To: mbarker@MIT.EDU
Cc: wdc@MIT.EDU, rbasch@MIT.EDU
Date: Fri, 30 May 1997 18:11:09 EDT

This is an initial list of possible requirements and issues to be
considered for a "long jobs" (batch) system.

First, the following are general considerations for a batch system:

- interactive vs. batch only
- shared file system vs. file staging
- parallel execution
- recreation of submit environment
- job dependency
- configurable dispatching factors:
  - system load
  - resources available and required:
    - CPU type
    - CPU load
    - memory
    - disk
- load balancing
- checkpoint
  - system
  - user
- administration:
  - job monitoring
  - rescheduling
  - suspension/resumption
  - cancel
  - change priority (nice)
  - migration
- runtime limits
- support/track subprocesses
- provide job statistics
- fault tolerance
- redundancy (i.e. no single point of failure)
- security
  - authentication
  - authorization/access control
  - encryption
- GUI/Web/command line client tools for submit, job status, modify/cancel
- ease of use
- cost
- licensing
- platform coverage


The following are possible additional questions to be addressed for the
Athena environment:

- AFS shared file system support
- Kerberos authentication
  - allow jobs > 1 day (i.e. more than Kerberos ticket limit)
  - submit for future execution
- Job encryption
- single- or multi-user execution systems
- limits on what applications can be run, monitor usage
- special authorization needs
  - sign-up
  - faculty control
- need root access within job?
- require password on submit?


Bob