Thanks for the link. I removed the send_interval call and retested but it didn't make much of a difference. It still takes about 7 and half minutes to create 100K connections. During creation I can see several time outs from client side. And the creation speed gets slower when the number of connections gets higher. Although the average CPU usage doesn't change much.<div>
<br></div><div>After creation, with 1.5MB/s traffic, the CPU usage is still about the same as before which is 30%.</div><div><br></div><div>Cheers.</div><div>Andy<br><br><div class="gmail_quote">On Wed, Jun 22, 2011 at 10:34 PM, Loïc Hoguin <span dir="ltr"><<a href="mailto:essen@dev-extend.eu" target="_blank">essen@dev-extend.eu</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Memory usage sounds a bit high for cowboy_http_req:compact. But maybe<br>
not. Good point though it's already in the examples just like hibernate.<br>
My bad.<br>
<br>
On the other hand if you have a timer per connection it's definitely not<br>
going to scale, timers are very slow so having 100k timers isn't going<br>
to work well. Remove the send_interval and it should work a lot better.<br>
<br>
For future reference:<br>
<a href="http://www.erlang.org/doc/efficiency_guide/commoncaveats.html#id58959" target="_blank">http://www.erlang.org/doc/efficiency_guide/commoncaveats.html#id58959</a><br>
<br>
Waiting for yet another comparison, have fun. :)<br>
<div><br>
On 06/22/2011 04:27 PM, Andy W. Song wrote:<br>
> I actually didn't patch. I want to keep my change as small as possible.<br>
> I will remove that send_interval and test it again. Sorry I didn't pay<br>
> much attention about this.<br>
><br>
> BTW, how can you tell I used or did not use cowboy_http_req:compact? And<br>
> I see hibernate is already the default behavior, isn't it?<br>
><br>
> Thanks<br>
> Andy<br>
><br>
> On Wed, Jun 22, 2011 at 10:06 PM, Loïc Hoguin <<a href="mailto:essen@dev-extend.eu" target="_blank">essen@dev-extend.eu</a><br>
</div><div><div></div><div>> <mailto:<a href="mailto:essen@dev-extend.eu" target="_blank">essen@dev-extend.eu</a>>> wrote:<br>
><br>
> Hello,<br>
><br>
> Did you fix the many issues in your cowboy application code I reported<br>
> in my previous emails? I see you removed the receive after from<br>
> misultin, did you remove it from cowboy too? You could very well just be<br>
> measuring Erlang's timer modules bad performance there.<br>
><br>
> Also I see you're not using cowboy_http_req:compact and the hibernate<br>
> option for cowboy that is doing it properly (unlike your patch).<br>
><br>
> And again, you should probably come to IRC and ask for advice from the<br>
> other people benchmarking it.<br>
><br>
> Cheers.<br>
><br>
> On 06/22/2011 03:14 PM, Andy W. Song wrote:<br>
> > OK, the competition comes again. I finally wrote a client that<br>
> performs<br>
> > decently. It's here<br>
> <<a href="https://github.com/awsong/Golang-WebSocket-Client" target="_blank">https://github.com/awsong/Golang-WebSocket-Client</a>>.<br>
> ><br>
> > I tested my previous Cowboy first( I changed send_interval from<br>
> 1000 to<br>
> > 900000 of line 64 of src/websocket_handler.erl in cowboy_examples ).<br>
> > Here is the version:<br>
> > *Cowboy_examples*: 343d002 Initial commit with a default and websocket<br>
> > handler.<br>
> > *Cowboy*: 5d69825 Add a max_connections transport option, Mon May 9<br>
> > 22:01:31 2011<br>
> ><br>
> > The command line I use on the client is "./conn -c 10000 -n 10". It'll<br>
> > create 100K connections, 10K per IP to the server. It takes about<br>
> 7 and<br>
> > half minutes to finish creating these connections. During the<br>
> creation,<br>
> > server CPU usage is about 110%(dual core), client cpu usage is<br>
