Irratic performance through switch 1

[MISAdmin]

At our site, we are running an intensive database application. We run the application over a Citrix Published Desktop with the app on the Citrix Server and the Database on a standalone box. To say the least, performance is very bad.

My Citrix Server, for those of you who have any experience with Citrix, is a dual 1.4ghz processored box with 2GB of RAM. ...more than enough for the dozen concurrent users that access it.

Beyond that, everything is connected into two 24 port Linksys Etherfast Hubs. On top of that, the database server is an old Win NT4 box with a 500Mhz processor, 256MB of RAM and a couple slow access 18GB IDE drives.

Knowing all of this, however, running database indexing utilities over the terminal server run painfully slow, almost to the point of timing out. If I run them on the local DB server, though, the run quite quickly. (I do have a 1.8ghz, 1GB ram SCSI disk RAID 5 arrayed server sitting here waiting to replace the 500Mhz box) I don't believe the server is really the point of the bottle neck though since the reindexing apps run fine if run local...

So... I wanted to get the servers out of the two auto-sensing hubs. I added a Netgear FastEthernet 10/100 switch and moved the Citrix Server and DB server into the switch. Once I did that, everything ran blindingly fast... for about 20 minutes. Users were even commenting on how fast the system was. It was very obvious. After about 20 minutes, everything got slow again. About an hour later, things again got faster, and then they slowed off again....

I'm a little puzzled by this. At this point, everything is running much slower, even that before, with the two servers in the switch rahter than the hub. THis statement holds true even for my workstation if I add it to the switch...

Any thoughts on the irratic switch performance? Or thoughts that I might be looking for the bottle neck in the wrong place?

 

[jimbopalmer]

Darn, unmanaged switch means you won't get much info from the switch. Try this, in the network control panels of the servers, set them both to 100 Half duplex, and see if that helps. if it does, try setting just one back to autonegotiate. I suspect one of the server NICs is not negotiating with the switch well. Full duplex is faster, but mismatched is way slower than half duplexed.


(if either NIC is not currently set to auto, try setting everything to auto first) I tried to remain child-like, all I acheived was childish.

 

[MISAdmin]

Well, good... at least we're thinking alike ;-)

I checked all of the servers earlier and they are all set to Auto... I figured it might be a negotiating problem so I had set them all back to 100 Full pending reboot (couldn't reboot during business hours today) It's just after 5:00 here, so I'll go back and set them all to 100 half and start from there...

Thanks.

 

[jimbopalmer]

Yes, Unless there is some sort of configuration ability for the netgear, 100 Meg full is doomed to misconfigure the NICs as you can't also tell the switch to be 100 Full, every thing defaults to half. I tried to remain child-like, all I acheived was childish.

 

[MISAdmin]

I think thats the key. I set all the servers to half (and my workstations that I have going directly into the switch to half) and restarted everything that wasn't W2K. Since then, I consistently load my problem app in 15-17seconds as opposed to the 1 1/2 - 5 minutes that it had been taking. A definate improvement ;-)

The only thing I'm still iffy about is the database server itself... Since I haven't upgraded it to the new server yet, I was unable to configure it's NIC to 100half. Went into NT... properties of NIC (3C905) and it said it couldn't configure the software for this device... So, right now, with no load on the network, everything is good. When a load hits that DB server tommorow... who knows... But I'm not going to lose sleep over it, the new server will take care of that problem.

 

[jimbopalmer]

Auto Negotiate may work on the old server, it usually does. We could have turned it off one at a time while the users were fuming, I just wanted the pain to stop NOW.

Now that it is working you can reset servers back to Auto until you find the one that hates it, your users will let you know when you find it! <G>

Autonegotiating a connection is great when every vender reads the standard in the same way, but on bad days there seem to be many ways to read a standard. I find I turn it off on about 1 out of 25 devices, but I am blessed with managed switches where I can adjust the switch as well. I tried to remain child-like, all I acheived was childish.

 

[MISAdmin]

Sounds like a plan! Again, thanks for your help. I appreciate it.

 

[jimbopalmer]

Auto to Auto works 24 out of 25 times (as a guess) 200 Mbps
Auto to Full Never works right, very slow throughput
Auto to Half always works 100Mbps
Full to Full always works 200 Mbps
Full To Half Never works right, very slow throughput
Half to Half always works 100 Mbps

Without some way to get the switch off of Auto, you only have the first 3 choices

in my experience, if it pings fast but throughput is slow, this is the issue I tried to remain child-like, all I acheived was childish.