Can you suggest me "how to calculate expected time to compute a task which requires heterogeneous multiple resources" ?
For example: if we consider only CPU requirement of a task, then the execution time of the task is task length in MI divided by CPU speed in terms of MIPS.
Can you suggest me for the consideration of multiple resources like CPU, RAM, Bandwidth, etc.
I do not know how you are modeling resource consumption in the multiple resources setting you are dealing with, but a simple approach might be to consider resource consumption to be sequential (even if the same resource appears multiple times in the sequence) and then take the expected value of the sum of all resource consumption times to be the sum of the respective expected values.
Dear Ram Pakath Sir,
Actually, resource like cpu time, we can calculate time needed for a task to execute.
But, when we consider, resources like memory, then problem creates. Suppose to execute a task, we need 10 data, then for searching and accessing those data, we need some time. I also need this time for my work to add with CPU time.
This is the problem.
You are referring to I/O time, I.e. time to read from and write to memory. If you are actually running a program, you can measure I/O time by taking suitable time stamps and using these to calculate time for different activities. Again, without details on what you are actually doing (eg, is thus a theoretical model, a simulation exercise, etc), one can only speculate.
Hi,
As you mentioned, the total CPU time is hardly estimated. Although the peak performance is easily obtained by the clock speed and the SIMD processors.
As for the memory, there are several level of memory devices. There are cache (L1,L2 ..) and main memory. The communication speed is different and neither the latency. If the necessary data is distributed on the cache, the delay to fetch the date should be small. Data missing cache causes the degradation. This is depend on your code, or the compiler you use.
So many HPC people have their benchmark software to estimate the actual computing time, but not so accurate.
Hi,
I find difficulties to understand what you want to do. In case of one task, you can refer to the overall soft chronometer of your task.
You can try queueing networks. You model each resource as a queue and define their interconnections.
Hi Mohammed Benalla,
We have a set of tasks and those task have different resource (like CPU time, RAM, Bandwidth, etc.) requirement. Then, how to calculate expected execution time of task ?
Hi Sambit,
As you have different materiel resources and different tasks to compute. I suggest to make a soft chronometer of each task and to sum the all task's time. You repeat the overall process 5 to 8 times and you will deduce the average computing time.
Hi Sambit,
I guess you can use such model as Universal Scalability Law, also often referred as USL. Originally it is introduced by Neil J. Gunther in some old paper, by I'd advise to find in the internet some updated information as shown on the attached page.
I hope that is what will help.
http://www.perfdynamics.com/Manifesto/USLscalability.html
hello sambit,
You mentioned that if we only consider, CPU it is easy to measure time, but i want to know about the CPU speed. How you find the CPU speed. and how it changes if another task arrives in parallel to the existing one?
Hi Sambit,
To compute the expected time, I divide the load by the available computing power. This is valid for any architecture type. To know more about the model I created, I recommend you read my publication on the following link and hope you will find the answer to your question:
http://onlinelibrary.wiley.com/doi/10.1002/spe.2258/abstract
Best regards,
Leila
- Badges
- Science topic
- Similar topics
- Computer Science