View Full Version : benchmark unix to win2k network?
10 October 2004, 02:57 PM
I'm working in a post house which uses a combination of irix flames, windows 2000 3d machines and some osx macs running to a unix file server. The flames are 100 base-t, the windows and macs are gigabit and the file server is fibre running directly to a gigabit switch which connects them all. The thing is though, that the speed is incredibly poor at a lot of times across the network and it seems like the entire thing isn't running at gigabit at all. Mainly we do a lot of network rendering, and pull a lot of after effects sequences from the file server but rarely all at once - even late at night when no one is using the systerm it's still not as quick as I reckon it should be.
Has anyone got any good ways of testing things like transfer rate from a unix file server to a pc hard drive, unix file server to pc memory and also maximum network bandwidth available? It really doesn't seem like were' getting the transfers we should be and I'd love to be able to chase it down to a single bottleneck. How does anyone else usuaully go about doing this?
Thanks in advance,
10 October 2004, 05:14 PM
I think you are oversimplifying it a bit.
You won't find many tools that can test bandwidth from a server to a distant system's memory or even to its hard drive.
You will have to test from server to the PC's data port and then from the data port into the PC and then make other tests.
I can guarantee one of the system's memory isn't slowing the network down. It might be slowing a transfer down from one of the machines to another machine but only because it can't get the data out of the PC quick enough.
However, I have yet to see memory that could slow a direct transfer down. If the direct transfers are being slowed at all from the system itself, they are being slowed by the drives.
When they wired up the network, did they use the correct cabling? This could very well be the issue.
Do some research on the sar command. You can use this from the IRIX machines to determine a few things about them.
A good tool for testing from the Windows machines is Intel's IOmeter. I'm not sure it'll work with the IRIX machines but I don't see why it wouldn't since it just puts a file on the distant machine and reads data from it.
You need a network engineer to come in and look at your network. This isn't something you can just "pick up" and learn. You may cause more damage, if you're not careful.
10 October 2004, 05:38 PM
Apologies for the thread title - it's misleading as you mentioned - It is as you mentioned from a unix hd to the pc - I mentioned memory as it's mainly transfer from the unix hd into pc memory as tga sequences loaded into ram - I wanted to take out the bottleneck of the pc HD to eliminate it from the benchmark. From what I'm told it's all cat 6 cabling and a gigabit switch - all the machines have intel gigabit cards but It really doesn't seem all that fast - I suppose a decent overall bandwidth test to give me the theoretical maximum transfer possible across the network and then the fastest possible from unix to windows would be the two main ones...
I'll check out iometer now :)
10 October 2004, 08:00 AM
There are a 100 different things that can be the problem, everything from the server being only connected through 1 gigibit port, to the gigabit switch being slowed down by having the 100 base T flames connected. Remember 1 Gigabit is at a theoretical max of 120 MB per sec. Personally the fastest I've ever seen a gigabit ethernet network go is 45 MB a second, so thats even more of a worry. Combine that with hitting your central file server for any kind of rendering which involves multiple TCp/IP requests and you're in trouble.
There are a couple of moves you can make the most logical we've found is to keep all the mission critical data on the RAID-5 + secondary back up file system. This is all the scene files, scripts and final outputs. For everything that has any heft to it (read anything that has anything vaguely to do with uncompressed) we use DAS with a raid 0+1 or a raid 5 (direct attached storage). Nothing else has the speed to keep up with our clients.
10 October 2004, 10:31 AM
yeah - it's definitely a pain in the ass - I'll try disconnecting the flames over the weekend andsee what difference it makes - cheers again folks :)
01 January 2006, 06:00 PM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.