From George Fuentes of UCAR: We are deploying an new batch system called Load Sharing Facility (LSF). We already have this product running on our Linux clusters from IBM. Any new systems that we buy will run this new batch system. On the larger Power IBM systems, we are still using LL along with our home grown job scheduler (BPS). This product is from Platform Computing, you can go to platform.com. Within this batch software package, we can deploy the NCAR scheduler functionality (BPS).
What I'm hearing is that maui on loadleveler can preempt jobs, but the preempted jobs are never resumed. The poster indicated that he was told that his fix for this problem would not be used in maui, but it moab would be fixed. Moab and Maui are under two different licenses. I believe Moab to be closed source.
I got a quote for LSF pricing from Platform, and forwarded it to Charlie. Charlie indicated that $30K was more than ESMF could afford.
SCD of NCAR is working on: BPS: a scheduler that works with loadleveler and LSF) BDS: a reduced-functionality scheduler UQS: "Unified Queuing System", which apparently is supposed to provide a single set of commands across a range of different queuing systems. I googled semithoroughly for BPS information, but did not find any downloads of it available.
Moab would be $3,964 for an 88 CPU IBM SP2 on loadleveler, and the salesrep indicates that Maob should be able to provide preemption, restarting, fairness guarantees, and resource reservation. The salesrep further indicates that we might be able to get a discount if we buy Moab through IBM. I sent e-mail asking if the job restarting would require additional coding. 2004-11-01
2004-11-01 Charlie tells me that the Moab purchase will have to go before the steering committee.
2004-11-01 Charlie also indicated that Maob on loadleveler is beta. Need to verify that. Charlie's right - portions of Moab are young with combined with loadleveler. I've sent an inquiry to Michael of clusterresources asking what parts are well tested, and what parts are young code. 2004-11-01.
2004-11-02 Charlie indicates in e-mail that $4000 is within the realm of possibility for the ESMF (meaning that a Moab purchase is a possibility).
From a clusterresources rep: Moab Cluster Suite would be considered a commercial subscription license where we will provide paying customers with confidential source code to modify if required. Our offering is priced to be based on a typical support pricing model. So that we can basically say that for the price of support, you get subscription licenses (with source code) as well as upgrade protection for 12 months. Follow on support/subscription years are 70% of the initial year. Example: Year 1 if someone had a $5,000 Support/Subscription license they receive the source code and upgrades. Year 2 would be $3,500, etc.