How can we calculate the bandwidth used by a workload sent to an SSD device with its average submission, completion and response time being available? Is there any formula for that? How can we implement it with C or C++?
An example of a workload is an asynchronous read-only one with data chunks of 4KB and a total data size of 128KB. The average response time of each request is 100usec after the simulation of it on a SSD device.
There would be some points to be noted:
1. The model need to be built firstly, you should build your testing environment based on your workload;
2. For eliminating deviation, there must be an interface between hardware and software, and make sure the interface is unimpeded, that means no other dataflow will affect the internection. E.g., there is not third APP to use CPU resource and make that internection be congested. Then the latency between software and hardware is fixed.
3. To develop the program to deal with the interrupt from the IO device and is called by OS. The time between interrupt reported by hardare and be aware by software is fixed when we have done step 2. The model we build in step 1 will be used to calculate the bandwidth since all the latency in the whole progress is fixed and calculated.
Well, you can do what is recommended above, but maybe analysing your problem in more depth could help. It looks like a server query model, with arrival times, service times, queueing times. Does M/M/1 ring a bell?
See
https://en.wikipedia.org/wiki/M/M/1_queue
- Badges
- Science topic
- Similar topics
- Philosophy
- Philosophy Of Science
- Reasoning