software running on the nodes?
Many of your questions may have already been answered in earlier discussions or in the FAQ. The search results page will indicate current discussions as well as past list serves, articles, and papers.
Robert G. Brown rgb at phy.duke.eduTue Jan 14 11:13:45 PST 2003
- Previous message: software running on the nodes?
- Next message: software running on the nodes?
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
On Tue, 14 Jan 2003, Mike Eggleston wrote: > Has this been asked for? Can we have a quick, informal poll > on what os/distribution/rpms/etc is running on people's nodes? RH 7.3, custom kickstart installation. Several hundred nodes (and lan workstations) in our department, but there are other departments who may or may not be using the same Dulug Beowulf group below as there are lots of clusters on campus now. Mostly dual nodes, both PIII and Athlon. See below (with encrypted root pw edited out:-) our "beowulf" ks file, and below it the relevant additions to the standard comps file to define the dulug beowulf group. Pretty straightforward -- sundry math/stats including blas/lapack, lam, pvm, gsl, xmlsysd/wulfstat to facilitate node monitoring, and a workstation-like install elsewhere. We (re)wrap everything local including proprietary site-license software in rpm form and restrict access accordingly. Perhaps this is a bit fatter than strictly necessary (installing X just so one can run x-tools on a node:-) but disk is "infinite" these days, our department network is basically flat wrt the cluster, and so who cares? We have a second image that doesn't create a rest-of-the-disk xtmp directory for use on small disk systems. Note that 3 GB is plenty for our OS image at least so far, and having a rest-of-disk, non-backed-up xtmp on each node provides many GB of scratch space for users without needing to know much about the node's specific disk other than it is bigger than 5 or 6 GB. There is also sundry magic worked in the post -- security, access, grub, and more. Note that a lot of this is due to Seth Vidal, our local linux genius who has set up and runs the dulug server(s). Anybody on campus (with a duke.edu address:-) can install linux consistently in a manner of minutes, automagically updated to the latest updates nightly via yum. This supports undergrads and grads, individual faculty at home (with duke.edu or proxy/tunnel) or in departments, department admins for their department lans, cluster owners and their clusters. Anybody on campus can install using the dulug beowulf group (and their own %post and additions/subtractions). Any admin who learns to use kickstart, grub, dhcpd and pxe can install or reinstall any cluster, cluster node, PC or workstation in a matter of minutes per node, where most of that time is spent waiting, not working. Just one person (Seth) does ALL the toplevel grooming of the actual distributions and support of the requisite tools. This scales pretty close to the theoretical maximum as a consequence, especially given that Seth's primary job ISN'T maintaining this, but running the physics department LAN. One fraction of an FTE (admittedly, a cosmically >>talented<< FTE:-) can and does maintain the entire tree structure that supports an entire University's use of linux -- thousands of systems -- including the completely automated distribution of e.g. security or functional updates as they are released. Seth actually provides a lot of detailed support for individual users and other linux admins via the dulug mailing list, as well, although there he has at least some help from the rest of the expert user base. All this and a real job too, gee! I'd like to see a single MCSE attempt this. Ever. Makes me want to laugh hysterically just thinking about it -- tracking the licenses alone for 1000+ systems would eat an FTE, and then, one has to pay for all those licenses. With Win(X < 2000) I think it would be openly impossible. With Win2000, there are at last some tools in MS for remote installs, but I suspect that server scaling and rising costs would utterly destroy you long before you could install a single 100+ node cluster, let alone a dozen LANS from a single common webserver. rgb Robert G. Brown http://www.phy.duke.edu/~rgb/ Duke University Dept. of Physics, Box 90305 Durham, N.C. 27708-0305 Phone: 1-919-660-2567 Fax: 919-660-2525 email:rgb at phy.duke.edu %< Snip snip snip ===================================================== install text lang en_US langsupport en_US keyboard us mouse none network --bootproto dhcp rootpw --iscrypted dontyouwishdontyouwithdontyouwish url --url http://install.dulug.duke.edu/pub/linux/dulug-7.3/i386 timezone US/Eastern skipx auth --useshadow --enablemd5 bootloader --location=mbr zerombr yes clearpart --all firewall --disabled part / --size 3000 --fstype ext3 part swap --recommended part /disk1 --size 2000 --fstype ext3 part /xtmp --size 1000 --grow --fstype ext3 reboot %packages @ Dulug Beowulf amanda amanda-client aspell htmlview ImageMagick joe jove mgetty openssh-server postfix pspell pump rdate-cron symlinks vim-enhanced wget xinetd -cups-drivers-hpijs -cups-drivers-pnm2ppa -cups-drivers -dateconfig -dhcpcd -ghostscript -ghostscript-fonts -gv -hpijs -logwatch -ntp -Omni -Omni-foomatic -printconf -printconf-gui -rhn_register -rhn_register-gnome -sendmail -sendmail-cf -sudo -up2date -up2date-gnome %post cd /tmp /usr/bin/wget -q -nd -nH http://install.phy.duke.edu/rh-7.3/post/beowulf.sh chmod +x /tmp/beowulf.sh /tmp/beowulf.sh 2>> /tmp/post-errors.log rm -f /tmp/beowulf.sh %< Snip ditto ------ 0 Dulug Stat/Math { ? X Window System { octave gnuplot root root-examples root-documentation postgresql-libs R-base R-recommended } gsl gsl-devel hpc blas blas-man lapack lapack-man } 0 Dulug Beowulf Misc { xmlsysd wulfstat ntp pvm pvm-gui lam } 0 --hide Dulug Beowulf { @ X Window System @ Network Support @ Utilities @ Emacs @ Software Development @ Dulug Stat/Math @ Dulug Beowulf Misc }
- Previous message: software running on the nodes?
- Next message: software running on the nodes?
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
