Re: Re: Allowing multiple, simultaneous, non-blocking queries.

Adam Richardson wrote: >> "Threading" is only realistically needed when you have to get data from > > multiple sources; you may as well get it all in parallel rather than > > sequentially to limit the amount of time your application / script is > > sitting stale and not doing any processing. > > >> In the CLI you can leverage forking to the process to cover this. > > >> When working in the http layer / through a web server you can leverage > > http itself by giving each query its own url and sending out every > > request in a single http session; allowing the web server to do the > > heavy lifting and multi-threading; then you get all responses back in > > the order you requested. > > > Regarding leveraging http to achieve multi-threading-like capabilities, I've > tried this using my own framework (each individual dynamic region of a page > is automatically available as REST-ful call to the same page to facilitate > ajax capabilities, and I tried using curl to parallel process each of the > regions to see if the pseudo threading would by an advantage.) > > In my tests, the overhead of the additional http requests killed any > advantage that might have been gained by generating the dynamic regions in a > parallel fashion. Do you know of any examples where this actually improved > performance? If so, I'd like to see them so I could experiment more with > the ideas. Hi Adam, Good question, and you picked up on something I negated to mention. With HTTP/1.1 came a little used addition which allows you to send multiple requests through a single connection - which means you can load up multiple requests and receive all the responses in sequence through in a single "call". Thus rather than the usual chain of: open connection send request receive response close connection repeat you can actually do: open connection send requests 1-10 receive responses 1-10 close connection The caveat is one connection per server; but it's also interesting to note that due to the "host" header you can call different "sites" on the same physical machine. I do have "an old class" which covers this; and I'll send you it off-list so you can have a play. In the context of this; it is also well worth noting some additional bonuses. By factoring each data providing source (which could even be a single sql query) in to scripts of their own, with their own URIs - it allows you to implement static caching of results via the web server on a case by case basis. A simple example I often used to use would be as follows: uri: http://example.com/get-comments?post=123 source: <?php // if the static comments query results cache needs updated // or doesn't exist then generate it if( file_exists('/query-results/update-comments-123') || !file_exists('/query-results/comments-123') ) { if( $results = $db->query($query) ) { // only save the results if they are good file_put_contents( '/query-results/comments-123', json_encode($results) ); } } echo file_get_contents('/query-results/comments-123'); exit(); ?> I say "used to" because I've since adopted a more restful & lighter way of doing things; uri: http://example.com/article/123/comments and my webserver simply returns the static file using os file cache and its own cache to keep it nice and speedy. On the generation side; everytime a comment is posted the script which saves the comment simply regenerates the said file containing the static query results. For anybody wondering why.. I'll let ab to the talking: Server Software: Apache/2.2 Server Hostname: 10.12.153.70 Server Port: 80 Document Path: /users/NMR/70583/forum_post Document Length: 10828 bytes Concurrency Level: 250 Time taken for tests: 1.432020 seconds Complete requests: 10000 Failed requests: 0 Write errors: 0 Total transferred: 110924352 bytes HTML transferred: 108323312 bytes Requests per second: 6983.14 [#/sec] (mean) Time per request: 35.800 [ms] (mean) Time per request: 0.143 [ms] (mean, across all concurrent requests) Transfer rate: 75644.20 [Kbytes/sec] received Yes, that's 6983 requests per second completed on a bog standard lamp box; one dual-core and 2gb ram. reason enough? Regards!