Lists wrote:
yeah, do a tcp thruput test between two boxes and tell me if you get more then 800mbs, then slap intel cards in the box and do it again, been there done that granted its better then it was in 5.x but still nothing close to what it should be for the bandwidth your giving up, let me know cause we tested 6.0 2 months ago and broadcom cards were running 670mbps tops.
I didn't say I performance tested it, I said it works reliably thus far. For the purposes of this box, it could be a 100 Mb NIC for all it matters, so I haven't had any need to assess network throughput. This really isn't a great test, since this isn't an isolated network environment, but it's accurate enough that I'll post the numbers. All Dell PowerEdge 2550's, dual P3 1.26 GHz. OS's tested are Windows 2000 Server SP4, with all patches and latest drivers, Windows Server 2003 SP1 with all patches and latest drivers, and FreeBSD 6.0-RELEASE. The 2550's have an onboard Broadcom BCM5700. Between a Windows 2000 and 2003 box, I can get about 450-500 Mb with iperf. Sitting at about 30-40% CPU utilization. Same if I only use one 2550 and connect it to a much bigger box, like a 2850 dual 3.6 GHz with Intel gig cards. From FreeBSD to any other box, I can get about 325-375 Mb. Granted, this FreeBSD box is about 60-70% loaded without running the test, capturing a bunch of packets off the onboard fxp card, and processing them through ntop, amongst a number of other duties. This is a busy box. The Windows boxes are relatively idle. I don't have anything better I can test right now, and I don't want to mess any further with the one production FreeBSD box. Given the numbers, and the load on the FreeBSD box that was not on the Windows boxes, I'd say that's a roughly even comparison. FreeBSD might be a little slower than Windows with the bge cards, but there certainly isn't a substantial difference. The best conclusion I could draw would not be "Broadcom on FreeBSD sucks", it'd be "Broadcom sucks". I expected much better on Windows, assuming it was a FreeBSD driver issue or a load related issue, but that seems to not be the case. I don't have any Intel PCI-X gig cards I could put in these boxes to compare, but I have no doubt they'd be faster and never disputed that. Intel cards are without a doubt the best choice. The only test I can come up with using all Intel gig cards is *way* different, but I get 900+ Mb through Intel PRO/1000 cards between two PowerEdge 2850's with Server 2003 SP1, dual Xeon 3.6 GHz, only 7-9% CPU utilization at that. I'd expect no different from FreeBSD.
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
