[AWS] Praise - and a very simple (and early) benchmark.

[Thomas Løcke]

I replied to Emmanuel's post earlier, but forgot to send the reply to
the AWS list..

So here it is. Sorry for the duplicate Emmanuel.  :o)


On Fri, Jan 21, 2011 at 4:35 PM, Emmanuel Briot <briot at adacore.com> wrote:
>
> A follow up to Thomas's benchmark.
>
> I did a test with a hello world server written in node.js (a javascript
> framework for writing faster servers). It has the reputation of being very fast.
>
> I compared with the hello world demo provided with AWS, compiled with -O2.
> The demo was modified to use 50 threads


Have you tried lowering the amount of AWS threads to equal your actual
amount of CPU cores?

I've found that AWS does not (on my system at least) scale well if you
go to far beyond the actual amount of CPU cores. On my Sun Ultra 27
the sweet spot seems to be 5.


> Here are the results for node.js
>
> Document Length:        12 bytes
> Time taken for tests:   0.143 seconds
> Complete requests:      1000
> Total transferred:      76000 bytes
> HTML transferred:       12000 bytes
> Requests per second:    7007.12 [#/sec] (mean)
> Time per request:       14.271 [ms] (mean)
> Time per request:       0.143 [ms] (mean, across all concurrent requests)
> Transfer rate:          520.06 [Kbytes/sec] received
>
> --------------------
> And the results for AWS
>
> Document Length:        12 bytes
> Time taken for tests:   0.189 seconds
> Complete requests:      1000
> Total transferred:      174590 bytes
> HTML transferred:       12324 bytes
> Requests per second:    5299.95 [#/sec] (mean)
> Time per request:       18.868 [ms] (mean)
> Time per request:       0.189 [ms] (mean, across all concurrent requests)
> Transfer rate:          903.63 [Kbytes/sec] received


There's a _huge_ difference in total transferred. Staggering. I wonder
how things would've looked if AWS hadn't had to push more than twice
as much data.


> Although AWS has a faster transfer rate, it is actually slower than a server
> written in javascript (fewer requests per second and time to process the
> requests longer).
>
> The strange thing is the amount of text that is transfered by Node.js compared
> to that of AWS. I suspect AWS loses time formatting headers correctly, whereas
> Node.js (in this basic hello_world demo) doesn't waste time with it. Maybe this
> accounts for the different, but it might be interesting to investigate if
> someone feels up to it.


Would you mind pastebinning your node.js stuff?