2

This curiosity came to me because I'm building a site similar to this one and needed an already existing site as a model, to compare the benchmarks with my site's own.

What's better than Stack Overflow/Stack Exchange?

I've tried nearly everything so far, but the results are far from the reality, corrupted someway, or more probably I've tried to benchmark the wrong thing.

So, out of curiosity I've pinged meta.stackoverflow.com which gave me the ip 198.252.206.140, made a benchmark of 100 requests with concurrency of 5:

ab -n 100 -c 5 http://198.252.206.140/

Tried with other ports and values but the result is always like this:

Server Software:        
Server Hostname:        198.252.206.140
Server Port:            80

Document Path:          /
Document Length:        3888 bytes

Concurrency Level:      5
Time taken for tests:   15.195 seconds
Complete requests:      100
Failed requests:        0
Write errors:           0
Non-2xx responses:      100
Total transferred:      399200 bytes
HTML transferred:       388800 bytes
Requests per second:    6.58 [#/sec] (mean)
Time per request:       759.759 [ms] (mean)
Time per request:       151.952 [ms] (mean, across all concurrent requests)
Transfer rate:          25.66 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:      175  312 107.5    296     620
Processing:   197  417 332.8    336    2109
Waiting:      178  346 233.4    309    1509
Total:        384  729 363.2    636    2422

Percentage of the requests served within a certain time (ms)
  50%    636
  66%    799
  75%    832
  80%    849
  90%    921
  95%   1731
  98%   2029
  99%   2422
 100%   2422 (longest request)

Obviously they are corrupted, which leads me to think I'm benchmarking the wrong thing.

For future visitors who may need this information, and for mere curiosity, is someone aware if such benchmark was ever made?

EDIT:

It looks like my results were "corrupted" due to latency that I didn't take account of (that's dumb I know).

  • 4
    Stack Exchange serves many sites from the same IP address. When you make a request to the IP address, without using a hostname, their servers don't know what site you're trying to access, and they return an error page. That's why all of your requests got non-200/OK responses. That should not be the case if you request an actual domain, such as stackoverflow.com, instead of the IP. (However, note that if you perform a lot of high-frequency requests like this, you're likely to be temporarily automatically blocked by Stack Exchange's network.) – Jeremy Banks Apr 14 '14 at 21:14
  • @JeremyBanks Got it. The results are the same since the first benchmark. The request time is 1 + ¼ of second, with 1000 requests it's even worse (1006.537 [ms] (mean)), it seems unrealistic because many more requests than that are served. Even my medium sized app is 10x times faster than the results you see on my question, so there must be something wrong. That's what you now told in your comment. – gw0 Apr 14 '14 at 21:27
  • I tried ab -n 100 -c 5 http://stackoverflow.com/, and I got a much faster result: Time per request: 87.505 [ms] (mean), with min=61 and max=128. I don't know where you're located; is it possible that your connection just has a high latency to reach any American servers? – Jeremy Banks Apr 14 '14 at 22:22
  • @JeremyBanks Uhm..that's a lot faster than what I had, somewhat more reliable so to say. Now that I think about it, it is possible that it has something to do with latency as you pinpoint. As of now I live in central Europe and connect through a remote wifi hotspot located circa 750m from my usual location. Latency doesn't show itself in a normal browsing session, but guess that with benchmarks it does.. it surely has to do a trip long 750m for every request. I feel dumb for not noticing such a trivial issue.. – gw0 Apr 15 '14 at 7:23

You must log in to answer this question.

Browse other questions tagged .