TCP-IP switching tip....

[cautery]

Just thought I'd throw this one out to you guys....

If you are using adapters that support "Auto_sense" for speed, like the Netgear FA310TX, DON'T LEAVE THEM ON "Auto"!

I just hooked up my RT314 and put 3 of my workstations directly into the switched ports. Just for grins I ran NetCPS.exe between different combinations of the workstations to verify Full-Duplex 100Mbit ops.... Well, it works of course, but....

With both NICs set to "Auto-sense" I only averaged 7.93 MB/sec throughput...
I explicitly set them to 100 Mbit FULL DUPLEX and my thoughput jumped all the way up to 9.89 MB/sec... pretty good improvement for a 15 second "tweak"!

I have not tested it yet, but I would supect that the same priciple applies to hubs as well.... explicity set your NICs to 100Mbit Half-Duplex to insure that you get best throughput.

Hope this helps...
Clay

 

[Dually]

It shouldn't make a difference, make sure you have the latest drivers, speeds depend more on hub than card in this case.

 

[spidey07]

Some NICs have problems with autonegotiation. (why I praise intels so highly) The best way to determine if you have a speed/duplex mismatch is to look at the errors on the NIC and switch. FCS and CRC and Alignment errors almost always point to duplex mismatch.

One last tip - autonegotiation is only supported when both side of a link are set to autonegotiate link parameters. If you set one side of the link the set the other side as well.

good tip though!

 

[cautery]

The NIC drivers are the latest offered for the FA310TX, and the Netgear has v.3.22 firmware (3.24 just came out last week... I'm about to flash it).

It would indeed seem to me that explicitly declaring the 100Mbit Half or Full Duplex does make a significant difference, since that was the only variable involved in the difference in throughput.

I ran the throughput test 5 times before the change and five times after the change, and the averages were consistently almost 2MB/sec higher afterwards.

I will run the tests again after I flash the switch, and it may well get better, but not likely a large increase as the RT314 is reportedly one of the fastest parts in its class.

Thanks,
Clay

 

[cautery]

Spidey...

That's what I suspect... that the economy FA310TX is just not very good at auto-sensing speed.

They were both set to auto, and now I have set ALL NICs explicitly.

Wish I could afford all new INTEL cards.... I got a great deal on a volume purchase of these FA310TXs a couple of years ago, and still have a few left....

Thanks,

 

[spidey07]

CAUTERY

That a good find you have there. I'm for sure not going to get into the "which NIC is best" hulabulu.

Then again, it might not be the NIC having trouble...it could be the switch. I'll try to find some specs on the RT314

<edit> What a second...you have netgear NICs and netgear switch. I'm amazed that there would be any autoneg problems with the same manufacturer?

 

[cautery]

The FA310TXs are good basic cards, but no where near the quality of the Intel Pro100+ (or the updated one with the encryption processor on board).

BUT, for use in WarpCore they were perfect, since the Beowulf guys had really tweaked a Linux driver for max line-speed for use on clusters....

I'm gonna go flash the switch with 3.24.... Then I'll test again... Wish the RT314 had SNMP like it's Zyxel? brother does! I have boot base 2.00, s I can't use the firmware from it's identical twin....

Thanks for helping out....