Coda File System

The ALINKA Linux Clustering Letter

From: Antoine Brenner <>
Date: Wed, 18 Apr 2001 18:04:07 +0200
Hello All,

Since one year now, the Alinka team has been writing
a weekly summary of the activity in mailing-lists 
relative to linux clustering (such as beowulf, linux
virtual server, linux-ha or journaling file systems) and 
general clustering news.

This "digest" is provided freely as a weekly email
from and is intended to give
a good overview of the linux clustering news for people
without enough time to read all about it.

As the ALINKA Linux Clustering Letter covers this 
mailing-list, and as many of you readers of this list
might not even know that the ALINKA Linux Clustering
Letter exists (We had an annoucement last year on the
mailing-lists, but we did not make any other posts),
I take the opportunity of the first birthday of the
ALINKA Linux Clustering Letter to tell you about it.

Here is a sample of the ALINKA Linux Clustering Letter.
If you like it, you can subscribe to the list: send an 
e-mail to from the address you 
wish to subscribe, with the word "subscribe" in the subject.

-----------The ALINKA Linux Clustering Letter-------------

The ALINKA Linux Clustering Letter, 

Wednesday, April the 11th. 2001
by Antoine Brenner (

Dear readers, I am happy to send you this week's edition of

This edition is the 52th since we started the Letter: this
is also first birthday of the ALINKA Linux Clustering Letter ! is a free weekly e-mail newsletter on linux 
clustering. It provides a summary of the weekly activity in
mailing-lists relative to linux clustering (such as beowulf, 
linux virtual server or linux-ha) and general clustering news.
For more information about ALINKA, see:

News from the High Performance world, by Dr Laurent Gatineau
Tips and tricks from the Beowulf mailing list
* Scott Shealy [m1] seeks comparaison  about P4 and Athlon. Some  CPUs
  benchmarks were given  [m3, m5], but as  Josh Fryman [m2] explain it
  without  a  good compiler benchmarks  could  be not  usefull. In the
  memory side, P4 have a good bandwith, see [m4, m6]

* There was a long  thread about what  tools to use for cloning nodes.
  Some  pointers were  given:   systemimager  [1],  lui  [2],   Scyld,
  kickstart from RedHat, Dolly [3], imgdisk [4], or just copy one hard
  drive on an other one... One can of course use Alinka Raisin [5] !
* Don Morton  [m1] was   looking  for information   about PBS  and PGI
  interactions. He answered  himself [m2]  and  gave a pointer to   an
  online documentation [1]

* J.W.Armstrong [m1] is looking for a parallel make for cluster.  Greg
  Lindahl  [m2]    warn  him because  compiles are     not  always CPU
  bounds. Dean  Carpenter  [m3]  gave  a  link  to ANTS [1]   (a  load
  balancing system).  Andreas  Boklund  [m4] gave results with  MPmake
  and Mosix.

* Javier  Iglesias [m1]  wants to build  a  cluster with two NICs  per
  node, one for administration and  one for computation. He is looking
  for advice.  Robert G.  Brown [m2] answered  that  on a cluster  two
  NICs could not help for some applications.  Peter H. Koenig [m3] and
  Dr O. Parchment [m4] gave two links two similar architecture [1,2].

* Wes  Wells [m1] is  looking for  an alternative  to NFS.  Rob Latham
  [m2]   gave  its opinion  about some   solutions.  Jag [m3] and Greg
  Lindahl [m4] wrote that AFS could have some limitations depending on
  what is needed.

News from the High Availability world, by Guillaume GIMENEZ
* Ard van Breemen said [m1] that drbd and PostgreSQL 7.1
will be made for eachother.

* Mike Debreceni advise us that PostgreSQL before 7.1 is
very slow with protocol C.

* Wensong Zhang announces [m4] ipvs release 0.2.9 for kernel 2.4
is available on LVS website [1].

* Mike keefe posted [m5] a performance evaluation of LVS.

News on the Filesystems front, by  Ludovic Ishiomin
* Clemens Hermann wanted to know if he could use Coda to do mail
Jan Harkes answered that Coda can give surprises on reintegration and
gave the URL of unisson [1] [2m].
* Steffen Neumann succeeded to build a prototype of a backup solution of
a Coda server, but had some concerns about large scale Backup [3m].
* Steffen Schaefer asked if it is possible to change the Venus timeout.
Jan Harles answered that it is possible but not adviseable because
of the protocol used by RPC2 [4m].
* Dermot Cassidy had problems to configure and run its
updateclnt/updatesrv daemon [5m].
* Steffen Neumann gave some tips about management of volumes on Coda and
ACLs [6m].
* Oliver had some configuration questions on how to setup a replicated
volume [7m]. Jan Harkes gave explanations about VSG and how it is 
used on clients and servers [8m].
* David Petterson had problems running procmail (can't read .procmailrc
in the home directory) [9m]. Jan Harkes described its configuration 
that runs without any problems [10m].
[5m] and
the followings.

