Can the workload rate be customized?

[WangTingZheng]

For example, we can simulate real-world workloads by customizing the number of load arrivals for each time period like this:

[ls4154]

Since YCSB is usually used to measure peak throughput, it does not have a request rate customization.
I think we could add some simple constant rate limit algorithms, but simulating a workload like the graph above seems complicated. Is there any popular workload trace available that can be used as an input?

[WangTingZheng]

Since YCSB is usually used to measure peak throughput, it does not have a request rate customization. I think we could add some simple constant rate limit algorithms, but simulating a workload like the graph above seems complicated. Is there any popular workload trace available that can be used as an input?

Thank you for your reply，QueryBot5000[1] has collected the number of accesses within certain intervals in certain scenarios, which might be useful?

[1] Ma, Lin, et al. "Query-based workload forecasting for self-driving database management systems." Proceedings of the 2018 International Conference on Management of Data. 2018.

[ls4154]

Hi, sorry for the late reply.

I've added a way to limit the client request rate.
Check out the rate_limit branch.

A constant rate can be specified using the limit.ops option. (initial rate)

The rate can be also set from the trace file by using limit.file options. The format of the file is as follows.

20 288
30 292
40 368
50 335
60 242

An example command using the above two options would be:

./ycsb -run -db basic -P workloads/workloada  -p threadcount=4 -p basic.silent=true -p limit.ops=100 -p limit.file=rate_trace.txt -s

Rate limiting only works for the run phase.

Please let me know if you have any issues or improvement ideas :)

[WangTingZheng]

Hi, sorry for the late reply.

I've added a way to limit the client request rate. Check out the rate_limit branch.

A constant rate can be specified using the limit.ops option. (initial rate)

The rate can be also set from the trace file by using limit.file options. The format of the file is as follows.

20 288
30 292
40 368
50 335
60 242

An example command using the above two options would be:

./ycsb -run -db basic -P workloads/workloada  -p threadcount=4 -p basic.silent=true -p limit.ops=100 -p limit.file=rate_trace.txt -s

Rate limiting only works for the run phase.

Please let me know if you have any issues or improvement ideas :)

thanks a lot!

[WangTingZheng]

how to get read throughput and latency when I set the rate limite? throughput and latency is fixed. @ls4154

[ls4154]

how to get read throughput and latency when I set the rate limite? throughput and latency is fixed.

If you limit the maximum rate, the rate should be fixed, unless your db is slower than limits.
Latency should not be fixed though.