Details on how COCOA was built
Setting up the hardware
Setting up the hardware was fairly straight-forward. Here are the main steps:
- Unpacked the machines, mounted them on the rack and numbered them.
- Set up the 24-port network switch and connected one of the 100 Mbit ports to the second ethernet adapter of the server which was meant for the private network. The rest of the 23 ports were connected to the ethernet adapters of the clients. Then an expansion card with 2 additional ports was added on the switch to connect the remaining 2 clients.
- Stacked the two 16-way monitor/keyboard switches and connected the video-out and the keyboard cables of each of the 25 machines and the server to it. A single monitor and keyboard were then hooked to the switch which controlled the entire cluster.
- Connected the power cords to the four UPS.
Setting up the software
Well, this is where the real effort came in! Here are the main steps:
- The server was the first to be set up. RedHat Linux 5.1 was installed on it using the bundled CD-ROM. Most of the hardware was automatically detected (including the network card), so the main focus was on partitioning the drive and choosing the relevant packages to be installed. A 3 GB growable root partition was created for the system files and the packages to be installed. Two 128 MB swap partitions were also created and the rest of the space (50 GB) was used for various user partitions. It was later realized that a separate /tmp partition of about 1 GB was a good idea.
- The latest stable Linux kernel (then #2.0.36) was downloaded and compiled with SMP support using the Pentium GNU CC compiler pgcc http://www.goof.com/pcg/ (which generates highly optimized code specifically for the Pentium II chipset) with only the relevant options required for the available hardware. The following optimization options were used: pgcc -mpentiumpro -O6 -fno-inline- functions. Turning on SMP support was just a matter of clicking on a button in the Processor type and features menu of the kernel configurator (started by running make xconfig).
- The new kernel-space NFS server for linux (knfsd) http://www.csua.berkeley.edu/~gam3/knfsd/ was installed to replace the earlier user-space NFS server to obtain improved NFS performance. For quick and hassle-free installation, a RedHat RPM package was obtained from http://rufus.w3.org/linux/RPM/, a popular RPM repository. The default options were used.
- ssh was downloaded from http://www.cs.hut.fi/ssh/, compiled and installed for secure access from the outside world. ssh-1.2.26 was preferred over the newer ssh-2.0.11 as ssh v2.x was much slower as well as backward incompatible. sshd daemon was started in runlevel 3 under
/etc/rc.d/rc3.d. Recently, RedHat RPMs for ssh have started appearing in http://rufus.w3.org/linux/RPM/ and several other RPM repositories, which make it much easier to install.
- Both the 3c905B ethernet adapters were then configured; one that connected to the outside world (eth1) with the real IP address 22.214.171.124, and the other which connected to the private network (eth0) using a dummy IP address 10.0.0.1. Latest drivers for the 3COM 3c905B adapters written by Donald Becker (3c59x.c v0.99H http://cesdis.gsfc.nasa.gov/linux/drivers/vortex.html) were compiled into the kernel to ensure 100 Mbit/sec Full-duplex connectivity. This was checked using the vortex-diag utility http://cesdis.gsfc.nasa.gov/linux/diag/vortex-diag.c. For the configuration, the following files were modified:
/etc/sysconfig/network, /etc/sysconfig/network-scripts/ifcfg-eth0 and
/etc/sysconfig/network-scripts/ifcfg-eth1. Here is how they looked for me after modification:
- For easy and automated install, I decided to boot each of the PCs from the network using the BOOT protocol. The BOOTP server was enabled by uncommenting the following line in /etc/inetd.conf:
bootps dgram udp wait root /usr/sbin/tcpd bootpd
A linux boot floppy was prepared with the kernel support for 3c905B network adapter which was used to boot each of the client nodes to note down their unique 96-bit network hardware address (eg. 00C04F6BC052). Using these address, the
/etc/bootptab was edited to look like:
- The /etc/hosts file was edited to look like:
127.0.0.1 localhost localhost.localdomain
# Server [COCOA]
126.96.36.199 cocoa.ihpca.psu.edu cocoa.aero.psu.edu cocoa
# IP address -- NAME mappings for the individual nodes of the cluster
10.0.0.1 node0.hpc.ihpca.psu.edu node0 # Server itself!
10.0.0.2 node1.hpc.ihpca.psu.edu node1
10.0.0.3 node2.hpc.ihpca.psu.edu node2
10.0.0.26 node25.hpc.ihpca.psu.edu node25
/etc/host.conf was modified to contain the line:
This was to force the lookup of the IP address in the /etc/hosts file
before requesting information from the DNS server.