> under 20%.<br>
> ><br>
> > Server memory usage(Cowboy residence memory) :<br>
> > Right after Cowboy starts: 525M<br>
> > After 100K connection established: 1842M<br>
> > After 5M packets passed: 2824M<br>
> > After 9M packets passed: 3231M<br>
> > After 87.7 M packets passed, 5746M<br>
</div></div>> > [{total,<a href="tel:4155518040" value="+14155518040" target="_blank">4155518040</a> <tel:<a href="tel:4155518040" value="+14155518040" target="_blank">4155518040</a>> <tel:<a href="tel:4155518040" value="+14155518040" target="_blank">4155518040</a><br>
<div><div></div><div>> <tel:<a href="tel:4155518040" value="+14155518040" target="_blank">4155518040</a>>>},<br>
> > {processes,1701318856},<br>
> > {processes_used,1701299096},<br>
> > {system,2454199184},<br>
> > {atom,497457},<br>
> > {atom_used,482279},<br>
> > {binary,1787851232},<br>
> > {code,4041219},<br>
> > {ets,41231176}]<br>
> > There seems a memory leak. If I don't stop, the memory usage will keep<br>
> > going higher. 40% CPU, 1.5 MB/s traffic.<br>
> ><br>
> > *Latest Cowboy and Cowboy_examples* (same send_interval change). It<br>
> > takes about 7 and half minutes to finish creating 100K connections.<br>
> > During the creation, server CPU usage is about 110%(dual core), client<br>
> > cpu usage is under 20%. After 158.9M packets passed, 1464M residence<br>
> > memory used, 30% CPU, 1.5 MB/s traffic, .<br>
> ><br>
> > *Latest Misultin* from github, with patch (see attachment):<br>
> > It takes about 45s to finish creating 100K connections, during<br>
> creation<br>
> > cpu usage is 120%. After creation, residence memory usage is<br>
> 2684M. And<br>
> > it goes down slowly to 1416M after 40M packets are passed, cpu<br>
> usage is<br>
> > about 50%<br>
> ><br>
> > Another thing worth to note is that when I hit CTRL-C on the client<br>
> > side, misultin will quickly destroy the connections and release memory<br>
> > and go back to normal. But Cowboy will consume almost all the cpu time<br>
> > and stay there for a long time (I didn't have the patience to measure<br>
> > how long, so just hit ctrl-c to quick erlang).<br>
> ><br>
> > Cheers<br>
> > Andy<br>
> ><br>
> ><br>
> > On Wed, Jun 1, 2011 at 8:26 PM, Andy W. Song <<a href="mailto:wsongcn@gmail.com" target="_blank">wsongcn@gmail.com</a><br>
> <mailto:<a href="mailto:wsongcn@gmail.com" target="_blank">wsongcn@gmail.com</a>><br>
</div></div><div>> > <mailto:<a href="mailto:wsongcn@gmail.com" target="_blank">wsongcn@gmail.com</a> <mailto:<a href="mailto:wsongcn@gmail.com" target="_blank">wsongcn@gmail.com</a>>>> wrote:<br>
> ><br>
> > BTW, the latest version uses less memory too, about 3.2G for 200K<br>
> > connections.<br>
> ><br>
> ><br>
> > On Wed, Jun 1, 2011 at 8:23 PM, Andy W. Song<br>
> <<a href="mailto:wsongcn@gmail.com" target="_blank">wsongcn@gmail.com</a> <mailto:<a href="mailto:wsongcn@gmail.com" target="_blank">wsongcn@gmail.com</a>><br>
</div><div>> > <mailto:<a href="mailto:wsongcn@gmail.com" target="_blank">wsongcn@gmail.com</a> <mailto:<a href="mailto:wsongcn@gmail.com" target="_blank">wsongcn@gmail.com</a>>>> wrote:<br>
> ><br>
> > Missed one attachment.<br>
> ><br>
> ><br>
> > On Wed, Jun 1, 2011 at 8:21 PM, Andy W. Song<br>
> <<a href="mailto:wsongcn@gmail.com" target="_blank">wsongcn@gmail.com</a> <mailto:<a href="mailto:wsongcn@gmail.com" target="_blank">wsongcn@gmail.com</a>><br>
</div><div>> > <mailto:<a href="mailto:wsongcn@gmail.com" target="_blank">wsongcn@gmail.com</a> <mailto:<a href="mailto:wsongcn@gmail.com" target="_blank">wsongcn@gmail.com</a>>>> wrote:<br>
> ><br>
> > I tested out the latest Misultin with this change, CPU<br>
> usage<br>
> > is about 2 to 3 percent lower.<br>
> ><br>
