Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S265608AbUAGTG5 (ORCPT ); Wed, 7 Jan 2004 14:06:57 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S265611AbUAGTG5 (ORCPT ); Wed, 7 Jan 2004 14:06:57 -0500 Received: from pop.gmx.de ([213.165.64.20]:6347 "HELO mail.gmx.net") by vger.kernel.org with SMTP id S265608AbUAGTGv (ORCPT ); Wed, 7 Jan 2004 14:06:51 -0500 X-Authenticated: #20450766 Date: Wed, 7 Jan 2004 20:06:07 +0100 (CET) From: Guennadi Liakhovetski To: Mike Fedyk cc: linux-kernel@vger.kernel.org Subject: Re: 2.6.0 NFS-server low to 0 performance In-Reply-To: <20040107181920.GN1882@matchmail.com> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 3017 Lines: 84 On Wed, 7 Jan 2004, Mike Fedyk wrote: > On Wed, Jan 07, 2004 at 07:13:46PM +0100, Guennadi Liakhovetski wrote: > > noticed is that 2.6 is trying to read bigger blocks (32K instead of 8K), > > You mean it's trying to do 32K nfs block size on the wire? Emn, no, if I understand it correctly. NFS-client requests 32K of data at a time, but that is sent in several fragments. Actually, client is the same (2.6 kernel), and it requests 32 or 8K depending on the kernel-version of the server... > Just post a few samples of the lines that differ. Any files should be sent > off-list. Well, I am afraid, I won't be able to identify the important differring packets. I did tcpdump -l -i eth0 -exX -vvv -s0 , so, the log contains complete packet dumps. Ok, I'll try just to quote headers. poirot is the server (PC1, 2.4 / 2.6), fast is the client (PC2, 2.6). Following the first request for data (diff only in length) 2.6: 18:42:28.374430 0:80:5f:d2:53:f0 0:50:bf:a4:59:71 ip 162: fast.grange.462443716 > poirot.grange.nfs: 120 read fh Unknown/1 32768 bytes @ 0x000008000 (DF) (ttl 64, id 15, len 148) 2.4: 18:48:57.794687 0:80:5f:d2:53:f0 0:50:bf:a4:59:71 ip 162: fast.grange.1972393156 > poirot.grange.nfs: 120 read fh Unknown/1 8192 bytes @ 0x000002000 (DF) (ttl 64, id 6, len 148) the server (PC1) sends the following packets: 2.6: 18:42:28.374554 0:50:bf:a4:59:71 0:80:5f:d2:53:f0 ip 1514: poirot.grange.nfs > fast.grange.445666500: reply ok 1472 read REG 100644 ids 0/0 sz 0x00007a120 nlink 1 rdev 0/0 fsid 0x000000000 nodeid 0x000000000 a/m/ctime 1073497348.374212040 2477.000000 1064093242.000000 32768 bytes (frag 40553:1480@0+) (ttl 64, len 1500) 18:42:28.374560 0:50:bf:a4:59:71 0:80:5f:d2:53:f0 ip 1514: poirot.grange > fast.grange: (frag 40553:1480@1480+) (ttl 64, len 1500) 2.4: 18:48:57.806270 0:50:bf:a4:59:71 0:80:5f:d2:53:f0 ip 962: poirot.grange > fast.grange: (frag 39126:928@7400) (ttl 64, len 948) 18:48:57.806291 0:50:bf:a4:59:71 0:80:5f:d2:53:f0 ip 1514: poirot.grange > fast.grange: (frag 39126:1480@5920+) (ttl 64, len 1500) Well, maybe important is this place in 2.6 log - when it got the first (2.5s) delay: 18:42:28.414903 1:80:c2:0:0:1 1:80:c2:0:0:1 8808 60: 18:42:31.033837 0:80:5f:d2:53:f0 0:50:bf:a4:59:71 ip 162: fast.grange.479220932 > poirot.grange.nfs: 120 read fh Unknown/1 32768 bytes @ 0x000010000 (DF) (ttl 64, id 18, len 148) 18:42:31.034244 0:50:bf:a4:59:71 0:80:5f:d2:53:f0 ip 1514: poirot.grange.nfs > fast.grange.479220932: reply ok 1472 read REG 100644 ids 0/0 sz 0x00007a120 nlink 1 rdev 0/0 fsid 0x000000000 nodeid 0x000000000 a/m/ctime 1073497351.33807720 2477.000000 1064093242.000000 32768 bytes (frag 40557:1480@0+) (ttl 64, len 1500) So, does it say anything? Thanks Guennadi --- Guennadi Liakhovetski - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/