From owner-freebsd-questions Mon May 31 13:33:40 1999 Delivered-To: freebsd-questions@freebsd.org Received: from spoon.beta.com (mcgovern.ne.mediaone.net [24.218.8.93]) by hub.freebsd.org (Postfix) with ESMTP id F17D814C8D for ; Mon, 31 May 1999 13:33:38 -0700 (PDT) (envelope-from mcgovern@spoon.beta.com) Received: from spoon.beta.com (localhost.beta.com [127.0.0.1]) by spoon.beta.com (8.9.2/8.9.2) with ESMTP id QAA16518; Mon, 31 May 1999 16:33:25 -0400 (EDT) (envelope-from mcgovern@spoon.beta.com) Message-Id: <199905312033.QAA16518@spoon.beta.com> To: freebsd@twin-city.net Cc: questions@freebsd.org Subject: Re: Cyclades Ze help... PLEASE! Date: Mon, 31 May 1999 16:33:25 -0400 From: "Brian J. McGovern" Sender: owner-freebsd-questions@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG Which version of the driver are you using? There have been varying levels of optimization for PPP over the last couple of years... Also, are you running in polled or interrupt mode? If you're running one of the 2.x drivers, are you running with in-host ram or board ram? PPP is probably not the best judge for performance on these boards. They are designed around moving massive amounts of data (more akin to FTP traffic) than sporatic small packets. In polled mode, you'll get data moved once every 1/n seconds, where n is the poll interval. In interrupt mode, you'll get data either when the queues are about 75% full (and I believe they're 4K or 8K queues, so a 56 byte ping packet will rarely trigger it unless you flood ping), or when you get 3 character times without new data (the no new data interrupt). Its quite possible, if you're careful, to miss the NND interrupt, and still not trigger the buffer full for a bit. In the end, I suggest you run 1.18 in interrupt mode, and test with a larger data set for throughput. We've found (through considerably testing) that overall performance has been on par with 16550 UARTS... -Brian To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-questions" in the body of the message