Maui Scheduler

Overview:

   Maui is an advanced job scheduler for use on clusters and supercomputers.  It is a highly optimized and configurable tool which allows sites to implement key mission policies, improve utilization and response time, track resource consumption, and improve overall cluster fault tolerance.

   Maui provides a wide array of fairness policies and dynamic prioritization settings to allow determination of which jobs should be allowed to run and in what order.  Reservations and node allocation policies allow control over when and how resources are made available to particular jobs.  Backfill, nodesets, and other optimizations allow sites to improve both response time and system utilization metrics.  Statistics are employed to allow tracking of delivered quality of service and address QoS targets. Diagnostic tools provide analysis regarding current cluster state, constraints, and failures.

   Maui is currently in use at many of the world's leading government, academic, and commercial sites.   It is highly scalable and improves the manageability and efficiency of machines ranging from clusters of a few processors to multi-teraflop supercomputers.

Features:

    Maui extends the capabilities of base resource management (or BQS) systems by adding the following features:
 

    Maui interfaces with numerous resource management systems supporting any of the following scheduling API's

PBS Scheduling API - OpenPBS and PBSPro (Veridian)
Loadleveler Scheduling API - Loadleveler (IBM)
SGE Scheduling API - Grid Engine (Sun)*
BProc Scheduling API - BProc (Scyld)**
SSS XML Scheduling API*
LSF Scheduling API - LSF (Platform)
Wiki FlatText Scheduling API (Wiki)
*partial support or under development
**supported under clubmask

    Maui is currently supported on all known variants of Linux, AIX, OSF/Tru-64, Solaris, HP-UX, IRIX, FreeBSD, and other UNIX platforms.
 
    The Maui scheduler is mature, fully documented, and supported.  It continues to be agressively developed and possesses a very active and growing user community.