- The filesystems to be exported were added to /etc/exports file
which looked like:
/boot node*.hpc.ihpca.psu.edu (ro,link_absolute)
/mnt/cdrom node*.hpc.ihpca.psu.edu (ro,link_absolute)
/usr/local node*.hpc.ihpca.psu.edu (rw,no_all_squash,no_root_squash)
/home1 node*.hpc.ihpca.psu.edu (rw,no_all_squash,no_root_squash)
/home2 node*.hpc.ihpca.psu.edu (rw,no_all_squash,no_root_squash)
/home3 node*.hpc.ihpca.psu.edu (rw,no_all_squash,no_root_squash)
/home4 node*.hpc.ihpca.psu.edu (rw,no_all_squash,no_root_squash)
- For rapid, uniform and unattended installation on each of the client nodes, RedHat 5.1 KickStart installation was ideal. Here is how my kickstart file /boot/install.ks looked like:
network --bootproto bootp
nfs --server 10.0.0.1 --dir /mnt/cdrom
part / --size 1600
part /local --size 2048
part /tmp --size 400 --grow
part swap --size 127
timezone --utc US/Eastern
rootpw --iscrypted kQvti0Ysw4r1c
lilo --append "mem=512M" --location mbr
@ Networked Workstation
rpm -i ftp://10.0.0.1/pub/CLUSTER/RPMS/wget-1.5.0-2.i386.rpm
rpm -i ftp://10.0.0.1/pub/CLUSTER/RPMS/xntp3-5.93-2.i386.rpm
/usr/bin/wget ftp://10.0.0.1/pub/CLUSTER/kernel/vmlinuz -O/boot/vmlinuz
/usr/bin/wget ftp://10.0.0.1/pub/CLUSTER/conf/lilo.conf -O/etc/lilo.conf
/usr/bin/wget ftp://10.0.0.1/pub/CLUSTER/conf/hosts.equiv -O/etc/hosts.equiv
sed "s/required\(.*securetty\)/optional\1/g" /etc/pam.d/rlogin /tmp/rlogin
mv /tmp/rlogin /etc/pam.d/rlogin
For more info on RedHat KickStart installation, look at: http://www- cache.ja.net/dev/kickstart/KickStart-HOWTO.html. In one of the post installation commands above, the first line of the /etc/pam.d/rlogin file is modified to contain:
auth optional /lib/security/pam_securetty.so
This is to required enable rlogin/rsh access from the server to the client without password which is very useful for the software mainte- nance of the clients. Also, the /etc/hosts.equiv file mentioned above looks like this:
The RedHat Linux 5.1 CD-ROM was then mounted as /mnt/cdrom on the server which was NFS exported to the client nodes. A new kernel with SMP support was compiled for the client nodes in very much the same way as for the server and was used to replace the existing kernel in the RedHat book diskette. This kernel however had lesser options com- piled in as it was only meant to act as a client. Additionally, option for ``kernel level autoconfiguration using BOOTP'' was enabled in the Networking options menu of the kernel configurator. This was required in order for the node to automatically get its IP address from the server at boot time. Support for The configuration file of the boot diskette was modified so as to boot directly in the KickStart mode. All that was needed to configure each client now was to insert the boot diskette, power-up the workstation and wait until the automatic installation was completed. Simple, eh ?!
- As soon as all the clients were rebooted after installation, the cluster was up and running! Some useful utilities like brsh ( http://www.beowulf.org/software/RPMS/beobase-2.0-1.i386.rpm) were installed to enable rsh a single identical command to each of the client nodes. This was then used to make any fine changes to the installation. NIS could have been installed to manage the user logins on every client node, but instead a simple shell script was written to distribute a common /etc/passwd, /etc/shadow and /etc/group file from the server.
- Most of the services were disabled in /etc/inetd.conf for each of the client nodes as they were unnecessary. The stripped down /etc/inetd.conf for the client nodes finally looked like:
auth stream tcp nowait nobody /usr/sbin/in.identd in.identd -l -e -o
- automount package was installed on each of the nodes to automatically mount the various user partitions on demand. Although this gave slightly improved NFS performance, it was found to be buggy and unstable. Finally, it was decided that automount for Linux was not yet ready for prime-time and was removed in favor of conventional NFS mounts.
- The Portland Group Fortran 77/90 and HPF compilers (commercial) were then installed on the server.
- Source code for freeware implementation of MPI library, MPI-CH was downloaded from http://www.mcs.anl.gov/mpi/ and compiled using pgcc. Installing it on the server on the /usr/local partition was quite straight-forward with no major hassles. The mpif77 script was modified to suit our needs and a similar mpif90 was created. The
/usr/local/mpi/util/machines/machines.LINUX was then modified to add two entries for each client node (as they were dual-processor SMP nodes). Jobs could now be run on the cluster using interactive mpirun commands!
- A queueing system, DQS v3.0 was downloaded from http://www.scri.fsu.edu/~pasko/dqs.html, compiled and installed as /usr/local/DQS/ making it available to all the client nodes through NFS. Appropriate server and client changes were then made to get it functional (i.e. adding the relevant services in /etc/services, starting qmaster on the server and dqs_execd on the clients) , although a few minor irritants were encountered. These were mainly owing to the bad documentation for DQS. It took a long time for me to figure out exactly how to configure the DQS to recognize a slave node, but once it was done, setting up the same for rest of the nodes was trivial. Wrapper shell scripts were then written by me for qsub, qstat and qdel which not only beautified the original DQS output (which was ugh to begin with!), but also added a few enhancements. For example, qstat was modified to show the number of nodes requested by each pending job in the queue. Also, three additional shell scripts qinfo, qload and qmem were written to give some useful load data for the nodes and the cluster resource utilization.
- COCOA was now fully-functional, up and running and ready for benchmarking and serious parallel jobs! As with the kernel, use of pgcc compiler was recommended for all the C/C++ codes. In particular, using pgcc with options ``-mpentiumpro -O6 -funroll- all-loops'' for typical FPU intensive number crunching codes resulted in 30 % increase in execution speed over the conventional gcc compiler.
This document is maintained by Suyog Shah firstname.lastname@example.org.
Mail me if you have any questions and/or suggestions.