Automatic installation
D0 farm
Installation procedure of a D0 farm node:
- set BIOS to load from network
- connect node to network
- boot
The server contains:
- a boot area in which /etc/fstab with mounts of D0 directories
- D0 directories
After a (re)boot programs in the D0 area which can execute on the server can
execute (after minor modifications) on the node with a rsh command from the
server.
Responsabilities of the system manager:
- the preparation of the boot area on the server
Responsabilities of the application manager:
- The installation of the software on the server
- The verification of the correct execution of the software on the server
- The necessary modifations for running the program on a node
The D0 software consists of:
- ups/upd
- fermi products (/usr/products)
- D0 products (/d0usr/products)
- D0 releases (/d0dist/dist)
- tar files
The procedure works fine on a dedicated farm.
NIKHEF farm
There are several ways to accomodate more experiments on a NIKHEF farm:
- install the applications of all experiments on the same server
- bootsector contains links to directories of all experiments
- each experiment has its own server
- use separate bootserver which boots the initial system and directs the
node to the proper application server
Gridification
|
The WP4 general picture:
- Configuration data is read from a database and compared with the status
information in monitor
- From this comprison actions for the installation are generated and executed
- Information about the applications is stored in a separate database
|
This picture leads to the following questions and remarks:
- Concerning the configuration database
- What is the format?
- Who is responsible for the content?
- Concerning the application database
- Does this contain a description of the s/w or the s/w itself?
- Is it possible to gridify the D0/NIKHEF solution?
Possible scenario
Let us imagine the following scenario:
- The user installs the application on a private system.
- The user is responsible for the correct execution of the application.
- The user provides information about the files which should be on the
disks of the nodes.
- The user creates a distribution tarfile.
- The distribution is shipped to a compatible system in the grid.
- The receiving centre installs the tarfile and verifies the
correct functioning of the application.
- The receiving centre generates a boot sector for the application.
- The tarfile and bootsector are installed on a farmserver.
- The centre supplies a farm batch system.
The following information has to be provided:
- the specifications of the hardware and the system software
- the directories on the server which has to be NFS mounted from the node
- the batch commands (which have to be translated in the equivalent local
commands)
To execute an application the user provides:
- job dependent configuration files with its location
- url's of input and output files
Remarks
The model should provide procedures for:
- the installation of the application
- the modification of the application
- the execution of the application
- in batch mode
- in debug mode
- with the compilation of user routines
- the replacement of a node in a farm
- before installation the machine parameters should be compared with the
specifications in the purchase database
- the replacement of a farm server
The computer center should publish information about:
- available platforms
- the versions of the installed applications
A Virtual Organization should specify:
- a limited number of certified platforms
- the available certified versions of applications
- the location of the applications
Willem van Leeuwen,
Tue Feb 13 17:30:56 MET 2001