performance Bottlenecks?

nel

Not a NP topic but i thought it would be best posted here.

At one of our sites we are experiancing some slowness issues when transfering files between the local site and a remote site. We have a 1gb link between the two sites. Ive checked the netflow on the 1gb wan link locally and its doing 20-50mbps as a rough average. In the lan there's 2 6500's w/ 2x sup720's. Each 3560 connects to the 6500 with a 1gb link. the switches are 3560-48PS and have 10/100 ports which connect to our 7200 on the 1gb link.

Im using cisco network assistant to pull some stats from the 3560 connecting to the wan link. for the bandwidth of the switch (i assume as a whole) is between 100-150mb. When using the link graph on the port connecting to the 7200 its averaging 12-30mb over the 5 sec interval.

Q's:

1. what kind of throughput should i be expecting from the switch shown in the bandwidth graph?
2. ive checked things like duplex mismatches, collisions and so on and havent found anything. are there any other errors worth looking out for?
3. do you think it would be worth trying a gig link between the 3560 and 7200 incase of a bottleneck?
4. i'll get the systems guy to check the server performance.
5. any other suggestions, tips or tools to troubleshoot this type of thing?

Thanks guys.

Forsaken_GA

Couple things could be in play -

#1 How many packets per second are we talking about here? That's one of those gotcha bottlenecks that can catch you by surprise.

#2 What operating systems are you dealing with and are you sure the bottleneck isn't there?

Last week, we had a debian server that was doing 130 mb/s. Pretty good, but there were complaints of slow down (it was a web server, streaming video). We altered the web daemon (lighttpd in this case) to put up more workers and turned off atime on the filesystem, and it's output skyrocketed to 340mb/s. At that point, the server was I/O bound and there was nothing we could do to make it go any faster.

I've seen similar bottlenecks on database servers (ie, someone sends a query that takes so long to run and/or locks tables, that it slows the effective throughput through no fault of the network. Optimizing the queries and doing some indexing of the tables, and magically, throughput increases)