Ram Samudrala (me@ram.org) v0.92, April 8, 2002 How to set up high-performance Linux computing clusters. This document describes how I set up my Linux computing clusters for high-performance computing which I need for my research. Use the information below at your own risk. I disclaim all responsibility for anything you may do after reading this HOWTO. The latest version of this HOWTO will always be available at http://www.ram.org/computing/linux/linux_cluster.html. Unlike other documentation that talks about setting up clusters in a general way, this is a specific description of how our lab is setup and includes not only details the compute aspects, but also the desktop, laptop, and public server aspects. This is done mainly for local use, but I put it up on the web since I received several e-mail messages based on my newsgroup query requesting the same information. Even today, as I plan another 64-node cluster, I find there is a dearth of information about exactly how to assemble components to form a node that works reliably under Linux. The main use of this HOWTO as it stands is that it's a report on what kind of hardware works well with Linux and what kind of hardware doesn't. This section covers the hardware choices I've made. Unless noted in the section, assume that everything works really well. Hardware installation is also fairly straight-forward unless otherwise noted, with most of the details covered by the manuals. For each section, the hardware is listed in the order of purchase (most recent is listed first). 32 machines have the following setup each: 2 AMD Palamino MP XP 1800+ 1.53 GHz CPUs Tyan S2460 Dual Socket-A/MP motherboard Kingston 512mb PC2100 DDR-266MHz REG ECC RAM 1 20 GB Maxtor UDMA/100 7200rpm HD 1 120 GB Maxtor 5400rpm ATA100 HD Asus CD-A520 52x CDROM 1.44mb floppy drive ATI Expert 98 8mb AGP video card IN-WIN P4 300ATX Mid Tower case Intel PCI PRO-100 10/100Mbps network card 32 machines have the following setup each: 2 Pentium III 1 GHz Intel CPUs Supermicro 370 DLE Dual PIII-FCPGA motherboard 2 256 MB 168-pin PC133 Registered ECC Micron RAM 1 20 GB Maxtor ATA/66 5400 RPM HD 1 40 GB Maxtor UDMA/100 7200 RPM HD Asus CD-S500 50x CDROM 1.4 MB floppy drive ATI Expert 98 8 MB PCI video card IN-WIN P4 300ATX Mid Tower case 1 server for external use (dissemination of information) with the following setup: 2 Pentium III 1 GHz Intel CPUs Supermicro 370 DLE Dual PIII-FCPGA motherboard 2 256 MB 168-pin PC133 Registered ECC Micron RAM 1 20 GB Maxtor ATA/66 5400 RPM HD 2 40 GB Maxtor UDMA/100 7200 RPM HD Asus CD-S500 50x CDROM 1.4 MB floppy drive ATI Expert 98 8 MB PCI video card Full-tower case with 300W PS 1 desktop with the following setup: 2 Intel Xeon 1.7 GHz 256K 400FS Supermicro P4DCE Dual Xeon motherboard 4 256mb RAMBUS 184-Pin 800 MHz memory 2 120 GB Maxtor ATA/100 5400 RPM HD 1 60 GB Maxtor ATA/100 7200 RPM HD 52X Asus CD-A520 INT IDE CDROM 1.4 MB floppy drive Leadtex 64 MB GF2 MX400 AGP Creative SB LIVE Value PCI 5.1 Microsoft Natural Keyboard Microsoft Intellimouse Explorer Supermicro SC760 full-tower case with 400W PS 2 desktops with the following setup: 2 AMD K7 1.2g/266 MP Socket A CPU Tyan S2462NG Dual Socket A motherboard 4 256mb PC2100 REG ECC DDR-266Mhz 3 40 GB Maxtor UDMA/100 7200 RPM HD 50X Asus CD-A520 INT IDE CDROM 1.4 MB floppy drive Chaintech Geforce2 MX200 32mg AGP Creative SB LIVE Value PCI Microsoft Natural Keyboard Microsoft Intellimouse Explorer Full-tower case with 300W PS 2 desktops with the following setup: 2 Pentium III 1 GHz Intel CPUs Supermicro 370 DLE Dual PIII-FCPGA motherboard 4 256 MB 168-pin PC133 Registered ECC Micron RAM 3 40 GB Maxtor UDMA/100 7200 RPM HD Asus CD-S500 50x CDROM 1.4 MB floppy drive Jaton Nvidia TNT2 32mb PCI Creative SB LIVE Value PCI Microsoft Natural Keyboard Microsoft Intellimouse Explorer Full-tower case with 300W PS 2 desktops with the following setup: 2 Pentium III 1 GHz Intel CPUs Supermicro 370 DLE Dual PIII-FCPGA motherboard 4 256 MB 168-pin PC133 Registered ECC Micron RAM 3 40 