From: Greg Banks Subject: Re: Different NFSSVC_MAXBLKSIZE for udp and tcp clients? Date: Thu, 27 Nov 2003 11:08:06 +1100 Sender: nfs-admin@lists.sourceforge.net Message-ID: <3FC54066.C04EBA49@melbourne.sgi.com> References: <482A3FA0050D21419C269D13989C6113020AC79F@lavender-fe.eng.netapp.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: James Pearson , nfs@lists.sourceforge.net Return-path: Received: from sc8-sf-mx1-b.sourceforge.net ([10.3.1.11] helo=sc8-sf-mx1.sourceforge.net) by sc8-sf-list1.sourceforge.net with esmtp (Cipher TLSv1:DES-CBC3-SHA:168) (Exim 3.31-VA-mm2 #1 (Debian)) id 1AP9hh-0006qV-00 for ; Wed, 26 Nov 2003 16:08:17 -0800 Received: from mtvcafw.sgi.com ([192.48.171.6] helo=rj.sgi.com) by sc8-sf-mx1.sourceforge.net with esmtp (Exim 4.24) id 1AP9hg-0006VU-Ou for nfs@lists.sourceforge.net; Wed, 26 Nov 2003 16:08:16 -0800 To: "Lever, Charles" Errors-To: nfs-admin@lists.sourceforge.net List-Help: List-Post: List-Subscribe: , List-Id: Discussion of NFS under Linux development, interoperability, and testing. List-Unsubscribe: , List-Archive: "Lever, Charles" wrote: > > here's a vote for leaving these values variable rather than > fixed on the server side. 48KB maximum for UDP is great on > a clean network, but some folks would probably prefer setting > it to 8KB to ensure clients on more realistically congested > networks retain reasonable UDP performance. Agreed. > is there any evidence to show that a 4MB TCP maximum will > have benefits over something smaller, like say 1MB ? None at all on Linux. I mention that specific number only because it's the hardcoded maximum on IRIX. However there is evidence that 8K isn't enough. In particular, it's smaller than an Altix client's PAGE_CACHE_SIZE so all writes between two Altix boxes go synchronous. > [...] there's probably a sweet spot where the sum > of both per-op and per-byte costs reach a minimum. Yes. > as > far as i can tell that value is around 16KB in the current > Linux client implementations. The value will be higher for Altix clients and servers, which have more RAM, CPU, and larger pages than the machines you're probably thinking of. Greg. -- Greg Banks, R&D Software Engineer, SGI Australian Software Group. I don't speak for SGI. ------------------------------------------------------- This SF.net email is sponsored by: SF.net Giveback Program. Does SourceForge.net help you be more productive? Does it help you create better code? SHARE THE LOVE, and help us help YOU! Click Here: http://sourceforge.net/donate/ _______________________________________________ NFS maillist - NFS@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/nfs