Patagonia






 
  -CoPs Project

  -Research
     Group
  -Institute
  -Departement
  -ETH

Patagonia -
A Dual Use Cluster of PCs
for Computation and Education

The COPS project

Computer clusters in major research universities are used for research and education. While both kinds of installations can be called clusters, the installations are far from looking identical. The PCs of an education cluster are workstations that are sparsely distributed across tables in a classroom and do have screens, keyboards and mice, while clusters of PCs for high performance computing are densely packed rack mounted systems kept in a cooled machine room and wired to one central operator console. Despite the completely different look of the two kinds of clusters, the current trends in technology mandate that they are built with nearly the same components.

Typically the education clusters are only used during the day and maybe in the evenings whereas the compute clusters are often idle. Complementary use would therefore hardly improve the cost effectiveness of the installations As a consequence we initiated the Patagonia cluster project at ETH Zürich to build a cluster that fits both the needs of education and the needs of research. While education has priority during the day on our cluster, research has priority during the night and with some limitations during vacations.

While our ideas and the corresponding experimental study clearly originates from a university environment we would like to mention the striking similarities to many modern corporate computing environments using PCs. In those environments the computing needs are quite similar. Most companies rely on a rapidly growing number of compute intensive tasks in data mining, combinatorial optimization and process simulations in addition to the typical personal computing needs of a large number of employees at their desks.

Patagonia Talk: acrobat, compressed postscript, html.


Patagonia CloneSys - A Tool to Install Multi-Boot Environments

Everybody knows the drudgery involved in manually setting up and rolling out new PCs, updating existing PCs, and recovering failed PCs. CloneSys makes a big dent in the time required to conduct these jobs. It first creates an exact image of a PC's hard drive or disk partitions, effectively taking a snapshot of all the files - hidden, visible, and active - that make up the operating system, applications, and configuration settings. The image can then be copied to any number of PCs, thereby creating completely identical installations. Moreover, it can be copied to many PCs simultaneously. This process is completely operating system independent which makes the system very easy and fast. As the raw disk data is copied, no file system initialization is needed, even the partitioning is implicitly done with a total clone. To support multi-boot environments the installation of single partitions is supported as well.

To keep the system even easier as other commercially available tools like Norton Ghost from Symantec, ImageCast from Innovative Software Ltd. or DriveImagePro form PowerQuest, we based it on freely available UNIX tools and Linux. For an initial boot of an uninstalled machine we use muLinux which is a minimalistic, but mostly complete, script-based Linux distribution that fits on a single 1722k floppy disk. It includes many basic system functions, such as Ethernet support, NFS, Samba, FTP, DHCP etc which is all we need. Further we install a small Linux distribution permanently on the harddisk which lets us update the cluster very fast by simply booting this Linux on all machines and remotely executing our cloning scripts.
Locking mechanisms on partitions allow to install special setups on a part of the cluster preventing the image to be overwritten by others.


Relevant Materials:

  • F. Rauch, Ch. Kurmann, T. Stricker: Partition Repositories for Partition Cloning - OS Independent Software Maintenance in Large Clusters of PCs. Proceedings of the IEEE International Conference on Cluster Computing 2000, Chemnitz, Germany, Nov 28 - Dec 2, 2000.
    Available formats: abstract, acrobat, compressed postscript.


  • F. Rauch, Ch. Kurmann, T. Stricker: Partition Cast - Modelling and Optimizing the Distribution of Large Data Sets in PC Clusters. Euro-Par 2000 -- Parallel Processing, Arndt Bode and Thomas Ludwig (Editors), Springer, Lecture Notes in Computer Science 1900, ISBN 3-540-67956-1. Presented as distinguished paper at: European Conference on Parallel Computing, Euro-Par 2000, Munich, Germany, August 29 - September 1, 2000.
    A similar paper appeared as technical report No. 343, Department of Computer Science, ETH Zürich.
    Available formats: abstract, acrobat, compressed postscript.
    Slides: acrobat, postscript, compressed postscript.


