Data center batch job quality of service control
A machine-controlled method can include determining an extended interval quality of service (QoS) specification for a batch job and determining a remaining data center resource requirement for the batch job based on the extended interval QoS specification. The machine-controlled method can also include determining an immediate QoS specification for the batch job based on the remaining data center resource requirement.
1. A machine-controlled method, comprising:
a processor determining an extended interval quality of service (QoS) specification for a given batch job, wherein determining the extended interval QoS specification comprises determining an allowed probability of failure q that the batch job will not be completed by the end of a time interval of n time slices;
for at least a first time slice of the n time slices, the processor determining a remaining data center resource requirement T for the batch job based at least in part on the extended interval QoS specification;
the processor determining an immediate QoS specification for the batch job, comprising:
determining a failure tolerance p for at least the first time slice; and
determining a smallest number of failures m such that a probability of m or fewer failures is greater than or equal to a required success rate (1−q);
determining a data center resource reservation S for the first time slice based on T, n and m; and
adjusting the data center resource reservation S, the failure tolerance p, or both, based at least in part on a progress of the batch job at the first time slice and the allowed probability of failure q.
2. The machine-controlled method of claim 1 , further comprising, for at least a second time slice of the n time slices, the processor determining the immediate QoS specification based at least in part on the progress of the batch job, wherein the second time slice occurs after the first time slice and before a completion time of the batch job.
3. The machine-controlled method of claim 1 , wherein determining the remaining data center resource requirement T comprises determining a stochastic model of data center resources needed to complete the batch job.
4. The machine-controlled method of claim 1 , wherein determining the immediate QoS specification comprises increasing the data center resource reservation S responsive to a determination that a progress of the batch job is slower than a predetermined amount.
5. The machine-controlled method of claim 1 , wherein determining an immediate QoS specification comprises decreasing the data center resource reservation S responsive to a determination that a progress of the batch job is faster than a predetermined amount.
6. The machine-controlled method of claim 1 , further comprising the processor adjusting the failure tolerance p for at least a second time slice of the n time slices based on the progress of the batch job, wherein the second time slice occurs after the first time slice and before a completion time of the batch job.
7. The machine-controlled method of claim 6 , wherein adjusting the failure tolerance p comprises decreasing the failure tolerance p responsive to a determination that the progress of the batch job is slower than a predetermined amount.
8. The machine-controlled method of claim 6 , wherein adjusting the failure tolerance p comprises increasing the failure tolerance p responsive to a determination that the progress of the batch job is faster than a predetermined amount.
9. The machine-controlled method of claim 1 , wherein determining the remaining data center resource requirement comprises estimating an amount of data center resources needed to complete the batch job based at least in part on registration information for a data center customer.
10. The machine-controlled method of claim 1 , further comprising the processor adjusting the data center resource reservation S and the failure tolerance p for at least a second time slice of the n time slices based on a progress of the batch job, wherein the second time slice occurs after the first time slice and before a completion time of the batch job.
11. A machine-controlled method, comprising:
a processor determining an extended interval quality of service (QoS) specification for a data center batch job, wherein determining the extended interval QoS specification comprises determining an allowed probability of failure q that the batch job will not be completed by the end of a time interval of n time slices;
for an immediate time slice, the processor determining a remaining data center resource requirement T for the data center batch job based at least in part on the extended interval QoS specification; and
the processor determining an immediate QoS specification for the data center batch job, comprising:
determining a failure tolerance p for the immediate time slice; and
determining a smallest number of failures m such that a probability of m or fewer failures is greater than or equal to a required success rate (1−q);
determining a data center resource reservation S for the immediate time slice based on T, n and m; and
adjusting at least one of the data center resource reservation S and the failure tolerance p based at least in part on a progress of the batch job at the immediate time slice and the allowed probability of failure q.