nvnet vs forcedeth

[fateswarm]

uhm, I'm reading on the forcedeth code file that it's supposed to be experimental compared to nvnet. However, has anyone made any benchmarking on those two? I'm currently back on using nvnet, but I was wondering if it's the right thing.

[Fisch.666]

Hi!

I just want to start a new thread about this question but found this thread here.
What is the difference between the forcedeth driver build in the kernel and the nvnet module which is available in this driver package:

http://www.nvidia.com/object/linux_nforce_1.0-0310.html

Which one is recommended?

Thanks in advance for an reply.

[etymxris]

"experimental" is an overused word in linux. I've been using forcedeth for a while with absolutely no problems. I was worried that it only ran at 100Mbps rather than 1000Mbps since it's in the 100Mbps section of the kernel drivers, but some informal file copies across the network shows that that does not appear to be the case.

[Kevin-B]

Originally Posted by fateswarm

uhm, I'm reading on the forcedeth code file that it's supposed to be experimental compared to nvnet. However, has anyone made any benchmarking on those two? I'm currently back on using nvnet, but I was wondering if it's the right thing.

I'd be interested in seeing some benchmarks.

In terms of stability, I've had no trouble with forcedeth. I've been using it for over a year and a half now with no problems.

[kmare]

Don't know about any benchmarks, but forcedeth is working great here. Also in the yet unreleased kernel 2.6.17 the driver is no more marked as experimental.

[Fisch.666]

Hi!

Thanks for this informations. I've switched from nvnet to forcedeth, because i think forcedeth is more up to date...

Originally Posted by Kevin-B

I'd be interested in seeing some benchmarks.

I've done some benchmarks with Netio from http://freshmeat.net/projects/netio/ (Don't know if this is a good benchmark for this)

Here are my results:

With nvnet:

./netio -t 192.168.178.3

NETIO - Network Throughput Benchmark, Version 1.26
(C) 1997-2005 Kai Uwe Rommel

TCP connection established.
Packet size 1k bytes: 8408 KByte/s Tx, 8259 KByte/s Rx.
Packet size 2k bytes: 8408 KByte/s Tx, 8310 KByte/s Rx.
Packet size 4k bytes: 8373 KByte/s Tx, 8354 KByte/s Rx.
Packet size 8k bytes: 8442 KByte/s Tx, 8337 KByte/s Rx.
Packet size 16k bytes: 8408 KByte/s Tx, 8132 KByte/s Rx.
Packet size 32k bytes: 8453 KByte/s Tx, 8238 KByte/s Rx.
Done.

./netio -u 192.168.178.3

NETIO - Network Throughput Benchmark, Version 1.26
(C) 1997-2005 Kai Uwe Rommel

UDP connection established.
Packet size 1k bytes: 11374 KByte/s (0%) Tx, 11440 KByte/s (0%) Rx.
Packet size 2k bytes: 11412 KByte/s (0%) Tx, 11478 KByte/s (0%) Rx.
Packet size 4k bytes: 11618 KByte/s (0%) Tx, 11667 KByte/s (0%) Rx.
Packet size 8k bytes: 11629 KByte/s (0%) Tx, 11674 KByte/s (0%) Rx.
Packet size 16k bytes: 11634 KByte/s (0%) Tx, 11675 KByte/s (0%) Rx.
Packet size 32k bytes: 11679 KByte/s (0%) Tx, 11705 KByte/s (0%) Rx.
Done.

and with forcedeth:

./netio -t 192.168.178.3

NETIO - Network Throughput Benchmark, Version 1.26
(C) 1997-2005 Kai Uwe Rommel

TCP connection established.
Packet size 1k bytes: 8497 KByte/s Tx, 8290 KByte/s Rx.
Packet size 2k bytes: 8407 KByte/s Tx, 8261 KByte/s Rx.
Packet size 4k bytes: 8372 KByte/s Tx, 8283 KByte/s Rx.
Packet size 8k bytes: 8366 KByte/s Tx, 8333 KByte/s Rx.
Packet size 16k bytes: 8434 KByte/s Tx, 8258 KByte/s Rx.
Packet size 32k bytes: 8440 KByte/s Tx, 8285 KByte/s Rx.
Done.

./netio -u 192.168.178.3

NETIO - Network Throughput Benchmark, Version 1.26
(C) 1997-2005 Kai Uwe Rommel

UDP connection established.
Packet size 1k bytes: 11375 KByte/s (0%) Tx, 11442 KByte/s (0%) Rx.
Packet size 2k bytes: 11403 KByte/s (0%) Tx, 11477 KByte/s (0%) Rx.
Packet size 4k bytes: 11611 KByte/s (0%) Tx, 11667 KByte/s (0%) Rx.
Packet size 8k bytes: 11628 KByte/s (0%) Tx, 11679 KByte/s (0%) Rx.
Packet size 16k bytes: 11632 KByte/s (0%) Tx, 11683 KByte/s (0%) Rx.
Packet size 32k bytes: 11658 KByte/s (0%) Tx, 11689 KByte/s (0%) Rx.
Done.

[kenyee]

Thanks, Fisch. I've been curious about that myself. Now I'm glad I'm using the open source forcedeth driver ;-)

[ismail]

Even Nvidia engineers are now working on forcedeth ( oh the irony ) so its the best choice.

[Lithorus]

Originally Posted by ismail

Even Nvidia engineers are now working on forcedeth ( oh the irony ) so its the best choice.

Actually even nvidia recommend using the forcedeth driver. I think the nvnet driver is only meant for older kernels which doesn't have an up-to-date forcedeth driver. Spending time on getting the forcedeth driver better is a natural choice for nvidia.

[PrakashP]

If using nforce3 and up I'd be interested to know whether cpu utilization is lower with nvnet, as this can do some hw offloading, while forcedeth doesn't.

[Fisch.666]

Hi!

Originally Posted by Lithorus

Actually even nvidia recommend using the forcedeth driver. I think the nvnet driver is only meant for older kernels which doesn't have an up-to-date forcedeth driver. Spending time on getting the forcedeth driver better is a natural choice for nvidia.

Thanks for this info. Maybe this should be noticed somewhere in the readme because i've used the nvnet over an half year...

[am2020]

Just my 2 cents:
I can't get the nvnet driver to work full duplex 100Mbps.
So I am using forcedeth...