Date: Sat, 7 Feb 2009 15:40:05 GMT From: Martin Birgmeier <martin@email.aon.at> To: freebsd-fs@FreeBSD.org Subject: Re: kern/131360: [nfs] poor scaling behavior of the NFS server under load Message-ID: <200902071540.n17Fe5nN062750@freefall.freebsd.org>
next in thread | raw e-mail | index | archive | help
The following reply was made to PR kern/131360; it has been noted by GNATS.
From: Martin Birgmeier <martin@email.aon.at>
To: bug-followup@FreeBSD.org
Cc:
Subject: Re: kern/131360: [nfs] poor scaling behavior of the NFS server under load
Date: Sat, 7 Feb 2009 16:31:21 +0100 (CET)
I am now very sure that this is an interaction with pppoa, and it is
also worse than I originally thought: it will even lead to failed NFS
transactions for the client.
Here is what I have:
Machine A ('server', a mini home server) does the following:
- connecting to the Internet using usermode ppp over pppoa over an
Alcatel ADSL modem
- NFS serving FreeBSD sources
Machine B does the following:
- Mounting the FreeBSD sources from A (using amd), under directory
/vol/SRC/FreeBSD/HEAD/src
- Compiling the FreeBSD sources: make -j4 buildworld, such that the
corresponding obj is local (via amd again)
Especially in the first phase of the buildworld (clean, depend, obj),
there is a lot of simultaneous NFS traffic from B to A. As soon as
a download is started at A (going via pppoa, of course), the load on
A rises to very high values (> 20 not uncommon).
This may lead to B aborting the compile, it just did that with
"directory not found".
Both machines are running 7.1.0.
No such problem happended when both were running 6.3.0.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200902071540.n17Fe5nN062750>