes: For files of 512MB and lower, I get 1,200 MB/Sec. It goes down to 154MB/Sec and 90MB/Sec for 1 and 2 GB files. Even with 90 MB/sec, it more than twice higher with the other systems.
Read: The read speed seems more realistic with 90 MB/s (70 for 2 GB). FreeNas come second with 70MB/sec average.

Any idea on why the given write speed with iozone is so high?
Any other feedback on my build is more than welcome.

Thanks

 

[Paco80]

I've made some more tests with iozone. The perf counter only show about 106,000,000 bytes/sec max sent on my client which is roughly 100 MB/s.

Not sure why iozone is showing such numbers...

 

[00Roush]

Well you have to understand how Windows deals with writes versus other OSes. From what I recall Iozone uses cached writes. This means that the data is stored in memory until it can be actually written to the disk, network, etc. The problem is that Iozone does not time how long it takes to actually write all of the data. It only times how long it takes to make the write requests to the OS. So for small files that fit into memory writes speeds normally reported by Iozone are very high. I believe using the -c (either that or -e can't remember) option with Iozone should give results that are closer to to actual transfer speeds.

I also want to mention that when testing on Vista and Windows 7 Iozone does not seem to give the same results as file copies. This is due to the fact that both Vista and Windows 7 have upgraded their file copy engine to use asynchronous reads and writes. Also the file copy engine will keep multiple reads and writes in flight at one time which can improve file copies over a network and locally. Iozone does have support for async IO under unix/linux but not windows to my knowledge. So while I have found Iozone a useful tool, with the latest Windows OSes it can provide weird results that do not accurately reflect actual file system performance.

00Roush

 

[thiggins]

You are seeing write cache effects. Use the -c switch in Iozone and you won't see the effect.

 

[Paco80]

Thanks for the info guys, I'll run the test again.