NAMD Cluster Tutorial Installation Instructions
The Rules
- This is not a race. You should have no problem finishing.
- The group member with the least Linux experience gets to drive.
- Please be patient if you have installed Linux before.
- The official documentation etc. is at http://www.scyld.com/.
- Scyld Beowulf meets our needs. It may not meet yours.
Master Node Installation
- Setup up the computer, monitory, keyboard, and mouse as usual.
- Connect the bottom network card (eth0) to the building network
and the top network card (eth1) to your private switch.
- Power on the computer and hit delete during the memory check to
enter the BIOS setup utility for the Asus A7A266 motherboard.
Confirm or change the following:
- Primary Master set to User Type HDD.
- Halt On set to All But Keyboard.
- CPU Level 2 Cache ECC Check set to Enabled.
- Boot order set to floppy, CD-ROM, hard drive.
- Boot Virus Detection set to Disabled
Insert the Scyld CD-ROM and Exit Saving Changes.
- The computer should boot off of the CD-ROM. When the initial
"Welcome to Scyld Beowulf Cluster Installation" screen appears, type
"text" to enter installation. If you wait too long the slave startup
will boot and then hang on "Sending RARP requests..."; in this case just
Ctrl-Alt-Del or hit the reset button and try again.
- What follows is mostly a normal RedHat Linux install. Here are some
hints for answering the questions:
- You have a "us" keyboard.
- You want to install a "Beowulf Controller based on GNOME".
- You want to initialize hda, default partitioning is fine.
- Hostnames and addresses for the tutorial are:
- access1.ks.uiuc.edu (130.126.120.98)
- access2.ks.uiuc.edu (130.126.120.99)
- access3.ks.uiuc.edu (130.126.120.100)
- access4.ks.uiuc.edu (130.126.120.101)
- access5.ks.uiuc.edu (130.126.120.102)
- Don't use bootp/dhcp; use netmask 255.255.255.0, gateway 130.126.120.1, nameserver 130.126.120.107.
- The defaults on the "Beowulf Configuration" page are fine, but lower the upper limit of the IP Range from 131 (32 slaves) to 102 (3 slaves) unless you
like looking at tables listing 29 down nodes.
- You have a generic PS/2 mouse (count the buttons yourself).
- The hardware clock should be set to GMT.
- I don't care what you pick for a root password, but you should write it down and tell the other members of your group.
- Create normal user accounts for everybody too.
- The video card should be recognized automatically.
- Formatting the filesystem takes about two minutes.
- Installing packages takes about five minutes (via a 52x CD-ROM drive).
- If this was your machine with your data on it you would want to create a custom boot disk. Skip it this time, though.
- If the monitor in front of you is on the list, go for it. Otherwise, Gateway EV700 worked for me.
- Pick your favorite video mode, ideally something 1280x1024.
- Yes, you want to start X on booting (for this tutorial at least).
- You're done! Hit OK and remove the CD-ROM before the system reboots.
Slave Node Installation
- Log in as root.
- The beostatus and beosetup programs should launch automatically.
- Connect the network on each slave node to your private switch.
- Move the monitor cord to the first slave, and power on and boot from the CD-ROM. (We could use a floppy for this, but the CD-ROM is much faster.)
- You would normally need to adjust the BIOS settings on your slaves too, but in this case they are known to be correct because they were taken from a working cluster.
- When the welcome screen times out, the machine will hang on "Sending RARP requests...".
- Switch the monitor to your master. BeoSetup should have an unknown hardware ethernet address listed. Drag it to "configured nodes" and click "apply".
- The state of the node will switch from "down" to "unavailable" and then finally to "up".
- Repeat for your other slave nodes, although there is no need to connect the monitor or to wait for one node to get all the way to "up" before ejecting the CD-ROM and starting the next node.
- Open a terminal and cd to /etc/beowulf.
- Run "beofdisk -q" to construct default partition tables, which you can inspect in the new directory /etc/beowulf/fdisk.
- Run "beofdisk -w" to actually format the hard disks on the slaves. Ignore the message about rebooting - you want to install the boot image first.
- Run "beofdisk -h" to see what the other options of this program are.
- Run "beoboot-install -a /dev/hda" to install the two-kernel monte boot image onto the hard drives on the slaves.
- Run "beoboot-install -h" to see what the other options of this program are.
- Run "bpsh -A df" to see what the nodes have mounted.
- Run "bpsh -h" to see what the other options of this program are.
- Edit /etc/beowulf/fstab, commenting out the $RAMDISK line and uncommenting the /dev/hda2 and /dev/hda3 lines. This tells the nodes to use their hard drives. You would need to temporarily undo these changes in order to format hard drives on additional nodes.
- Run "bpctl -S all -s reboot" to reboot the nodes.
Stupid Scyld Tricks
- Look in /var/log/beowulf if you want to see the slave boot logs.
- Look at /etc/beowulf/config if you want to see the file that BeoSetup modifies.
- Open a terminal. Run "bpstat", "bpstat -h" and "bpstat -u".
- In a different terminal run "namd-demo". Now run "bpstat -p" and "top".
(Kill namd-demo before you leave or gnuplot will interfere with your timings later.)
- Run "beostat -h" and try some of the options.
- Experiment with the bpsh command to explore the Linux environment on the slave nodes. Which commands don't work? Why?
- Run "bpctl -h". Be sure to try "bpctl -S all -s pwroff". (Then turn the slave nodes back on.)