1000 requests per second?

[John Moore]

luben karavelov wrote:
> owkaye wrote:
>> Greetings,
>>
>> Can nginx -- running on one server -- deliver 1000 requests per 
>> second without "bogging down" and pushing more and more requests into 
>> a queue? 
>> Here's my reason for asking:
>>
>> I'm designing a live auction website that needs to respond to 
>> 500-1000 requests per second for about an hour.  Each request will 
>> post only 20 bytes of data so the volume being posted is low.  
>> Nevertheless the HTTP headers still need to be parsed and they will 
>> have far more volume than the actual post data -- so it seems I 
>> should do everything I can to reduce the HTTP header overhead.  This 
>> will substantially reduce the load and speed up nginx's response 
>> times, correct?
>>
>> I'm wondering if nginx has the ability to use "Web Sockets" 
>> technology to eliminate all but the first HTTP header, and maintain a 
>> connection with the browser so data can be passed back and forth faster?
>>
>> http://www.w3.org/html/wg/html5/#network
>>
>> If this is not possible, can you tell me the best way to reduce the 
>> HTTP header overhead so I can make sure that each of those 1000 
>> requests per second are responded to as fast as they come in?  Or am 
>> I concerned about something that's a non-issue, perhaps because nginx 
>> is so blazing fast that it can handle this kind of load without 
>> breaking a sweat?
>>
>> The worst problem I can imagine is that during one of these live 
>> auctions the server will begin to respond slowly and push requests 
>> into a queue.  If this happens, bidders will not receive timely 
>> updates from the server and then the whole service loses credibility.
>>
>> If Web Sockets is not an option, perhaps using Javascript in the 
>> visitor's browsers to send requests via XMLHttpRequest is the 
>> next-best option for reducing overhead?
>>
I suspect that nginx would be able to handle this without at problem, 
although I would suggest that the best way to find out is to do some 
load testing. How are the responses built? If that involves database 
lookups, for example, it's more likely that your bottleneck will be 
there. Incidentally, I don't think XMLHttpRequest will help you - from 
the server's point of view, it's just another request, albeit one 
serving a smaller amount of data.