Date: Wed, 1 Nov 2000 11:01:30 -0700 From: "Kenneth D. Merry" <ken@kdm.org> To: Jin Guojun <jin@george.lbl.gov> Cc: freebsd-hardware@FreeBSD.ORG Subject: Re: What is the main task chewing CPU in GigE driver Message-ID: <20001101110130.A81809@panzer.kdm.org> In-Reply-To: <200010312241.e9VMfDr07348@portnoy.lbl.gov>; from jin@george.lbl.gov on Tue, Oct 31, 2000 at 02:41:13PM -0800 References: <200010312241.e9VMfDr07348@portnoy.lbl.gov>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, Oct 31, 2000 at 14:41:13 -0800, Jin Guojun wrote: > This is following up a previous SysKonnect/NetGear/Intel GigE discussion > on CPU utilization. I heard people talking that using Jumbo Frame on these > GigE adapters will use less CPU time. Since I have not gotten a chance > to paly the Jumbo Frame on these NICs, and it seems all these cards > have the similar CPU utilization issue, I would like to know which part > of the drive process that chews more CPU when using the smaller MTU than > using the larger MTU (assume without zero memory copy)? The CPU utilization increase will mainly come in the network stack, since it has 6 times the number of packets to deal with for the amount of data transmitted. The per-packet overhead will likely get (mostly) lost in the cost of copying the data from userland into the kernel and vice versa, though. Checksumming is another thing that'll cost CPU cycles if you don't have checksum offloading. Ken -- Kenneth Merry ken@kdm.org To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hardware" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20001101110130.A81809>