GB Maxtor UDMA/100 7200 RPM HD Mitsumi 8x/4x/32x CDRW 1.4 MB floppy drive Jaton Nvidia TNT2 32mb PCI Creative SB LIVE Value PCI Microsoft Natural Keyboard Microsoft Intellimouse Explorer Full-tower case with 300W PS 4 desktops with the following setup: 2 Pentium III 1 GHz Intel CPUs Supermicro 370 DE6 Dual PIII-FCPGA motherboard 4 256 MB 168-pin PC133 Registered ECC Micron RAM 3 40 GB Maxtor UDMA/100 7200 RPM HD Ricoh 32x12x10 CDRW/DVD Combo EIDE 1.4 MB floppy drive Asus V7700 64mb GeForce2-GTS AGP video card Creative SB Live Platinum 5.1 sound card Microsoft Natural Keyboard Microsoft Intellimouse Explorer Full-tower case with 300W PS Backup: 2 Sony 20/40 GB DSS4 SE LVD DAT Monitors: 1 22" Viewsonic P220F 0.25-0.27m monitor 4 21" Sony CPD-G500 .24mm monitor 2 18" Viewsonic VP181 LCD monitor 1 17" Viewsonic VE170 LCD monitor We use KVM switches with a cheap monitor to connect up and "look" at all the machines: 15" .28dp XLN CTL Monitor 3 Belkin Omniview 16-Port Pro Switches 40 KVM cables While this is a nice solution, I think it's kind of needless. What we need is a small hand held monitor that can plug into the back of the PC (operated with a stylus, like the Palm). I don't plan to use more monitor switches/KVM cables. Networking is important: 1 Cisco Catalyst 3448 XL Enterprise Edition 48 port network switch. 1 Netgear FS524 24 port network switch Our vendor is Hard Drives Northwest (http://www.hdnw.com). For each compute node in our cluster (containing two processors), we paid about $1500-$2000, including taxes. Generally, our goal is to keep each node to below $2000.00 (which is what our desktop machines cost). We use Linux systems with a 2.4.9-7 kernel based on the KRUD 7.2 distribution, and 2.2.17-14 kernel based on the KRUD 7.0 distribution. These distributions work very well for us since updates are sent to us on CD and there's no reliance on an external network connection for updates. They also seem "cleaner" than the regular Red Hat distributions. We use our own software for parallelising applications but have experimented with PVM and MPI. In my view, the overhead for these pre-packaged programs is too high. I recommend writing application-specific code for the tasks you perform (that's one person's view). Linux is freely copiable. This section describes disk partitioning strategies. farm/cluster machines: hda1 - swap (2 * RAM) hda2 - / (remaining disk space) hdb1 - /maxa (total disk) desktops (without windows): hda1 - swap (2 * RAM) hda2 - / (4 GB) hda3 - /home (remaining disk space) hdb1 - /maxa (total disk) hdd1 - /maxb (total disk) desktops (with windows): hda1 - /win (total disk) hdb1 - swap (2 * RAM) hdb2 - / (4 GB) hdb3 - /home (remaining disk space) hdd1 - /maxa (total disk) laptops (single disk): hda1 - /win (half the total disk size) hda2 - swap (2 * RAM) hda3 - / (4 GB) hda4 - /home (remaining disk space) Install a minimal set of packages for the farm. Users are allowed to configure desktops as they wish. I believe in having a completely distributed system. This means each machine contains a copy of the operating system. Installing the OS on each machine manually is cumbersome. To optimise this process, what I do is first set up and install one machine exactly the way I want to. I then create a tar and gzipped file of the entire system and place it on a CD-ROM which I then clone on each machine in my cluster. The commands I use to create the tar file are as follows: tar -czvlps --same-owner --atime-preserve -f /maxa/slash.tgz / I use have a script called go that takes a hostname and IP address as its arguments and untars the slash.tgz file on the CD-ROM and replaces the hostname and IP address in the appropriate locations. A version of the go script and the input files for it can be accessed at: http://www.ram.org/computing/linux/linux/cluster/. This script will have to be edited based on your cluster design. To make this work, I also use Tom's Root Boot package http://www.toms.net/rb/ to boot the machine and clone the system. The go script