We have a group of Solaris 8 machines that are having some real network performance issues, they all go through a switch which *should* be set at 100 mb full duplex on every port which goes back to a master switch. I have checked the machines and the NIC cards are all set at 100 full. I had a networking benchmark script that copyed files back and forth from the server and outputted the times it took to transfer them, I did this for a couple of other solaris machines to get something to compare to which arent in the same location as the other group of solaris machines. The windoze admin monkeys had also done the same sort of thing testing performance but looking at the results the windoze machines destroys the Solaris machines for speed. The group of solaris machines had really bad test times whcih were alot worse then the other solaris boxes. I was wondering if there was any test I could do for bottleneck performance and general issues relating to the machines before I confront the network manager. I know theres netstat but I dont really know what im looking for.
Since I'm doing such QA tests all day long, I will ask the following :
what is the arch of those Suns, what is the hardware, what are the type of files that you are doing tests with, how big are they ? These are factors that will give us more clarity on this issue.
You have said "should", you have to be 100% sure for this, because this may be the main problem.
there are few tools you can use that are more graphical like se tools by adrian cockroft but it's comprehensive and covers networks disks, memory cpu. Not sure if se tools are still being distributed haven't used them in quite a while. Try sunfreeware.com
actually I feel a lot of times the boxes we get after we install them comes with very basic default system values which very often you will need to tweak to meet your own performance requirements. I'm pretty sure many folks know this site but it's still worth a read to refresh memory from time to time http://www.sean.de/Solaris/soltune.html
or if you want quick and dirty you can try these values, they r mostly tcp windows sizes, tcp send/receive congestion sizes. cwnd_max should throttle up yur thoroughput.
take a reading of the existing values first before making any changes.
no harm to try find out, but if it is a network problem I think it will also tell the same story on the wintel boxes which could be using the same network.
And if I was the comms guy that's what I will generally say in defence if a unix guy approached me and told me he suspected it's a network problem, I would say "the wintel guys don't have this problem, how come only you guys have it"
most likely the wintel boxes are themselves tuned and have their tcp RWIN values set higher than the unix boxes hence performing a lot better.
It could be an MTU Problem, but i dont know where is this parameter in Sun Solaris, also as what sparcguy said, why you dont just pay some respect to the Windows Admins who seems did there work better than you ?
