Gigabit performance degradation over time?

[MysticLlama]

I'm curious as to if any of you have noticed an occurence like this.

I've been getting ready to get my network upgraded to Gig-E, and also to move to Exchange 2003.

While reading the white paper from MS about OTGs move to Exchange 2003, there was a brief mention that they were having performance degradation problems with their Gig-E NICs, and that they actually swapped them out for 100BT after they moved to the SAN because they didn't need the throughput anymore, and there was less administrative overhead.

Here is the actual comment fromt the white paper:
"OTG?s experience with Gigabit Ethernet showed a gradual trend of network adapter performance degradation. The administration effort required to manage and resolve the degradation was quite time and resource consuming. ... Moreover, the 100 Mbps Ethernet adapters required much less maintenance overhead."

I asked the author of the paper if he had any more details on this (he's a friend of mine) and here is what I got back:
"As I recall during the interviews I had, Gigabit Ethernet adapters regularly suffered from a progressive degradation of throughput. The servers had to be reset to restore full performance. The manufacturers could do nothing to resolve the issue, despite much effort, so since the high-speed throughput was not necessary with the new configuration, OTG went back to 100 Mbps network adapters."

Has anyone else run across this sort of thing in a production environment? Does it maybe have to do with particular adapters, or some bug in Windows?

 

[wlee]

I don't think this is restricted to just Gigabit. I notice "slowdowns" on 100Mbit as well. A power cycle of all switches/routers ( flush MAC/Route Tables ), and reboot of all servers clears it up. I think it's just that the Error tables from packet collision, timeouts,etc. build up over time. I've taken to rebooting servers once a week and powercycling everything else once every couple of months. More often if the opportunity is there.

 

[cmetz]

wlee, there is no reason why you should need to reboot your equipment on a regular basis, especially if it claims to be server/business grade equipment. Anything that needs that is a toy (Windows especially .

MysticLlama, I haven't seen the behavior you described. I'm strongly inclined to blame Windows and suggest if you can spare a couple like boxes to try Linux or BSD on the same hardware and see if that works reliably - unfortunately that will simply point the finger at Windows (Microsoft) or the NIC vendor's driver, neither of which are likely to fix a problem on your account. At least if it turns out to be the NIC vendor, you can replace the NIC with one from someone else without too much trouble. You don't mention what actual hardware was used, making it difficult for anyone to more specifically relate their experiences.

 

[MysticLlama]

I know I don't mention any specific equipment, that's just because I haven't managed to get that info yet. Sometimes it's a little hard to get certain info out of MS, for instance, the tidbit that internally they use BackupExec most of the time, simply because they don't want to alienate other developers.

What I do know is that most of the Exchange systems were based on Compaq hardware, and anything in OTG is typically the most common, well-tested stuff.

Based on those assumptions, I would guess that they were using either 1000BT Compaq Netelligent NICs, or a Server-level Intel or 3Com.

I'll try to figure out what exactly it was, since that information will of course be more useful.

I just wondered if anyone had run into it in testing or anything.