Platform LSF Administration Guide Version 6.2

Welcome
Administering Platform LSF
21
Platform LSF HPC
Interruptible
backfill
Designed to improve cluster utilization, the new interruptible backfill scheduling policy
allows reserved job slots to be used by low priority small jobs that will be terminated
when the higher priority large jobs are about to start.
An interruptible backfill job:
◆ Starts as a regular job and is killed when it exceeds the queue runtime limit
OR
◆ Is started for backfill whenever there is a backfill time slice longer than the specified
minimal time, and killed before the slot-reservation job is about to start
Applies to compute-intensive serial or single-node parallel jobs that can run a long time,
yet be able to checkpoint or resume from an arbitrary computation point.
Resource
granularity
Allows for greater flexibility on how numeric resources are reserved by jobs. Resources
may be reserved on a slot, job, or host basis.
The cluster-wide RESOURCE_RESERVE_PER_SLOT parameter in
lsb.params is
obsolete. Resource reservation is now configured on a per-resource basis as PER_JOB,
PER_SLOT, PER_HOST in the ReservationUsage section in
lsb.resources. This
configuration overrides RESOURCE_RESERVE_PER_SLOT if it also exists.
Time-based slot reservation and greedy slot reservation
Existing LSF slot reservation works in simple environments, where host-based MXJ
limits are the only constraint to job slot requests. In complex environments, where more
than one constraint exists:
◆ Estimated job start time becomes inaccurate
◆ The scheduler makes a reservation decision that can postpone estimated job start
time or decrease cluster utilization.
Current slot reservation (RESERVE_BY_STARTTIME) resolves several reservation
issues in multiple candidate host groups, but it cannot help on other cases:
◆ Special topology requests, like span[ptile=n]
◆ Only calculates and displays reservation if host has free slots. Reservations may
change or disappear if there are no free CPUs; for example, if a backfill job takes all
reserved CPUs.
◆ For HPC machines containing many internal nodes, host-level number of reserved
slots is not enough for administrator and end user to tell which CPUs the job is
reserving and waiting for.
With time-based reservation, a set of pending jobs will get future allocation and
estimated start time. The size of job set is determined by cluster capacity and
LSB_TIME_RESERVE_NUMJOBS in
lsf.conf.
If a job cannot be placed in a future allocation, scheduler uses greedy slot reservation to
reserve slots. Existing LSF slot reservation is a simple greedy algorithm:
◆ Only considers current available resource and minimal number of requested job
slots to reserve as many slots as it is allowed
◆ For multiple exclusive candidate host groups, scheduler goes through those groups
and makes reservation on the group that has the largest available slots