> > With the attached Misultin package, I can still reproduce<br>
> > the large packet problem.<br>
> ><br>
> > With latest version, Misultin doesn't report any<br>
> error. But<br>
> > it doesn't reply to the client either. Seems it just drops<br>
> > the packet. Another attached file is the tcpdump result.<br>
> ><br>
> > I will try Mochiweb later just to compare.<br>
> ><br>
> ><br>
> > On Wed, Jun 1, 2011 at 12:09 AM, Roberto Ostinelli<br>
> > <<a href="mailto:roberto@widetag.com" target="_blank">roberto@widetag.com</a> <mailto:<a href="mailto:roberto@widetag.com" target="_blank">roberto@widetag.com</a>><br>
</div><div><div></div><div>> <mailto:<a href="mailto:roberto@widetag.com" target="_blank">roberto@widetag.com</a> <mailto:<a href="mailto:roberto@widetag.com" target="_blank">roberto@widetag.com</a>>>> wrote:<br>
> ><br>
> ><br>
> > On May 31, 2011, at 12:54 PM, Andy W. Song wrote:<br>
> ><br>
> >> Would you mind testing it using my C client? Search<br>
> >> "->len" and give<br>
> >> it a value between 1500 and 2000, hopefully<br>
> you'll see<br>
> >> the result I<br>
> >> saw.<br>
> >><br>
> >> Regards<br>
> >> Andy<br>
> ><br>
> > Hi Andy,<br>
> ><br>
> > will do that.<br>
> ><br>
> > Meanwhile, I've added an undocumented option,<br>
> > {ws_no_header, true} so that headers are not<br>
> duplicated<br>
> > in the record and should use less memory. plus, I've<br>
> > optimized binary code for websockets.<br>
> ><br>
> > can you please retry the test so we have a matter of<br>
> > comparison?<br>
> ><br>
> > the only thing you need to change is to add this<br>
> option<br>
> > in misultin initialization, i.e.:<br>
> ><br>
> > % start misultin http server<br>
> > start(Port) -><br>
> > misultin:start_link([<br>
> > *{ws_no_header, true},*<br>
> > {port, Port}, {loop, fun(Req) -> handle_http(Req,<br>
> Port)<br>
> > end},<br>
> > {ws_loop, fun(Ws) -> handle_websocket(Ws) end},<br>
> > {ws_autoexit, false}<br>
> > ]).<br>
> ><br>
> > please let me know...<br>
> ><br>
> > r.<br>
> ><br>
> ><br>
> ><br>
> ><br>
> > --<br>
> ><br>
> ---------------------------------------------------------------<br>
> > 有志者,事竟成,破釜沉舟,百二秦关终属楚<br>
> > 苦心人,天不负,卧薪尝胆,三千越甲可吞吴<br>
> ><br>
> ><br>
> ><br>
> ><br>
> > --<br>
> ><br>
> ---------------------------------------------------------------<br>
> > 有志者,事竟成,破釜沉舟,百二秦关终属楚<br>
> > 苦心人,天不负,卧薪尝胆,三千越甲可吞吴<br>
> ><br>
> ><br>
> ><br>
> ><br>
> > --<br>
> > ---------------------------------------------------------------<br>
> > 有志者,事竟成,破釜沉舟,百二秦关终属楚<br>
> > 苦心人,天不负,卧薪尝胆,三千越甲可吞吴<br>
> ><br>
> ><br>
> ><br>
> ><br>
> > --<br>
> > ---------------------------------------------------------------<br>
> > 有志者,事竟成,破釜沉舟,百二秦关终属楚<br>
> > 苦心人,天不负,卧薪尝胆,三千越甲可吞吴<br>
> ><br>
> ><br>
> ><br>
> > _______________________________________________<br>
> > erlang-questions mailing list<br>
</div></div>> > <a href="mailto:erlang-questions@erlang.org" target="_blank">erlang-questions@erlang.org</a> <mailto:<a href="mailto:erlang-questions@erlang.org" target="_blank">erlang-questions@erlang.org</a>><br>
<div><div></div><div>> > <a href="http://erlang.org/mailman/listinfo/erlang-questions" target="_blank">http://erlang.org/mailman/listinfo/erlang-questions</a><br>
><br>
><br>
> --<br>
> Loïc Hoguin<br>
> Dev:Extend<br>
><br>
><br>
><br>
><br>
> --<br>
> ---------------------------------------------------------------<br>
> 有志者,事竟成,破釜沉舟,百二秦关终属楚<br>
> 苦心人,天不负,卧薪尝胆,三千越甲可吞吴<br>
><br>
<br>
<br>
</div></div><font color="#888888">--<br>
Loïc Hoguin<br>
Dev:Extend<br>
</font></blockquote></div><br><br clear="all"><br>-- <br><div>---------------------------------------------------------------</div><div>有志者,事竟成,破釜沉舟,百二秦关终属楚</div><div>苦心人,天不负,卧薪尝胆,三千越甲可吞吴</div><br>
</div>