Date: Wed, 9 Mar 2005 04:15:51 +0000 From: Joseph Koshy <joseph.koshy@gmail.com> To: Peter Hessler <phessler@coverity.com> Cc: freebsd-performance@freebsd.org Subject: Re: NFS small files Message-ID: <84dead720503082015c11bb53@mail.gmail.com> In-Reply-To: <20050309015119.44566908B7@coverity.dreamhost.com> References: <20050309005628.5C6D1908B8@coverity.dreamhost.com> <422E541C.8040108@telus.net> <20050309014632.49895908B7@coverity.dreamhost.com> <20050309015119.44566908B7@coverity.dreamhost.com>
next in thread | previous in thread | raw e-mail | index | archive | help
> I should mention that copying the same directory on the nfs server takes Filename resolution over NFS involves multiple round-trip requests, one for each component of the file's pathname. For example, assuming that '/mnt' is an NFS mount-point, accessing a file named '/mnt/FreeBSD/src/sys/sys/file.h,v' results in the following NFS requests: [Edited output from ethereal] 1 0.000000 ::1 ::1 NFS V3 LOOKUP Call (Reply In 2), DH:0xa9a212b8/FreeBSD V3 Procedure: LOOKUP (3) what: dir Name: FreeBSD 2 0.000116 ::1 ::1 NFS V3 LOOKUP Reply (Call In 1), FH:0xcd6610c4 Status: OK (0) 3 0.000162 ::1 ::1 NFS V3 LOOKUP Call (Reply In 4), DH:0xcd6610c4/src V3 Procedure: LOOKUP (3) what: dir Name: src 4 0.000194 ::1 ::1 NFS V3 LOOKUP Reply (Call In 3), FH:0x5c9e0ec6 Status: OK (0) 5 0.000224 ::1 ::1 NFS V3 LOOKUP Call (Reply In 6), DH:0x5c9e0ec6/sys V3 Procedure: LOOKUP (3) what: dir Name: sys 6 0.000253 ::1 ::1 NFS V3 LOOKUP Reply (Call In 5), FH:0xb4d2113a Status: OK (0) 7 0.000281 ::1 ::1 NFS V3 LOOKUP Call (Reply In 8), DH:0xb4d2113a/sys V3 Procedure: LOOKUP (3) what: dir Name: sys 8 0.000312 ::1 ::1 NFS V3 LOOKUP Reply (Call In 7), FH:0xb0f9f738 Status: OK (0) 9 0.000338 ::1 ::1 NFS V3 LOOKUP Call (Reply In 10), DH:0xb0f9f738/file.h,v V3 Procedure: LOOKUP (3) what: dir Name: file.h,v 10 0.000366 ::1 ::1 NFS V3 LOOKUP Reply (Call In 9), FH:0x58ea0f3a Status: OK (0) 11 0.002220 ::1 ::1 NFS V3 ACCESS Call, FH:0xa9a212b8 V3 Procedure: ACCESS (4) object access: 0x3f I can imagine that the extra latency induced by these round-trips would be adding to elapsed time for your test. -- FreeBSD Volunteer, http://people.freebsd.org/~jkoshy
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?84dead720503082015c11bb53>