11–20 of 41 posts
There is nothing new in "modern HTTP servers". They are event-driven programs and this has existed for a long time.
Quickly looking over SoftLayer's price sheet, it looks to be a little over $1k a month for Intel 5670 with 24GB of RAM
You can get a dual E5-2609 with 32gb of ram + bandwidth from a good host like WebNX for $369 / month.
And if you wanted to spend $1k per month, you can get a dual e5-2687W (16 cores @ 3.1ghz) with 128gb ram; with a 4x SAS RAID 10, plus a 120gb ssd drive, plus a 2tb backup freebie, plus 100mbps unmetered bandwidth (30tb). That literally blows the doors off Softlayer.
Granted this is also being tested on localhost with static content (no disk I/O) -- but shows that event-driven servers are not that novel or difficult to write: my code weighs around 1700 LOC of (might I say) readable C.
Static file serving is also fast (using sendfile(), etc), but needs an overhaul to achieve usable concurrency. Currently there's a ~4x performance drop while serving files, but I'm working on this.
(The sources are at http://github.com/lpereira/lwan by the way.)
Quickly looking over SoftLayer's price sheet, it looks to be a little over $1k a month for Intel 5670 with 24GB of RAM
Softlayer is a lot more expensive than most hosts these days. They're closer to being in Rackspace territory now. You can get a dual E5-2609 with 32gb of ram + bandwidth from a good host like WebNX for $369 / month. And if you wanted to spend $1k per month, you can get a dual e5-2687W (16 cores @ 3.1ghz) with 128gb ram; with a 4x SAS RAID 10, plus a 120gb ssd drive, plus a 2tb backup freebie, plus 100mbps unmetered b…
Earlier quoted context omitted.
Softlayer is a lot more expensive than most hosts these days. They're closer to being in Rackspace territory now. You can get a dual E5-2609 with 32gb of ram + bandwidth from a good host like WebNX for $369 / month. And if you wanted to spend $1k per month, you can get a dual e5-2687W (16 cores @ 3.1ghz) with 128gb ram; with a 4x SAS RAID 10, plus a 120gb ssd drive, plus a 2tb backup freebie, plus 100mbps unmetered b…
Literally? Did you literally fall asleep in English class?
Looking over your comments on HN, that latest comment seems to be the peak of your contribution here. In over two years you've picked up 394 karma points. I'm going with the assumption that all of those karma points are from being a pedant about the English language. You're such a valuable member of this community! Rock on.
Sigh, another completely synthetic benchmark with big iron: A dual Intel Xeon X5670 with 24GB of RAM from SoftLayer. The X5670 has 6 cores @ 2.93 GHz, 2 threads per core, /proc/cpuinfo shows 24 CPUs. Serving a plain HTML page over localhost via nginx. A medium length blog post with pretty much no real-world information.
Earlier quoted context omitted.
Literally? Did you literally fall asleep in English class?
Douche much? Looking over your comments on HN, that latest comment seems to be the peak of your contribution here. In over two years you've picked up 394 karma points. I'm going with the assumption that all of those karma points are from being a pedant about the English language. You're such a valuable member of this community! Rock on.
nginx saturates at ~18k req/sec / core with a latency of ~24ms. This saturation is not coming from nginx in particular but from OS limits (mode switching, stack copying etc.. for read/write system calls). There is nothing new in "modern HTTP servers". They are event-driven programs and this has existed for a long time.
Hence in telcom environments where you get such requirements the TCP/IP stack is processed closer to the card - but no out-of-the-box web servers could handle such streams - most of that is handled with e.g. C-libs closely bound to the networking HW.
With such specialized NW cards (16 core card) plus libs you get to > 1'500'000 teardowns p.s. / connections established and roughly 15Gbps - there are now 100Gbps solutions on the market on the top-end - your actual throughput depends mostly what kind of processing you're doing. That kind of equipment is of course not normally used for web serving - more for (transparent) proxying and inspection / traffic shaping, etc as more-and-more telcom cores are completely TCP/IP based.
Earlier quoted context omitted.
Literally? Did you literally fall asleep in English class?
Douche much? Looking over your comments on HN, that latest comment seems to be the peak of your contribution here. In over two years you've picked up 394 karma points. I'm going with the assumption that all of those karma points are from being a pedant about the English language. You're such a valuable member of this community! Rock on.