Simulation Servers of Systems Ecology - ETH Zurich

The shown cluster consisted originally of Macintosh Xserve G5 rack mounted machines. The cluster contains a central file server (se-server) and an arbitrary number of attached, clustered nodes acting as simulation servers. All nodes can be used in the same manner and are all identical except for the first node, i.e. superior. First node superior offers additional functionality, i.e. it stores the user's home directories on behalf of the entire cluster. Recently this cluster was reduced to the bare minimum and consists now only of the central file server and the first node. As the rack mounted Xserve G5 Macs have stopped working, they were replaced by standalone G5 Macs. The latter Macs are fully compatible and can still run the identical software. Note, for security reasons this cluster is no longer accessible as described below.

• Simulationists have their home directory on a single node of the cluster, i.e. on superior. As a simulationist you always work with this very same directory, regardless on which clustered node you are currently logged in;
• The cluster offers ssh (includes scp, sftp) and afp services. All these services provide restricted access only available to members of the Systems Ecology group at ETH Zurich;
• To use these services it is recommended to use following means from a Mac OS X machine:
  • afp: the Finder's menu command 'Go -> Connect to Server...' (Command-K)
  • ssh: the preinstalled application 'Terminal', e.g. via 'TelnetLauncher', and
  • sftp: 'Fetch 5.x'
• Any access to the cluster is done with the same login and password, since it is controlled by a Open Directory / LDAP master. Access is offered only to registered users of the group simulationists;
• Every node of the cluster is protected by a firewall;
• No backup is provided on any node of the cluster. As a simulationist it is your personal responsibility to save the results and remove them from the cluster once your calculations are completed;
• Any afp volume (afp share point, afp stands for Apple File Sharing protocol) can be accessed from any Macintosh computer (given it supports AFP services via TCP/IP, i.e. MacOS 8.1 and later). The names of all these volumes are shown on above graph with dark blue letters. Ex.: Access the centrally stored models and data storage by afp://se-server.ethz.ch/Models-Data. Your access to individual folders may be restricted, depending on the project on which you are currently working;
• To access your personal home directory on superior, use afp to connect from any Macintosh to the superior (afp://superior.ethz.ch). Alternatively you can also use sftp from any host, e.g. from your Mac or any of our Suns. Connect to any node of the cluster, e.g. connect to michigan by sftp sftp://michigan.ethz.ch and you can access your home directory on superior;
• On the superior you can access also the directory 'Simulationists' (afp, sftp). That directory contains all user's home directories. Note, however, your access is restricted to the publicly available parts of the other user's home directories. Any other data are fully protected. This allows you to share data with other users, while preserving privacy;
• Data stored centrally in the folder Data of the afp volume Models-Data are stored only once on the se-server. They are made available for reading-only to any clustered simulation server (nfs export shown in above graph as arrows). Thus there is no need to first copy redundantly data to the the cluster node where you will execute simulation experiments;
• Any execution of programs is done only on the basis of so-called commando tools, i.e. the ssh service is the only mean available to simulationists to issue commands;
• The clustered simulation servers provide mainly RASS-OSX as modeling and simulation software;
• RASS-OSX plus other software is provided to the cluster via a single installation on the se-server (nfs export shown as arrows in above graph). This warrants identical behavior throughout the entire cluster. Clustered simulation servers find that software in directory /usr/local, which is by default accessible to any simulationist (global /etc/profile). It is not necessary to have a local '.profile' in your home directory and it is actually recommended to have none as long as you are working only with the standard simulation software;
• Some software is also installed locally on every clustered simulation server, but this is kept to an absolute minimum to minimize maintenance costs;
• Each cluster offers an individual large data pool. It is available via afp and sftp. This is a "commons" data storage to which all users have access. Its name begins with 'SimServ...' and ends with the first portion of the node's name. E.g on superior it is called 'SimServSuper'. As a simulationist you find in your home directory links to the currently available data pools. If you have currently several ssh sessions open to several cluster nodes, you will find symbolic links to all those nodes to which you are currently logged in. Note, however, from any given ssh session, you will have working access only to that pool on the machine to which that session is logged in. This data pool storage is the most efficient manner to store simulation results. For efficiency reasons you should use your home directory only sparingly. It is recommended to store in your home directory only models and settings files, but no large files resulting from simulation experiments. The only exception to these usage rules is the node superior, where this does not matter;
• The se-server can not be used to execute simulations;
• Use the se-server for any permanent storage of data, models, and simulation results. Store the files via afp, sftp, or scp at afp://se-server.ethz.ch/Models-Data or sftp://se-server.ethz.ch/Models-Data;
• The se-server offers backup (netbackup);
• The se-server offers various other services, which are all completely independent from the server's role for the cluster. An example is the central FileMaker data base (LiteratureSE) to store and exchange literature references;
• Benchmarks for ForClim 2.6-4.1 tests:
  

  Test	Runs (1200 a)	Quadra 950 (68K and FPU, 33 Mhz)	PPC mac (G4, ca.1 GHz)	superior (G5, dual 2.3 GHz) no optimization	superior (G5, dual 2.3 GHz) fast option
  *-test	6	ca. 4'12"	ca. 2'	19"	10"
  **-test	600	ca. 7h	2h 45'	33'37"	3'07"

• Should you encounter inconsistencies in your home directory, e.g. disfunctional symbolic links to the data pool storage, try running simulation cluster utility 'fixHome'. In addition it may also help to either 'source .profile' or even log out and log back in to make sure all works as expected (restores all environment variables should you have tampered with them).
• You can work with StuffIt archives using the command line tools 'stuff' and 'unstuff'.

• RASS-OSX is generated on a dedicated node, i.e. the first cluster node superior;
• RASS-OSX can be easily installed for the cluster by running script doRassInstallation as a system administrator on se-server;
• The cluster can be easily extended by further nodes, cloning the first ordinary cluster node, i.e. michigan, using 'NetRestore Helper' and 'NetRestore' (cf. afp://se-server.ethz.ch/SE_Software/Unix_Software/OS_X/NetRestore-ncutil, in particular consult 'Cloning cluster nodes.txt'). Alternatively it should also be possible to NetBoot a new node by pressing key 'n' during very first startup (se-server is also a NetBoot server for the cluster) after having added the new node's hardware address, IP#, and Server serial number into file 'machine_specific_data.csv';
• Restart sequence: 1) se-server, 2) superior, 3) rest of nodes. Restarting any machine early in this sequence typically requires a restart of the machines in the remainder of the sequence. Within 'rest of nodes' you need to observe no restart sequence;

Specifications of all our Unix machines (includes the Suns)

af, ETHZ - 2/19/26