  • F. Rauch, C. Kurmann, B. M. Müller-Lagunez, T. M. Stricker Patagonia - A Dual Use Cluster of PCs for Computation and Education Proc. of the second workshop on Cluster-Computing, 25./26. March 1999, Karlsruhe, Germany.
    Available formats: abstract, acrobat, postscript, compressed-postscript.
    Talk: acrobat, postscript, compressed-postscript.


Download

Download The Patagonia CloneSys is designed to provide easy system installations on multi-boot and multi-purpose Clusters of PC's. Do keep it simple and circumvent the development of a proprietary system was our aim. Therefore it is based on standard UNIX tools and can easily be ported to other flavours of UNIXes.

The scripts and ideas are subject to change and are provided as they are without any support.

You can download our Cloning System Scripts clonesys_v0.9.tar.gz (91 kB). It runs on standard UNIX systems, but it is especially designed for Linux-Systems to install multiboot PC Clusters. Read the Quick Reference (quickinit.pdf) and the CloneSys Documentation (clonesys.pdf) for further information. 


Other tools and helpful documentation:

Please note that the materials listed here are helpful for our own cluster maintenance. We provide them in the hope that they can be helpful for others too, but we can not give any support for them.
  • CAT: Cluster Administration Tool. It helps to gain an overview of the current state of a multi-use cluster and shows the currently running operating system, machine load and users with their corresponding idle times for each node. The client part of the tool as well as the daemon for Linux was mostly written in Tcl/Tk, while the daemon for Windows NT is in C.

    Files: The tool including source CAT.tar.gz (234 kB), the documentation rspuler.pdf (95 kB, in german! Includes some screenshots), the english abstract.

  • Dolly. This is a program to clone harddisks or partitions over a fast switched network. It does so by building a virtual TCP ring between the machine with the disk/partition to be cloned and all the machines where the disk/partition should be written. It works with raw partitions as well as (possibly compressed) image files. We were able to clone a 2 GB Windows NT partition using dolly to 15 machines over Gigabit Ethernet in less than 4 minutes.

    Files: The latest stable version of Dolly including source and README is 0.57: dolly.0.57.tar.gz (21 KB). A newer, but not yet so thoroughly tested version is 0.58C: dolly.0.58C.tar.gz (25 KB). The most notable feature of version 0.58 is that it now allows the use of the standard input and standard output to read and write files. I.e. you can now use tar to clone directory trees. This version might break some third-party scripts, because dolly now prints all of its regular output to standard error instead of standard output. Version 0.58C introduces a flag that tells Dolly not to sync() before it exits, thereby reducing the runtime for smaller files. The current documentation is in HTML or ASCII. There is also a directory with other versions of Dolly.

    A research paper on Dolly was presented at Euro-Par 2000 (29.8.-1.9.2000 in Munich, Germany). See above (relevant materials) for more information.

    Atsushi Manabe of KEK (Tsukuba, Japan) provides an alternative implementation called Dolly+ with some other features (e.g. multi-file transfers and a fail-safe mechanism to bypass crashed nodes).

    A new utility based on Dolly is nettee, a network 'tee' program maintained by David Mathog. It is a simpler, cleaned-up program based on Dolly's source code.

  • Zip bootdisk which runs Linux in a ramdisk. This documentation describes the required steps the generate a pair of a floppy- and a Zip-disk which boots Linux completely from a RAM-disk. This is useful to boot a whole cluster of new (or empty) machines to a small Linux system and install the full installation over the network.

    Available formats: HTML

Dolly together with the Zip bootdisk can be useful to install the harddisks of all machines in a new cluster within reasonable time.
  • System Commander We use V-Communications' System Commander to boot the multiple installed systems. After trying different boot managers we evaluated this tool as the one fitting our specifications best. The System Commander utility offers menu controlled selection of the operating system of choice and lets protects systems and administration setup with a sophisticated password protection scheme, which allows user groups to be defined. Furthermore, it protects from booting from external attached devices and floppy disks.

    See our whitepaper for V-Communications' System Commander , also available as Acrobat PDF.


ICS - Lab for Computersystems
DINFK - Departement of Computer Science
ETHZ - Swiss Institute of Technology

Departement Informatik der ETH Zürich
May 2005
Christian Kurmann, < kurmann@inf.ethz.ch >
Felix Rauch, < rauch@inf.ethz.ch >