Re: Question on Unix FIFOs - why slow when on NFS?

From: Richard L. Hamilton (Richard.L.Hamilton_at_mindwarp.smart.net)
Date: 08/03/04 

Date: Tue, 03 Aug 2004 13:47:10 -0000

In article <cen70t$m1j$1@reader1.panix.com>,
        gerg@panix.com (Greg Andrews) writes:
> v_borchert@despammed.com (Volker Borchert) writes:
>>Casper H.S. *** wrote:
>>
>>|>>I've noticed that a FIFO (that is, a pseudo-file that shows up with a 'p'
>>|>>in the "permission bits") is very slow if it is created on an NFS share.
>>
>>|> The file only exists in memory,
>>
>>Probably a stupid question, but... how can that be? If the server sees
>>it and shows it, it _must_ have gone over the wire?
>>
>
> The filename and inode information went over the wire, yes.
> But the interpretation of the major and minor device numbers
> happens in the NFS client machine's kernel. The mapping of
> the major/minor numbers to a device driver happens in the
> NFS client machine's kernel. The invocation of the device
> driver happens in the NFS client machine's kernel. So reads
> and writes are handled by the device driver in the NFS client
> machine's kernel - not the kernel in the NFS server.
>
> Reading the info about the device file happens over the wire.
> Other I/O happens in the local kernel, not over the wire.

But ISTR the timestamp updates going over the wire. That's very
little volume, but with a slow connection or server, it could still
slow things down. 

-- 
mailto:rlhamil@smart.net  http://www.smart.net/~rlhamil
Lasik/PRK theme music:
    "In the Hall of the Mountain King", from "Peer Gynt"