<div dir="ltr">Micael, are there similar tests for udp? </div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Tue, Oct 29, 2019 at 12:12 PM Micael Karlberg <<a href="mailto:micael.karlberg@ericsson.com">micael.karlberg@ericsson.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">Hi,<br>
<br>
Its not a secret :) You will find it in the emulator test catalog.<br>
The only problem is that it was never intended for public "consumption"<br>
and is therefor not documented:<br>
<br>
socket_test_ttest_tcp_socket.erl (uses socket.erl)<br>
socket_test_ttest_tcp_gen.erl (uses gen_tcp.erl)<br>
<br>
The idea here is that they should have an "identical" interface (so<br>
the users should work with both transports).<br>
Obviously the "socket" is different (a gen_tcp socket is a port and<br>
a socket socket is a term()).<br>
<br>
The other socket_test_ttest_tcp* modules "can" be used to see how to<br>
use these...<br>
<br>
Also, there is work on making gen_tcp work with the new socket module.<br>
This is ongoing. There is a lot of special options that gen_tcp/inet<br>
manages (packaging for instance).<br>
<br>
Regards,<br>
/BMK<br>
<br>
On 2019-10-29 10:54, Loïc Hoguin wrote:<br>
> Hello,<br>
> <br>
> I would like to do a similar experiment with RabbitMQ. Would you mind sharing the code that makes <br>
> 'socket' work similar to 'gen_tcp'? That may give me a head start, even if the code is not complete.<br>
> <br>
> Thanks,<br>
> <br>
> On 25/10/2019 12:39, Micael Karlberg wrote:<br>
>> Hi,<br>
>><br>
>> Its early days still, but the goal is definitely that it should<br>
>> be faster.<br>
>><br>
>> Here is some figures from a (time) test tool which is part of the<br>
>> test suite (its basically a ping-pong case), both server and client<br>
>> running on the same host (but in different VMs):<br>
>><br>
>> Two transports:<br>
>> gen: gen_tcp<br>
>> sock: socket (tcp)<br>
>><br>
>> The socket active mode is a simulation of gen_tcp active mode<br>
>> (no active = N).<br>
>><br>
>> The tables are the exchange from the client side.<br>
>><br>
>> With server side using gen (active = false):<br>
>><br>
>> Transport Active Data<br>
>> gen false 10192 byte/ms, 154 msgs/ms<br>
>> gen true 10383 byte/ms, 157 msgs/ms<br>
>> gen once 6003 byte/ms, 90 msgs/ms<br>
>> sock false 14050 byte/ms, 212 msgs/ms<br>
>> sock true 14772 byte/ms, 223 msgs/ms<br>
>> sock once 14050 byte/ms, 210 msgs/ms<br>
>><br>
>><br>
>> With server side using gen (active = true):<br>
>><br>
>> Transport Active Data<br>
>> gen false 9447 byte/ms, 143 msgs/ms<br>
>> gen true 22345 byte/ms, 338 msgs/ms<br>
>> gen once 5532 byte/ms, 83 msgs/ms<br>
>> sock false 15316 byte/ms, 232 msgs/ms<br>
>> sock true 23693 byte/ms, 358 msgs/ms<br>
>> sock once 22068 byte/ms, 334 msgs/ms<br>
>><br>
>><br>
>> With server side using sock (active = false, async = true):<br>
>><br>
>> Transport Active Data<br>
>> gen false 11260 byte/ms, 170 msgs/ms<br>
>> gen true 22273 byte/ms, 337 msgs/ms<br>
>> gen once 7703 byte/ms, 116 msgs/ms<br>
>> sock false 15211 byte/ms, 230 msgs/ms<br>
>> sock true 24778 byte/ms, 375 msgs/ms<br>
>> sock once 23086 byte/ms, 349 msgs/ms<br>
>><br>
>><br>
>> With server side using sock (active = true, async = true):<br>
>><br>
>> Transport Active Data<br>
>> gen false 11351 byte/ms, 171 msgs/ms<br>
>> gen true 22469 byte/ms, 340 msgs/ms<br>
>> gen once 7407 byte/ms, 112 msgs/ms<br>
>> sock false 15484 byte/ms, 234 msgs/ms<br>
>> sock true 24885 byte/ms, 377 msgs/ms<br>
>> sock once 23570 byte/ms, 357 msgs/ms<br>
>><br>
>><br>
>> There is of course a bit of overhead since the socket transport<br>
>> is trying to emulate (part of) gen_tcp.<br>
>><br>
>> This is run on a Dell Precision T1700 running SLES 12 SP2 (not that<br>
>> that effects the relative performance).<br>
>> Run with a snapshot from the maint branch.<br>
>><br>
>> This is obviously not a real use case, but it can be a guideline.<br>
>> Also no UDP test (at the moment).<br>
>><br>
>> /BMK<br>
>><br>
>> On 2019-10-24 18:48, Frank Muller wrote:<br>
>>> Hi Thomas<br>
>>><br>
>>> Is the socket module faster than gen_tcp/gen_udp in your case? If yes, can you please share some <br>
>>> numbers.<br>
>>><br>
>>> /Frank<br>
>>><br>
>>> > case socket:recvfrom(Sock, 0, nowait) of<br>
>>> > {ok, {#{family := Domain,<br>
>>> > port := Port,<br>
>>> > addr := Addr}, Data}} -><br>
>>> > .<br>
>>><br>
>>> Hi Micael, Mark,<br>
>>><br>
>>> thanks for your replies. The snippet above helped me getting the record<br>
>>> matched.<br>
>>><br>
>>> I'm quite impressed with the socket module, it seems to be pretty<br>
>>> complete, at least for my application.<br>
>>><br>
>>> Thanks,<br>
>>> Thomas<br>
>>><br>
> <br>
</blockquote></div>