You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@arrow.apache.org by Louis C <lc...@outlook.fr> on 2020/09/30 08:35:56 UTC
[C++] Flight performance
Hello,
I compiled the flight benchmarks on windows (using vcpkg and adding some options to the port file because it is not done by default), version 1.0.0.
While running it I only get about 700 MB/s max with the default parameters. My CPU is i7-4790K, with 32Gb of Ram (ddr3) at 1600 Mhz.
However I see that in the blog post concerning Flight, the performance which is reached is about 3GB/s.
Does such a difference in performance seems normal to you ?
Example of output :
>arrow-flight-benchmark.exe
Using standalone server: false
Server running with pid 27500
Testing method: DoGet
Server host: localhost
Server port: 31337
Server host: localhost
Server port: 31337
Batch size: 131040
Batches read: 9768
Bytes read: 1280000000
Nanos: 2737466000
Speed: 445.924 MB/s
Throughput: 3568.26 batches/s
Latency: 948.912 usec/batch
I also tested on similar machines, with similar performances (I reach 1.2GB/s max on more recent computers (windows) )
Cheers,
Louis C
Re: [C++] Flight performance
Posted by Wes McKinney <we...@gmail.com>.
Since the TCP/networking stack is different between Linux and Windows,
I would certainly expect the performance to be *different* but I'm not
sure of the expected magnitude of the difference. I'm copying dev@ in
case someone knows more about gRPC and TCP performance on Windows.
There might be some default gRPC settings which should be set
differently on Windows to achieve better performance out of the box
On Wed, Sep 30, 2020 at 3:36 AM Louis C <lc...@outlook.fr> wrote:
>
> Hello,
>
> I compiled the flight benchmarks on windows (using vcpkg and adding some options to the port file because it is not done by default), version 1.0.0.
> While running it I only get about 700 MB/s max with the default parameters. My CPU is i7-4790K, with 32Gb of Ram (ddr3) at 1600 Mhz.
> However I see that in the blog post concerning Flight, the performance which is reached is about 3GB/s.
> Does such a difference in performance seems normal to you ?
> Example of output :
> >arrow-flight-benchmark.exe
> Using standalone server: false
> Server running with pid 27500
> Testing method: DoGet
> Server host: localhost
> Server port: 31337
> Server host: localhost
> Server port: 31337
> Batch size: 131040
> Batches read: 9768
> Bytes read: 1280000000
> Nanos: 2737466000
> Speed: 445.924 MB/s
> Throughput: 3568.26 batches/s
> Latency: 948.912 usec/batch
> I also tested on similar machines, with similar performances (I reach 1.2GB/s max on more recent computers (windows) )
>
> Cheers,
> Louis C
Re: [C++] Flight performance
Posted by Wes McKinney <we...@gmail.com>.
Since the TCP/networking stack is different between Linux and Windows,
I would certainly expect the performance to be *different* but I'm not
sure of the expected magnitude of the difference. I'm copying dev@ in
case someone knows more about gRPC and TCP performance on Windows.
There might be some default gRPC settings which should be set
differently on Windows to achieve better performance out of the box
On Wed, Sep 30, 2020 at 3:36 AM Louis C <lc...@outlook.fr> wrote:
>
> Hello,
>
> I compiled the flight benchmarks on windows (using vcpkg and adding some options to the port file because it is not done by default), version 1.0.0.
> While running it I only get about 700 MB/s max with the default parameters. My CPU is i7-4790K, with 32Gb of Ram (ddr3) at 1600 Mhz.
> However I see that in the blog post concerning Flight, the performance which is reached is about 3GB/s.
> Does such a difference in performance seems normal to you ?
> Example of output :
> >arrow-flight-benchmark.exe
> Using standalone server: false
> Server running with pid 27500
> Testing method: DoGet
> Server host: localhost
> Server port: 31337
> Server host: localhost
> Server port: 31337
> Batch size: 131040
> Batches read: 9768
> Bytes read: 1280000000
> Nanos: 2737466000
> Speed: 445.924 MB/s
> Throughput: 3568.26 batches/s
> Latency: 948.912 usec/batch
> I also tested on similar machines, with similar performances (I reach 1.2GB/s max on more recent computers (windows) )
>
> Cheers,
> Louis C