Has anybody ever benched their un-raid setup read/write performance with one of the drive read/write utilities?

 

When I first setup my basic 100 Mbps windows network I ran something that did a real, simple fairly good sized file write/read across the network.

 

I got the typical 8 MB/sec read/write performance.  I hear it's +10% if using linux, the 10% apparently being the windoze i/o overhead....

 

I haven't yet tried 1000 Mbps but as I understand it, I can expect roughly 20 MB/sec read/write nominal?

 

Understanding people are running all different hardware combos and networks here, what is a 'typical' read/write figure for a nominally functioning UnRaid 1000 Mbps configuration?

 

I have yet to jump in to my own UnRaid rig if you haven't guessed and haven't seen any specific discussion on this which is why I'm asking...

 

--Don

 

Hi Don-

 

We consistently see around 25MB/s read, and around 14MB/s write (between a Windows machine and an UnRaid server).  I'd be interested as well in what others are seeing.  Also, there's a known issue with reading and writing at the same time - not so much with overall throughput, but that writing across GigE while viewing video can cause the video to jitter.

 

While on the subject, I'll soon be seeking volunteers to use a utility that I'll provide to "tweak" the linux buffer cache.  I think the right combination of settings will greatly reduce video jitter which occurs when writing is taking place.  The goal of course is to completely eliminate the jitter, and I think the right buffer cache tuning along with possible UnRaid driver changes will reach the goal.

Hey thanks Tom I was kind of hoping you might see this and comment, that's exactly the kind of information I was hoping for.

 

--Don

Tom, I don't think i'm seeing near that performance but rather than go with a gut instinct here I'd like to do a test. What are you using to test this? I think a utility that didn't rely on Windose or a stopwatch would be ideal If it could show a nice graph of throughput through the test that would also be nice since I seem to see jerky up and down performance. I'm using an onboard GigE port on my desktop through an SMC GigE switch to the array.

 

I'll look for something myself in the meantime. Right now I often have a torrent client R/W to the array while I do other things so I'd have to stop that for accuracy I guess. Getting some real world benchmarks to compare with others sounds like a pretty good idea to me...

 

Edit: Am trying out the tool discussed here -> http://www.tomsnetworking.com/2004/03/02/how_we_test/  One note - when it says to specify a file don't be stupid like I was and specify an existing file. I figured it needed one to work with and thus blew away one of the few P0rn movies I had on the array - doh!

My brother does this type of testing all the time and uses this program.

 

http://www.iozone.org/

 

 

My brother does this type of testing all the time and uses this program.

 

http://www.iozone.org/

 

 

 

That's exactly the tool Tom's hardware uses. I've run the tests once on my box using the same parametwer Tom's does. Howevwer because I have 2Gig of RAM I may not have used a large enough file size based on what I'm reading from the docs. I also had Torrents running in the background doing some slow R/W on a different disk - we know this impacts unRAID performance a great deal. These tests take a good while to run so if anyone is doing them be prepared.

 

Lastly, I haven't a clue how to interpret the data results. It can apparently be put into pretty graphs that knowledgable folks can interpet - I'm apparently not one of those <sigh> Attached is my results file if someone wants to twiddle it. Be advised it's not completely representative of the array's performance sincemy system was lightly tasked doing other things while the test was run and my array was also slightly busy. If this is of great interest to folks I could try to run the test again following the "rules" more closely or if an "easier" tool is available I'd be happy to run that as well.

 

P.S. Ithought it interesting that Tom's apparently doesn't bother to graph GigE's potential throughput on their graphs because they have never had one come close to meeting that kind of speed Their example graph almost meets the 12,500K speed of 100meg ethernet...

 

Edit: I meant to include my results file here - doh! Looking at some of the other tests Tom's has done the results I got are making more sense but I'd like to hear from tohers. We seem to do well in some areas but not so well in others so I'm still a bit confused. I'd be interested in other's running this for comparison's sake and we can compare it to the results found here http://www.tomsnetworking.com/2006/08/08/diy_nas_smackdown/page10.html.

I'd love to see some user benchmarks. I sometimes boot my Windows box with a Linux Live CD (Simply Mepis) and according to that I get transfer speeds of 4MB/s even when I just move stuff from one unRaid drive to another. (My parity check speed also seems to be pretty slow - I'm using the starter kit.)

 

If we could standardise the benchmark tool and some parameters (e.g. unRaid version, network speed, external OS?). I wonder what other factors affect the speed: onboard controller vs. Promise? SATA vs. PATA? brand of disk? disk cache size?

I did another test with nothing accessing the drives and my desktop CPU freed up. I was going to contact the tom's Networking guy but their board is hosed up. If I get a chance I'll post it up here but interpeting it is something I'll leave to others. I sort of understand some of the data but graphing it clearly would probably help <shrug>