Method of mixed workload high performance scheduling
View Patent ↗A method of scheduling in a mixed workload environment. A high priority workload requiring bounded response times is executed on the same system with a low priority workload that is capable of tying up the CPU and multiple volume storage resources of the system by causing multiple concurrent I/O operations, thereby increasing the response times of the high priority workload beyond acceptable bounds. The method of scheduling prevents the response times of the high priority workload from increasing beyond the acceptable bounds by deferring the dispatch of processes servicing the current low priority workload for a time that depends on the priority of the low priority work and by not performing concurrent I/O operations that are requested by the current low priority work, when there is sufficient higher priority activity on the storage resource.
1. A method of scheduling in a mixed workload environment on a computing system having a CPU resource and a permanent storage resource, the computing system servicing requests from one or more clients, comprising:
executing a current process on the CPU resource and the storage resource, the current process having been dispatched to service a current client request;
performing a contention check while executing the current process to determine whether a new client request has a transaction priority that is greater than the transaction priority of the current client request;
if the transaction priority of the new client request is greater than that of the current request, dispatching a process to service the new client request;
If the transaction priority of the new request is not greater than that of the current request, determining whether the transaction priority of the current request is less than a predetermined threshold priority;
if the transaction priority of the current client request is lower than the predetermined threshold priority and there is higher priority I/O activity present on the storage resource:
delaying the servicing of the current client request and forgoing the servicing of any read aheads for the current client request; and
dispatching a process to service the highest priority client request that is available for service; and
if the transaction priority of the current client request is greater than the predetermined threshold or the priority of the current client request is lower than the predetermined threshold and there is no higher priority I/O activity present on the storage resource:
determining whether the current client request requires any read aheads;
dispatching one or more helper processes to service any required read aheads; and
returning to the current process to service the current client request.
2. A method of schedulling in a mixed workload environment as recited in claim 1 ,
wherein a maximum priority in the system is 255 and a minimum priority is 1; and
wherein the threshold priority is 151.
3. A method of scheduling in a mixed workload environment as recited in claim 1 , wherein the step of delaying the servicing of the current client request includes
delaying the servicing of the current client request by an amount of time that depends on the transaction priority of the current client request, higher priority requests being delayed less than lower priority requests, and the amount of the delay being the sum of a fixed delay and a priority dependent delay.
4. A method of scheduling in a mixed workload environment as recited in claim 3 ,
wherein a maximum transaction priority in the system 255 and a minimum priority is 1 and the threshold priority is 151;
wherein the fixed delay is about 0.2 seconds; and
wherein the priority dependent delay is the product of a constant and the difference between the threshold priority and the priority of the current client request.
5. A method of scheduling in a mixed workload environment as recited in claim 4 , wherein the constant is approximately 0.02.
6. A method of scheduling in a mixed workload environment as recited in claim 1 , wherein the step of delaying the servicing of the current client request includes delaying the servicing by a fixed delay.
7. A method of scheduling in a mixed workload environment as recited in claim 6 , wherein the fixed delay is approximately 10 milliseconds.
8. A method of scheduling in a mixed workload environment as recited in claim 1 , wherein the step of performing a contention check occurs once every time a physical block is transferred from the storage resource.
9. A system, comprising:
a central processing unit (“CPU”);
an interface coupled to the CPU and adapted to receive client requests, said CPU further adapted to:
execute a current process on the CPU, the current process dispatched to service a current client request;
determine whether a new client request has a transaction priority that is greater than the transaction priority of the current client request;
if the transaction priority of the new client request is greater than that of the current request, to dispatch a process to service the new client request;
if the transaction priority of the new request is not greater than that of the current request, to determine whether the transaction priority of the current request is less than a threshold;
if the transaction priority of the current client request is lower than the threshold and higher priority input/output (“I/O”) activity is present on a storage resource, to delay the servicing of the current client request and to dispatch a process to service the highest priority client request that is available for service; and
if the transaction priority of the current client request is grater than the threshold or the priority of the current client request is lower than the threshold and no higher priority I/O activity Is present, to determine whether the current client request requires any read aheads and to dispatch one or more helper processes to service any required read aheads.
10. The system of claim 9 wherein the CPU is further configured to delay the servicing of the current request by a variable amount.
11. A system, comprising:
a central processing unit (CPU) adapted to receive client requests and to:
execute a current process dispatched to service a current client request;
determine whether a new client request has a transaction priority that is greater than the transaction priority of the current client request;
if the transaction priority of the new request is not greater than that of the current request, to determine whether the transaction priority of the current request is less than a threshold; and
if the transaction priority of the current client request is lower than the threshold and higher priority input/output (“I/O”) activity is present on a storage resource, to delay the servicing of the current client request and to dispatch a process to service the highest priority client request that is available for service.
12. The system of claim 11 wherein if the transaction priority of the new client request is greater than that of the current request, the CPU is adapted to dispatch a process to service the new client request.
13. The system of claim 11 wherein if the transaction priority of the current client request is greater than the threshold or the priority of the current client request is lower than the threshold and no higher priority I/O activity is present, the CPU is adapted to determine whether the current client request requires any read aheads and to dispatch one or more helper processes to service any required read aheads.