5 ms·

Very nice work, competition is always good. However, it seems worth mentioning that webservers haven't been a bottleneck for a long time now. Your bottleneck i

by moe 12y ago

Very nice work, competition is always good. However, it seems worth mentioning that webservers haven't been a bottleneck for a long time now. Your bottleneck is always disk I/O, the network, or the slow application server that you're proxying to. For reference: Wikipedia[1] serves roughly 8k pageviews/sec on average for a total of ~20 billion pageviews/month. Assuming each pageview consists of ~10 webserver hits we're looking at ~80k requests/sec. This is within the realm of a single instance of either nginx or h2o on a beefy machine [on a very beefy network]. So, unless you plan to serve Wikipedia or Facebook from a single server, you're probably fine picking your webserver software on the basis of features rather than benchmarks. [1] http://reportcard.wmflabs.org/graphs/pageviews http://reportcard.wmflabs.org/graphs/pageviews

sparkzilla 12y ago
You may be a bit off with your calculation. Wikipedia uses 15,000 cpus on 750 hosts. http://ganglia.wikimedia.org/latest/ http://ganglia.wikimedia.org/latest/
- walterbell 12y ago
  Thanks. Do you know of any references on Wikipedia's architecture for the host types named in the Ganglia report?
- moe 12y ago
  You may be a bit off with your calculation Where? Wikipedia uses 15,000 cpus on 750 hosts. These are not all webservers. Also, even though they could run their load on a fraction of the hardware it would probably not make a lot of sense to optimize for that, as the potential cost savings are relatively small (servers are cheap).
  - jaytaylor 12y ago
    The costs incurred for a system compromised of 750 hosts or 15,000 CPUs are most certainly nontrivial.
    - moe 12y ago
      First off, 750 hosts don't have 15k CPUs. The ganglia metric likely refers to cores and doesn't account for hyperthreading. And, again, 750 hosts is not too bad for an operation the size of Wikipedia. A running server doesn't cost much in the grand scheme of things. You can infer that from the fact that ISPs will rent you one for under $30/mo and still make a profit on you. Consequently the potential savings may not be trivial, but they're hardly big enough to justify tying up large parts of the team in an effort to squeeze out the last x%.

by moe 12y ago

sparkzilla 12y ago

You may be a bit off with your calculation. Wikipedia uses 15,000 cpus on 750 hosts. http://ganglia.wikimedia.org/latest/ http://ganglia.wikimedia.org/latest/

walterbell 12y ago
Thanks. Do you know of any references on Wikipedia's architecture for the host types named in the Ganglia report?
moe 12y ago
You may be a bit off with your calculation Where? Wikipedia uses 15,000 cpus on 750 hosts. These are not all webservers. Also, even though they could run their load on a fraction of the hardware it would probably not make a lot of sense to optimize for that, as the potential cost savings are relatively small (servers are cheap).
- jaytaylor 12y ago
  The costs incurred for a system compromised of 750 hosts or 15,000 CPUs are most certainly nontrivial.
  - moe 12y ago
    First off, 750 hosts don't have 15k CPUs. The ganglia metric likely refers to cores and doesn't account for hyperthreading. And, again, 750 hosts is not too bad for an operation the size of Wikipedia. A running server doesn't cost much in the grand scheme of things. You can infer that from the fact that ISPs will rent you one for under $30/mo and still make a profit on you. Consequently the potential savings may not be trivial, but they're hardly big enough to justify tying up large parts of the team in an effort to squeeze out the last x%.