* Anne Malicia had some questions about page cache management in GFS for
the 2.4 kernel [11m]. It appears she found a bug [12m].
* There was a thread about the implementation of flock() in GFS which
leaves the process in D-state (that increase the load of the machine)
[13m]. Eventually, Mike Tilstra modified the code so that only one 
process is in a D state [14m].
* Gil Shallom asked about Callback policy and pointed that it has a big
impact about performance [15m]. Ken Preslan detailed the one adopted
in GFS [16m].
* David Zafman detailed features of the in-kernel membership service
written by Compaq and announced that it should be released in Open 
Source [17m]. David Teigland enjoyed the announce [18m]. Jeff Darcy
was not convinced by in-kernel membership, and felt that it is better
to stay in user space [19m]. 
* There was a technical thread about integration of IBM DLM and GFS
[13m] and
the followings.
[20m] and
the followings.

* Vladimir Ivaschenko was not able to get to the b1_0_0 branch of the
cvs [21m].
Shirish Patak gave the tip [22m].

* Andreww Klaasseen wanted to know how close to production release is
JFS. Steve Best gave the plans [23m].
* James Lewis Nance reported the same problems he found with earlier
drop when building mozilla with the latest JFS drop [24m].

* Arthur Viak reported problems with vgcreate [25m]. Holger Grothe
asnwered that a kernel upgrade is needed [26m].
* Jonathan DeHaan upgraded from beta5 to beta6 and had problems with
[27m]. Andreas Dilger advices to use beta5 patch with beta5 tools [28m].
* There was a thread about the way to patch 2.2.19 with LVM [29m].
* Jorg de Jong announced lvmgui 0.3 [30m].
* Terje Kvernes announced the development of a LVM library in Perl
* Heinz Mauelshagen announced LVM 0.9.1beta7 [32m].
* Erik Bågfors was not able to remove a logical volume after using a
snapshot [33m]. Heinz Mauelshagen confirmed that it is an old bug
which is fixed in the latest version [34m].

* Brandy Dawkins asked if Oracle runs fine on XFS [35m]. Adam
Cioccarelli related its experience [36m] and Steve Lord added that
direct IO available with XFS should be a great interest to Oracle [37m].
* Stephen Adler reported its very good experience with XFS [38m].
* Ed McKenzie was unable to mount an XFS partition by label. After a bit
of investigation, he had the same problem with ext2 partition, and
concluded that the problem comes from mount and devfsd [39m].
* Ryo Ono wanted to partition a disk with XFS so that it would be
readable on both Irix and Linux [40m]. Nathan Scott recommended 
to use the fx utility on Irix [41m], because fdisk on Linux does 
not handle properly the SGI disklabel [42m], although the kernel 
recognizes themp properly [43m].
* Thomas Graichen asked if LVM snapshoting is planned for XFS. Steve
Lord answered that this feature is planned, bu there still a lot of 
work to stabilize XFS. He also announced that Peter Braam is working on
integrating snapfs and Intermezzo with XFS [44m].
* Lennert Buytenhek found a bug with XFS and named pipe [45m]. Steve
Lord confirmed it and hacked the cvs to correct it [46m].
* Doug wanted to know if CXFS (Clustered XFS filesystem) would be ported
to Linux. Steve Lord announced that a port is planned [47m].
* Justin Tripp reported a weird behavior on a NFS exported XFS
filesystem [48m].
Steve Lord confirmed the problem and sent a patch [49m] which resolved
the problem [50m]. Later Ajay Shekhawat reported a oops on its
NFS server exporting an XFS volume [51m].
* Kenneth Leung tried the new RedHat Wolverine ISO image with XFS
support provided by SGI (0.10 prerelease) and found a problem with 
automount [52m]. Russel Catalan confirmed this is a RedHat Wolverine
issue [53m].
* Petru Paler was unable to compile XFS on a sparc64 box. He had to add
a #include in a XFS file [54m]. Later he reported an oops during XFS
recovery [55m].


To subscribe to the list, send e-mail to from the
address you wish to subscribe, with the word "subscribe" in the subject.

To unsubscribe from the list, send e-mail to from 
the address you wish to unsubscribe from, with the word "unsubscribe" in
the subject.

Alinka is the editor of the ALINKA ORANGES and ALINKA RAISIN 
administration software for Linux clusters. 
(Web site: )
Received on 2001-04-18 11:59:19