On Fri, Aug 29, 2008 at 12:45 PM, Daniel Gruner <[EMAIL PROTECTED]> wrote:
>
> On 8/29/08, Abhishek Kulkarni <[EMAIL PROTECTED]> wrote:
> >
> >
> >
> > On Fri, Aug 29, 2008 at 10:13 AM, Daniel Gruner <[EMAIL PROTECTED]>
> wrote:
> > >
> > > Hi Ab
> > >
> > >
> > > On 8/29/08, Abhishek Kulkarni <[EMAIL PROTECTED]> wrote:
> > > > Hi Daniel,
> > > >
> > > > Understand the way in which XCPU is supposed to integrate with oneSIS
> > and/or
> > > > Perceus. It uses these as a "launch vehicle" to build minimal images
> > with
> > > > xcpufs running on them, and provision the nodes with these images. In
> > the
> > > > best case, that's all that you need to be running on the compute
> nodes.
> > >
> > > I understand.
> > >
> > >
> > > >
> > > > On Fri, Aug 29, 2008 at 8:46 AM, Daniel Gruner <[EMAIL PROTECTED]>
> > wrote:
> > > > >
> > > > > Hi Greg,
> > > > >
> > > > > I definitely have additional questions! :-)
> > > > >
> > > > > Ok, here we go:
> > > > >
> > > > > - assume I am totally new to this - what would one do in order to
> set
> > > > > up a perceus/xcpu cluster?
> > > >
> > > > As Greg said, you have two ways to go about it. You could choose
> either
> > of
> > > > them or try both to see what works for ya. It's just a matter of
> playing
> > > > with different configurations and rebooting your nodes to try them.
> > > >
> > > > >
> > > > >
> > > > > - now, I am not totally new to this game, and my background is with
> > > > > bproc clusters, so I would like to have a replacement for these,
> but
> > > > > with the same basic principle of having a minimal node
> installation,
> > > > > and basically no management of nodes needed. I definitely do not
> want
> > > > > to go to a model where the nodes have password files, and you ssh
> into
> > > > > them in order to run your codes.
> > > > >
> > > > > - in the caos-NSA installation, the warewulfd is started by
> default.
> > > > > I assume it needs to be stopped and perceus started, correct?
> > > >
> > > > You can enable Perceus from "sidekick" in NSA. Warewulf focuses on
> > cluster
> > > > monitoring starting with 3.0.
> > >
> > > Ok, I am concentrating on my RHEL5 machine for now. It seems to be
> > > working, at least insofar as the nodes boot. I haven't been able to
> > > contact them to try to do anything, other than running xstat with a
> > > positive response:
> > >
> > > n0000 tcp!10.10.0.170!6667 /Linux/x86_64 up 0
> > > n0001 tcp!10.10.0.185!6667 /Linux/x86_64 up 0
> > >
> > > I'd like the nodes to get sequential IP addresses, for ease of
> > > identification and management, and I have yet to find out how you do
> > > that in perceus.
> >
> > Take a look at the ipaddr module in Perceus.
>
> Thanks for the pointer. I am looking at it, but the manual leaves a
> lot to be desired in terms of describing what each module does and how
> to configure them. I'll try the perceus list if I keep getting stuck
> on this.
>
> >
> > >
> > >
> > > Now, when I try to do anything on the nodes I get, for example:
> > >
> > > xgroupset 10.10.0.170 root 0
> > > xgroupset: Error: Connection refused:10.10.0.170
> >
> > Whoops! What about telnet 10.10.0.170 6667?
> > Perceus might possibly be running xcpufs on some non-standard port. I'm
> not
> > sure about that but I remember seeing something like that a while back.
>
> You seem to have hit it! I can in fact telnet using port 6667
> explicitly (can't do anything while in there...:-). I thought that
> was the default port anyway, correct?
>
> >
> > >
> > >
> > > similarly with xrx.
> > >
> > > xrx 10.10.0.170 /bin/date
> > > Error: Connection refused:10.10.0.170
> >
> > Ditto with this, if it's running on a different port you would want to do
> > xrx 10.10.0.170!port /bin/date
> >
> > Alternatively you could specify the "-a" flag to retrieve the nodes from
> the
> > statfs.
>
> I can get xgroupset and xuserset to work with the -a flag, with no
> complaints. However, when I try to run anything on the nodes using
> xrx, whether I use the -a flag or explicitly set the port (xrx
> 10.10.0.170\!6667 /bin/date) the command just hangs. I realize this
> is some progress, but no cigar yet.