can be placed on a CD-ROM or on the floppy containing Tom's Root Boot package (you need to delete a few programs from this package since the floppy disk is stretched to capacity). More conveniently, you could burn a bootable CD-ROM containing Tom's Root Boot package, including the go script, and the tgz file containing the system you wish to clone. You can also edit Tom's Root Boot's init scripts so that it directly executes the go script (you will still have to set IP addresses if you don't use DHCP). Thus you can develop a system where all you have to do is insert a CDROM, turn on the machine, have a cup of coffee (or a can of coke) and come back to see a full clone. You then repeat this process for as many machines as you have. This procedure has worked extremely well for me and if you have someone else actually doing the work (of inserting and removing CD-ROMs) then it's ideal. Rob Fantini has contributed modifications of the scripts above that he used for cloning a Mandrake 8.2 system accessible at http://www.ram.org/computing/linux/cluster/fantini_contribution.tgz. If you have DHCP set up, then you don't need to reset the IP address and that part of it can be removed from the go script. DHCP has the advantage that you don't muck around with IP addresses at all provided the DHCP server is configured appropriately. It has the disadvantage that it relies on a centralised server (and like I said, I tend to distribute things as much as possible). Also, linking hardware ethernet addresses to IP addresses can make it inconvenient if you wish to replace machines or change hostnames routinely. The hardware in general has worked really well for us. Specific issues are listed below: The AMD dual 1.2 GHz machines run really hot. Two of them in a room increase the temperature significantly. Thus while they might be okay as desktops, the cooling and power consumption when using them as part of a large cluster is a consideration. The AMD Palmino configuration described previously seems to work really well. This section is still being developed as the usage on my cluster evolves, but so far we tend to write our own sets of message passing routines to communicate between processes on different machines. Many applications, particularly in the computational genomics areas, are massively and trivially parallelisable, meaning that perfect distribution can be achieved by spreading tasks equally across the machines (for example, when analysing a whole genome using a single gene technique, each processor can work on one gene at a time independent of all the other processors). So far we have not found the need to use a professional queueing system, but obviously that is highly dependent on the type of applications you wish to run. For the single most important program we run (our ab initio protein folding simulation program), using the Pentium 3 1 GHz processor machine as a reference frame, the Athlon 1.2 GHz processor machine is about 16% faster on average, the Pentium 4 1.7 GHz machine is about 25-32% faster on average, and the Athlon 1.5 GHz processor is about 80% faster on average (yes, the Athlon 1.5 GHz is faster than the Xeon 1.7 GHz since the Xeon executes only six instructions per clock (IPC) whereas the Athlon executes nine IPC (you do the math!)). These machines are incredibly stable both in terms of hardware and software once they have been debugged (usually some in a new batch of machines have hardware problems). Reboots have generally occurred when a circuit breaker is tripped. The first machine I installed has been up since its birth! ~ ram@fp1 % uptime 4:49am up 374 days, 2:47, 1 user, load average: 2.08, 2.02, 2.01 The following people have been helpful in getting this HOWTO done: Michael Levitt (Michael Levitt) The following documents may prove useful to you---they are links to sources that make use of high-performance computing clusters: RAMBIN web page RAMP web page Ram Samudrala's research page (which describes the kind of research done with these clusters)