You might want to take a look at the work done a few years ago in GSA-RG and recently in GRAAP-WG on this topic. There is a document on "Scheduling Attributes" in the document store of GSA-RG, which specifies various attributes that you mentioned, and some. This might be a good starting point for taking up older, more or less abandoned work and follow up on the basis of a real use case. Recently, GRAAP-WG has been working on an Advance Reservation profile for WS-Agreement, which basically is WS-* agnostic and just plain XML. Let me know if you have problems to find the documents. Also, I would welcome if you would cc: me, since scheduling-related issues are my key research area. Best, Alexander Am 29.03.2010 um 11:55 schrieb Mehdi Sheikhalishahi:
Dear All:
After "energy efficient scheduling in OCCI" presentation in OGF28, I am sending this message to introduce some Job/Scheduling parameters that can be incorporated into OCCI. According to the complexity of interface and requirements, all or some of these can be incorporated into the OCCI reference model. In Energy Efficient Scheduling, energy consumption for running all jobs is another metric to be optimized; an energy-aware scheduler reduces or minimizes energy consumption to run all jobs in a data center by considering other scheduling metrics. Contemporary modern processors offer a few important features to control energy consumption, e.g. cpuidle and Dynamic Voltage and Frequency Scaling (DVFS, cpufreq).
Here comes Job/Scheduling parameters and QoS metrics from user's perspective.
hpc-workload characteristic: {io-intensive, cpu-intensive, memory-intensive}, useful for scheduling algorithms like Gang programming model: {serial, MPI, openmp, others} scheduling goal: {energy, performance}, which one should be optimized preemption: {yes,checkpoint,no} allocation type: {best-effort, advance-reservation, immediate, deadline}, type resource allocation for job arrival time: wait time: start time: real execution time: estimated execution time: required by Backfilling algorithms finish time: response time CPU time slowdown algorithm: main scheduling algorithm {aggressive-backfilling, conservative-backfilling, probabilistic-backfilling, lookahead-backfilling, fcfs, sjf, gang} frequency-list: a list of frequencies that a job can run with electricity: how much electricity is used to finish job heat: how much heat is generated during job execution suspension/resumption: suspension of Virtual Machine by scheduler
After reviewing parameters related stuff, we will define new namespaces to cover these parameters.
Any comments appreciated.
Regards, Mehdi _______________________________________________ occi-wg mailing list occi-wg@ogf.org http://www.ogf.org/mailman/listinfo/occi-wg
-- Alexander Papaspyrou alexander.papaspyrou@tu-dortmund.de