From owner-freebsd-hackers Mon Dec 3 2: 3:41 2001 Delivered-To: freebsd-hackers@freebsd.org Received: from apollo.backplane.com (apollo.backplane.com [216.240.41.2]) by hub.freebsd.org (Postfix) with ESMTP id 66D1937B416 for ; Mon, 3 Dec 2001 02:03:36 -0800 (PST) Received: (from dillon@localhost) by apollo.backplane.com (8.11.6/8.9.1) id fB3A31S96528; Mon, 3 Dec 2001 02:03:01 -0800 (PST) (envelope-from dillon) Date: Mon, 3 Dec 2001 02:03:01 -0800 (PST) From: Matthew Dillon Message-Id: <200112031003.fB3A31S96528@apollo.backplane.com> To: Danny Braniss Cc: Richard Sharpe , freebsd-hackers@FreeBSD.ORG Subject: Re: Patch #3 (TCP / Linux / Performance) References: Sender: owner-freebsd-hackers@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.ORG :this is what i get, it's between a dell 2540 (dual PIII/900) and an Intel Sl2 :(dual PIII/1g). both are using intel's em driver. : :majadara> ./tbench 1 bagel :Throughput 12.4785 MB/sec (NB=15.5981 MB/sec 124.785 MBit/sec) :... :Throughput 37.002 MB/sec (NB=46.2526 MB/sec 370.02 MBit/sec) : ./tbench 20 bagel :Throughput 37.0574 MB/sec (NB=46.3217 MB/sec 370.574 MBit/sec) : :it starts very nice, but maxes out at about 37MB/sec. Yah. Anything running over GigE is going to be extremely finicky about the cpu/hardware/motherboard design. The best tbench results I got was 47 MB/sec between two DELL2550's(1.1GHz). It was definitely cpu-saturated. 37 MB/sec sounds about right for a 900 MHz box (with everything else being equal). This is actually fairly good performance considering the Big Giant Lock issue with SMP. :btw, before the patch tbench was moving bits (not bytes :-) at turtle speed :(i thought that it was hung, but tcpdump showed some trafffic :-) : :good work! now lets see if nfs/tcp is ok again. : :danny It should be. I now get 7.8 MBytes/sec with a TCP mount and *no* nfsiod's running where before I got buckus. And with nfsiod's I get 12.3 MBytes/sec over 100BaseTX (100% saturation). On the GigE I get 16 MBytes/sec without nfsiod and around 22 MBytes/sec with nfsiod running. That's doing NFS reads... a 'dd' to read a large file over NFS, and very good for a TCP mount. UDP mounts over the GigE get 19 MB/s and 31 MB/s reading, which is also very good for a single-file read. I wish we'd gone after these problems earlier. If we had our ducks lined up in a row we could have had these (trivial) fixes in as early as FreeBSD-4.3 and would have got much more favorable numbers in the several Linux<->FreeBSD comparison articles that have come out in the last six months. Oh well. -Matt Matthew Dillon To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message