Is there any statistics about the RPC data?

[Johnson Li]

Hi all,

I am doing some research on the statistics of RPC. For example, what's the distribution of request frequency, response payloads, and the deadline in the normal case? 

I wonder if there is any real trace data available on the Internet or just some statistics.

Thanks

Johnson

[Michael Rose]

This seems like a very broad ask. All of those quantities vary per vary per (technology, endpoint, service, team, company). Maybe you can refine what you're looking for. Is this comparing and contrasting RPC against something else? Or trying to grasp what a "typical" RPC looks like?

Best,

Michael

I will say, RPCs and how they're used 

[Carl Mastrangelo]

There is research that is very similar for Internet packets.  Inter-arrival time is a power distribution, but can be approximated as exponential.  We use this in our benchmarks.  Payloads are usually small.  We also use this in our benchmarks.  

You are probably better off looking for published papers for this info.

[Johnson Li]

Hi Michael,

I just want to grasp what a "typical" RPC looks like. I have made some change in gRPC and HTTP/2， and got some performance improvement in the dataset that I made up. But I can not prove that the dataset is reasonable. So I am looking for one kind of reasonable data for my evaluation, or just some statistics about what kind of data is reasonable.

Thanks
Johnson

[Johnson Li]

Hi Carl,

I knew some data like http://pages.cs.wisc.edu/~tbenson/IMC10_Data.html. And papers like dctcp also collected network traces in their data center. The problem is that these packages and their conclusions are in the transport layer, thus TCP. I want to find some that are in the application layer, HTTP, to be exactly.

Besides, I want to find the data in RPC because I care about the deadline information. That seems to be rare for general HTTP traces.

Thanks
Johnson