Weird. Did you use the above xrx command as-is? If yes, notice the
back-slash in your command.
I have not ever seen xrx hang that way. Try passing the "-d" switch
alongwith -a, to enable the debug mode.
If not, do the strace and post the output. It's not that tough. Really. It
should just work.
>
>
> >
> > >
> > >
> > > I also don't get name resolution for the nXXXX names assigned to the
> > > nodes by perceus.
> > >
> > >
> >
> > Check your /etc/resolv.conf.
> > Probably try adding the following to it.
> > nameserver 127.0.0.1
> >
> > If that doesn't work, the right place to ask this would be the Perceus
> ML.
>
> Doesn't work. I'll try the perceus gurus...
>
> >
> > >
> > >
> > >
> > > >
> > > > >
> > > > >
> > > > > - what initialization of perceus needs to be done (the first time
> it
> > > > > runs)? I know about the network interface specification, and that
> I
> > > > > want it to use xget (the default), but is running the "perceus
> module
> > > > > activate xcpu" enough to get the nodes booting into xcpu?
> > > >
> > > > Yes, it is enough to get xcpufs running on the compute nodes.
> > > >
> > > > >
> > > > >
> > > > > - what about configuring the resource manager (e.g. slurm) for use
> in
> > > > > the perceus/xcpu environment?
> > > >
> > > > XCPU only supports Moab Torque for now.
> > >
> > > Is this the open source torque, or just the commercial product?
>
>
> Who would know about which version of Torque this is? I can't afford
> the commercial Moab right now...
>
>
> > >
> > >
> > >
> > >
> > > >
> > > > >
> > > > >
> > > > > - I don't see the xcpufs and statfs daemons running on the master
> > > > > after starting perceus even though I told it to activate xcpu. I
> > > > > haven't tried to boot nodes yet, but I'd like to understand what I
> am
> > > > > doing first (I hate black boxes...).
> > > > >
> > > >
> > > > You shouldn't need to run xcpufs on the master. As for statfs, you
> can
> > start
> > > > it manually if it is not running already.
> > > >
> > > > Again, considering that you have fully configured the master and have
> > the
> > > > nodes provisioned to the init state, this is what I would do to
> generate
> > my
> > > > statfs.conf --
> > > >
> > > > perceus node status | awk 'NR > 2 {print $1 "=tcp!" $3 "!6667"}' >
> > > > /etc/xcpu/statfs.conf
> > >
> > > I had to replace the part "NR>2" with "NR>0" for the above incantation
> > > to work (??).
> >
> > Strange, I might probably be running a different version of Perceus.
>
> Actually, what happens is that the first two lines of output from the
> "perceus node status" command are output to stderr, and the rest to
> stdout. If your '|' redirection included stderr then the command as
> you wrote it would work. What shell are you using?
>
Ahh, maybe. I'm using bash. I didn't realize that.
Thanks,
-- Abhishek
>
>
> Thanks,
>
> Daniel
>
>
>
> >
> > >
> > >
> > >
> > > >
> > > > And then,
> > > >
> > > > statfs -c /etc/xcpu/statfs
> > >
> > > statfs seems to work. Here is the output from xstat:
> > >
> > > n0000 tcp!10.10.0.170!6667 /Linux/x86_64 up 0
> > > n0001 tcp!10.10.0.185!6667 /Linux/x86_64 up 0
> > >
> > > In any case, there is some progress, but it is not quite there yet...
> >
> > I'm glad you are almost there.
> >
> > Thanks,
> > -- Abhishek
> >
> >
> > >
> > >
> > > Thanks,
> > > Daniel
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > > >
> > > >
> > > > >
> > > > > etc.
> > > > >
> > > > > I guess the main problem I have is not with perceus itself (I have
> > > > > read the manual), but rather with its integration and provisioning
> for
> > > > > xcpu, and for the subsequent configuration of those pieces that
> make
> > > > > the cluster useable in a production environment.
> > > > >
> > > > >
> > > > > Thanks for your help,
> > > > > Daniel
> > > >
> > > > Thanks
> > > > -- Abhishek
> > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > On 8/29/08, Greg Kurtzer <[EMAIL PROTECTED]> wrote:
> > > > > >
> > > > > > You have multiple choices on how to move forward.
> > > > > >
> > > > > > First you can run the xcpu Perceus module like:
> > > > > >
> > > > > > # perceus module activate xcpu
> > > > > >
> > > > > > That will interrupt the node provisioning process and instead of
> > > > > > copying the VNFS to the node it will just start up xcpu and
> start
> > > > > > accepting connections.
