[AWS] Further info on the weird "AWS is returning too much data" issue

[Thomas Løcke]

On Tue, Jan 25, 2011 at 10:53 AM, Pascal Obry <pascal at obry.net> wrote:
> I just cannot reproduce on my GNU/Linux Debian box (a dual-core). I'll
> try later today
> on a quad core.
> Document Path:          /
> Document Length:        16 bytes
>
> Concurrency Level:      8
> Time taken for tests:   20.986 seconds
> Complete requests:      100000
> Failed requests:        0
> Write errors:           0
> Total transferred:      17400000 bytes
> HTML transferred:       1600000 bytes
> Requests per second:    4765.11 [#/sec] (mean)
> Time per request:       1.679 [ms] (mean)
> Time per request:       0.210 [ms] (mean, across all concurrent requests)
> Transfer rate:          809.70 [Kbytes/sec] received
> HTML transferred is correct and on the AWS side:
>
> $ grep "200" hello_world-2011-01-25.log | wc
>  100000  900000 6100000
>
> $ grep "200 16" hello_world-2011-01-25.log | wc
>  100000  900000 6100000


Here are my results using the same 16 byte document length:

Server Software:        AWS
Server Hostname:        thomas
Server Port:            8080
Document Path:          /
Document Length:        16 bytes
Concurrency Level:      8
Time taken for tests:   4.964 seconds
Complete requests:      100000
Failed requests:        0
Write errors:           0
Total transferred:      17400348 bytes
HTML transferred:       1600032 bytes
Requests per second:    20146.05 [#/sec] (mean)
Time per request:       0.397 [ms] (mean)
Time per request:       0.050 [ms] (mean, across all concurrent requests)
Transfer rate:          3423.32 [Kbytes/sec] received

$ grep "200 16" hello_world-2011-01-25.log | wc -l
100004

Here's a further discrepancy, because AWS is logging 100004 requests,
but as can be seen in the ab data, only 32 bytes of excessive HTML has
been received by ApacheBench, which equals 2 x 16 byte documents

But what is really fascinating, is the fact that this computer
completes the test more than 5 times faster than the one you've used
for your test. A lot of bad things can happen once things get fast
enough, and 20146.05 requests per second is _very_ fast. And as stated
earlier, I cannot reproduce this on slower systems.

Are there anybody on the list with some wicked fast computers that
would like to chime in with their results?

:o)
Thomas Løcke