From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-ew0-f46.google.com ([209.85.215.46]:32889 "EHLO mail-ew0-f46.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752112Ab1DQNVY (ORCPT ); Sun, 17 Apr 2011 09:21:24 -0400 Received: by ewy4 with SMTP id 4so1139340ewy.19 for ; Sun, 17 Apr 2011 06:21:23 -0700 (PDT) In-Reply-To: References: Date: Sun, 17 Apr 2011 22:21:22 +0900 Message-ID: Subject: Re: NFSv3 client caches dozens or hundreds of megabytes before sending a write request to NFS server. From: crocket To: linux-nfs@vger.kernel.org Content-Type: text/plain; charset=ISO-8859-1 Sender: linux-nfs-owner@vger.kernel.org List-ID: MIME-Version: 1.0 It seems "noac" mount option enables direct write to the NFS server. Now the NFS client doesn't cache but write directly to the server. Please refer to "man nfs" if you use linux. On Sun, Apr 17, 2011 at 6:23 PM, crocket wrote: > Let me explain my network configuration. > > Internet <--------------------> eth0 | the linux router | eth1 > <------------------------------> eth0 | the desktop. > > The linux router is actually a proper computer but not a router that > has linux firmware. > eth1 on the router and eth0 on the desktop are both full-duplex at 100Mbps. > The full-duplex connection is confirmed using netcat like > http://imagebin.org/148650 > And they also remain full-duplex during NFS transfers according to "ethtool". > > The router runs an NFS server, and the desktop uses an NFS client. > > On the desktop, if I download files from the internet to the NFS > server, the client caches dozens or hundreds of megabytes before > sending data to the server like http://imagebin.org/148648 > As a result, the download pauses frequently to send data to the NFS > server until the download is finished. > I want NFS client to forward downloaded data to the server immediately > instead of after caching. > In other words, I want NFS transfer to be http://imagebin.org/148650 > but not http://imagebin.org/148648 > > Does anybody know how to solve this issue? >