> > > > > >
> > > > > > The second option would be to run xcpu from within the VNFS of
> your
> > > > > > choice. That mechanism basically involves installing xcpu into
> the
> > > > > > mounted VNFS image and then provision your nodes with that.
> > > > > >
> > > > > > Let me know if that helps or if you have additional questions.
> :)
> > > > > >
> > > > > >
> > > > > > Greg
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > > On Fri, Aug 29, 2008 at 6:45 AM, Daniel Gruner <
> [EMAIL PROTECTED]>
> > > > wrote:
> > > > > > >
> > > > > > > Hi Kevin,
> > > > > > >
> > > > > > > Well, I've just completed installing xcpu2 and perceus into my
> > RHEL5
> > > > > > > machine, but now I am stumped with the configuration. How do
> you
> > > > tell
> > > > > > > perceus that you want your cluster to run xcpu? I sure don't
> > > > > > > understand where this is configured (I assume somewhere in the
> > > > > > > /etc/perceus .conf files), and there is no mention of that in
> the
> > > > > > > manual other than saying that xcpu works.
> > > > > > >
> > > > > > > If you install xcpu2 you surely would need 9p, right?
> > > > > > >
> > > > > > > Also, how does slurm integrate into the perceus/xcpu world?
> > > > > > >
> > > > > > > I have also installed this on a caos-NSA test machine, but
> again
> > I
> > > > > > > don't know how to configure the provisioning.
> > > > > > >
> > > > > > > Any help with this would be much appreciated...
> > > > > > >
> > > > > > > Daniel
> > > > > > >
> > > > > > >
> > > > > > > On 8/28/08, Kevin Tegtmeier <[EMAIL PROTECTED]> wrote:
> > > > > > >> We used RHEL5 + perceus successfully. I had to modify the
> > perceus
> > > > boot
> > > > > > >> image for x86_64, but it may have been a kexec/hardware
> specific
> > > > issue I ran
> > > > > > >> into. If you run into an issue with it I can help you along.
> > > > > > >>
> > > > > > >> I don't think the 9P module was built in, but I don't think
> you
> > > > would use
> > > > > > >> it.
> > > > > > >>
> > > > > > >>
> > > > > > >> On Thu, Aug 28, 2008 at 11:31 AM, Daniel Gruner
> > <[EMAIL PROTECTED]>
> > > > wrote:
> > > > > > >>
> > > > > > >> >
> > > > > > >> > Thanks, Abhishek.
> > > > > > >> >
> > > > > > >> > I will try it and report on my success/lack thereof.
> > > > > > >> >
> > > > > > >> > Just for info, I am using a RHEL5 distribution, but with
> the
> > > > 2.6.26
> > > > > > >> > kernel so that it supports 9p. Has anybody been successful
> > with
> > > > this
> > > > > > >> > distribution? Otherwise, is there a preferred one?
> > > > > > >> >
> > > > > > >> > Daniel
> > > > > > >> >
> > > > > > >> >
> > > > > > >> >
> > > > > > >> >
> > > > > > >> > On 8/28/08, Abhishek Kulkarni <[EMAIL PROTECTED]> wrote:
> > > > > > >> > >
> > > > > > >> > > Daniel,
> > > > > > >> > >
> > > > > > >> > > It is _not_ necessary to install cAos Linux to use
> Perceus.
> > > > Perceus
> > > > > > >> > > supports most, if not all, distributions.
> > > > > > >> > >
> > > > > > >> > > XCPU is bundled up as a module within Perceus. The
> > > > documentation at
> > > > > > >> > >
> > > > > > >>
> > > > http://www.perceus.org/docs/perceus-userguide-1.4.0.pdf
> > is
> > > > > > >> quite
> > > > > > >> > > extensive at that and has details on importing and
> > activating
> > > > modules.
> > > > > > >> > > It's quite simple even if you find yourself wanting to
> > tinker
> > > > with the
> > > > > > >> > > XCPU Perceus module (it's just a shell script that runs
> at
> > a
> > > > specified
> > > > > > >> > > provisioning state/level)
> > > > > > >> > >
> > > > > > >> > >
> > > > > > >> > > -- Abhishek
> > > > > > >> > >
> > > > > > >> > >
> > > > > > >> > > On Thu, 2008-08-28 at 14:17 -0400, Daniel Gruner wrote:
> > > > > > >> > > > Yes, that is a possibility. Instructions on that,
> > please?
