[erlang-questions] Cowboy feedback needed

[Michael Truog]

A comparison (summary) of cowboy 0.6.1 and misultin 0.9 final in the context of CloudI is here:
https://github.com/okeuday/CloudI/blob/master/src/tests/http_req/loadtest/results_v1_1_0/201210_summary.pdf
(the raw Tsung results are also within the same directory)

The results are just showing the latency when putting cowboy and misultin under 10kreq/s load with both 20k and 40k connections, when the requests go through CloudI messaging into all the supported programming languages (Erlang, C/C++, pure-Java, Python/C, pure-Python, and pure-Ruby.... the "pure" part is where only the target language is used to create the CloudI API, which does the Erlang binary term format CloudI request encoding/decoding).  So, the test is showing the performance of a simple HTTP GET query parameter parse/response using XML (the XML is based on historical misultin testing).

For these tests, it showed cowboy always has less latency which is significant, if the programming language internal latency is not significant.  The cpu usage of cowboy was slightly lower than misultin for high connection counts (40k instead of 20k).  The memory usage of cowboy was significantly lower than misultin.  The test results are for R15B01 and R15B02, just due to when I did the loadtesting.

I will continue to do similar loadtests in the future to make sure and evaluate performance with more recent Erlang releases, as time allows, but the Tsung configurations are within the repository for people to test their own (hardware) environments.

On 01/24/2013 01:23 PM, Loïc Hoguin wrote:
> Hey,
>
> I'm looking into perhaps starting a project related to Cowboy and could use some feedback from users, particularly in the realm of numbers.
>
> If you use Cowboy and have it in production where:
>
>  *  Latency is vital
>  *  Throughput is vital
>  *  Concurrent number of connections is huge
>  *  Load is huge (or would be with another solution)
>
> Then I'd like to hear from you!
>
> Please send me average numbers, statistics, graphs or anything where I can see how well it performs for you! In private if you prefer. Tell me if I can quote you or your company about it. Please answer even if we briefly discussed it in the past.
>
> (If you found that it didn't perform enough for your needs you should probably open a ticket, or, if you can't, send me a private email.)
>
> Looking forward to the feedback. Thanks!
>