Benchmarking Thrift

[Tamás Szelei <he...@gmail.com>]

Hello all,

I'm creating a benchmark of RPC solutions and Thrift is one of them.
Similarly to unit tests, each RPC solution has a fixture where the
infrastructure for performing calls is set up. The benchmarks then measure
the performance of various scenarios. In all cases, the fixtures create
both a client and server and synchronous calls are performed.

It is still a WIP, but I would like to get your input on the setup code I
implemented, i.e. whether or not this is a fair entry in such a benchmark
and if there are any possible areas of improvement. The code is here:
https://github.com/rpclib/benchmarks/blob/master/include/thrift/fixture.h

FWIW Thrift seems to be performing quite well in these early measurements.

Also, if there are any known scenarios where Thrift is expected to
outperform outhers, please let me know.


Regards,
Tamás Szelei

Re: Benchmarking Thrift

[Tamás Szelei <he...@gmail.com>]

Hi,

Thanks. I haven't seen this, but I also don't speak Chinese (Google
translate seems ok-ish though). In any case, I forgot to mention that my
benchmark measures the C++ implementations, whereas the post you linked
measures Java.

Regards,
Tamás

On Sun, Feb 5, 2017 at 11:54 AM, Jens Geyer <je...@hotmail.com> wrote:

> Hi,
>
> have you seen this?
>
> http://www.useopen.net/blog/2015/rpc-performance.html
>
> have fun,
> JensG
>
> -----Ursprüngliche Nachricht-----
> From: Tamás Szelei
> Sent: Sunday, February 5, 2017 10:52 AM
> To: user@thrift.apache.org
> Subject: Benchmarking Thrift
>
> Hello all,
>
> I'm creating a benchmark of RPC solutions and Thrift is one of them.
> Similarly to unit tests, each RPC solution has a fixture where the
> infrastructure for performing calls is set up. The benchmarks then measure
> the performance of various scenarios. In all cases, the fixtures create
> both a client and server and synchronous calls are performed.
>
> It is still a WIP, but I would like to get your input on the setup code I
> implemented, i.e. whether or not this is a fair entry in such a benchmark
> and if there are any possible areas of improvement. The code is here:
> https://github.com/rpclib/benchmarks/blob/master/include/thrift/fixture.h
>
> FWIW Thrift seems to be performing quite well in these early measurements.
>
> Also, if there are any known scenarios where Thrift is expected to
> outperform outhers, please let me know.
>
>
> Regards,
> Tamás Szelei
>
>

Re: Benchmarking Thrift

[Jens Geyer <je...@hotmail.com>]

Hi,

have you seen this?

http://www.useopen.net/blog/2015/rpc-performance.html

have fun,
JensG

-----Ursprüngliche Nachricht----- 
From: Tamás Szelei
Sent: Sunday, February 5, 2017 10:52 AM
To: user@thrift.apache.org
Subject: Benchmarking Thrift

Hello all,

I'm creating a benchmark of RPC solutions and Thrift is one of them.
Similarly to unit tests, each RPC solution has a fixture where the
infrastructure for performing calls is set up. The benchmarks then measure
the performance of various scenarios. In all cases, the fixtures create
both a client and server and synchronous calls are performed.

It is still a WIP, but I would like to get your input on the setup code I
implemented, i.e. whether or not this is a fair entry in such a benchmark
and if there are any possible areas of improvement. The code is here:
https://github.com/rpclib/benchmarks/blob/master/include/thrift/fixture.h

FWIW Thrift seems to be performing quite well in these early measurements.

Also, if there are any known scenarios where Thrift is expected to
outperform outhers, please let me know.


Regards,
Tamás Szelei