> > > > > > >> > > > I tried installing caos linux, but it doesn't quite
> > finish
> > > > doing the
> > > > > > >> install.
> > > > > > >> > > >
> > > > > > >> > > > Daniel
> > > > > > >> > > >
> > > > > > >> > > > On 8/28/08, ron minnich <[EMAIL PROTECTED]> wrote:
> > > > > > >> > > > >
> > > > > > >> > > > > Use perceus.
> > > > > > >> > > > >
> > > > > > >> > > > > Ron
> > > > > > >> > > > >
> > > > > > >> > > > >
> > > > > > >> > > > > On 8/28/08, Daniel Gruner <[EMAIL PROTECTED]>
> wrote:
> > > > > > >> > > > > >
> > > > > > >> > > > > > Hi All,
> > > > > > >> > > > > >
> > > > > > >> > > > > > The list has been very quiet lately... :-)
> > > > > > >> > > > > >
> > > > > > >> > > > > > I've been trying, yet again, to install the
> latest
> > xcpu2
> > > > in a
> > > > > > >> test
> > > > > > >> > > > > > cluster. Ron's instructions on the xcpu.orgsite
> > seem
> > > > to be
> > > > > > >> outdated,
> > > > > > >> > > > > > and partly buggy too. For instance, here are a
> > couple
> > > > of
> > > > > > >> points:
> > > > > > >> > > > > >
> > > > > > >> > > > > > - After doing:
> > > > > > >> > > > > >
> > > > > > >> > > > > > make xcpu-tarball
> > > > > > >> > > > > >
> > > > > > >> > > > > > make ramfs-tarball
> > > > > > >> > > > > >
> > > > > > >> > > > > > make install
> > > > > > >> > > > > >
> > > > > > >> > > > > > I don't know whether xcpu2 has actually been
> built
> > (I
> > > > suspect
> > > > > > >> not),
> > > > > > >> > > > > > and it certainly has not been installed (e.g. no
> > xrx, or
> > > > xcpufs,
> > > > > > >> or
> > > > > > >> > > > > > any of that stuff has been installed).
> > > > > > >> > > > > >
> > > > > > >> > > > > > - The command
> > > > > > >> > > > > >
> > > > > > >> > > > > > export u=`uname -r`
> > > > > > >> > > > > > ./mk-initramfs-oneSIS -f initrd-$u.img $u -nn -rr
> \
> > > > > > >> > > > > > -o ../overlays/xcpu-64 \
> > > > > > >> > > > > > -w e1000 \
> > > > > > >> > > > > > -w forcedeth \
> > > > > > >> > > > > > -w ext3
> > > > > > >> > > > > >
> > > > > > >> > > > > > should really be
> > > > > > >> > > > > >
> > > > > > >> > > > > > ./mk-xcpu-oneSIS ....
> > > > > > >> > > > > >
> > > > > > >> > > > > > in order that the 9p and 9pnet modules get loaded
> > into
> > > > the
> > > > > > >> initrd.
> > > > > > >> > > > > >
> > > > > > >> > > > > > Can someone please take a look and revise the
> > > > instructions (and
> > > > > > >> let us
> > > > > > >> > > > > > mere mortals know what to do)?
> > > > > > >> > > > > >
> > > > > > >> > > > > >
> > > > > > >> > > > > > Furthermore, is xcpu2 actualy useable for
> production
> > > > work? What
> > > > > > >> about
> > > > > > >> > > > > > its integration with a scheduler/resource
> manager?
> > What
> > > > about
> > > > > > >> MPI?
> > > > > > >> > > > > >
> > > > > > >> > > > > > Regards,
> > > > > > >> > > > > > Daniel
> > > > > > >> > > > > >
> > > > > > >> > > > >
> > > > > > >> > > > >
> > > > > > >> > > > > --
> > > > > > >> > > > > Sent from Gmail for mobile | mobile.google.com
> > > > > > >> > > > >
> > > > > > >> > >
> > > > > > >> > >
> > > > > > >> >
> > > > > > >>
> > > > > > >>
> > > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > > --
> > > > > > Greg Kurtzer
> > > > > > http://www.infiscale.com/
> > > > > > http://www.runlevelzero.net/
> > > > > > http://www.perceus.org/
> > > > > > http://www.caoslinux.org/
> > > > > >
> > > > >
> > > >
> > > >
> > >
> >
> >
>
