Date: Fri, 11 Oct 2002 17:24:42 -0700 From: Terry Lambert <tlambert2@mindspring.com> To: Cheng Jin <chengjin@cs.caltech.edu> Cc: "freebsd-smp@freebsd.org" <freebsd-smp@freebsd.org>, Xiaoliang Wei <weixl@cs.caltech.edu> Subject: Re: SMP + dual GbE Message-ID: <3DA76BCA.B4EF9D79@mindspring.com> References: <Pine.LNX.4.33L0.0210111710110.29183-100000@fast2.cs.caltech.edu>
next in thread | previous in thread | raw e-mail | index | archive | help
Cheng Jin wrote: > > It may be your SysKonnect card. It's very easy to get full wire > > speed with 1500 MTU packets on an 800MHz machine with Tigon III > > cards (for example). It's all in the tuning. > > Sorry, I didn't make myself clear in my previous message. We were able to > get 1 Gbps when we pump data through one SysKonnect card (SMP or not). > The problem is when we try to use both cards under FreeBSD SMP kernel, the > throughput on each card is only 640 Mbps. In fact, using a > single-processor kernel, we could get 560 Mbps going through both cards. > > The cards seem fine, the problem seems to be with the SMP kernel. FreeBSD never routes interrupts to more than a single processor at a time. I believe Linux does the same, though it does not select the CPU to target the same way. NT assigns cards to CPUs to distribute the load. As things currently stand, you should always get better overall numbers on tests like this with a UP FreeBSD system, unless you are willing to take suggestions on your test configuration. 33Mhz x 32 bits ~= 1.32Gbit burst rate; this could be a PCI bus limitation. You didn't say how fast your PCI bus was, whether you were using 64 bit slots for each card, whether they were/were not sharing a PCI interrupt with other cards, etc.. You might also consider contacting the driver's author, since he would best know how to get good numbers out of the hardware. -- Terry To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-smp" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3DA76BCA.B4EF9D79>