Re:Next year's computers same speed but more cores/cpus: dev...
by jsankey on Thursday October 01, 2009 @04:46AM
(#854545)
A mitigating factor is that not much software is CPU-bound: many desktop apps spend most of their time idly waiting for input, and most web apps have already been dealing with concurrency by handling multiple users (and scaling out).
[ Parent
| on HN ]
Re:Next year's computers same speed but more cores/cpus: dev...
by jrockway on Thursday October 01, 2009 @05:00AM
(#854565)
The reason people have such trouble scaling web apps is that they think the above is true.
Re:Next year's computers same speed but more cores/cpus: dev...
by jsankey on Thursday October 01, 2009 @06:43AM
(#854676)
Could you elaborate? Are you saying that apps are CPU bound or that web apps have not been dealing with concurrency even before multi-core? Perhaps you mean that web app authors care about CPU usage too little, which may be true, but I still don't think that multi-core marks a huge change for them.
Re:Next year's computers same speed but more cores/cpus: dev...
by dagw on Thursday October 01, 2009 @08:16AM
(#854772)
The thing to remember is that there are two sides to a web-app. The client and the server. The web-app I'm currently working on uses very little CPU client side, but on the server side the number of simultaneous users I can handle scales more or less linearly with the number of cores the server has (assuming enough memory). So personally multiple cores (and using the correctly) makes huge difference to me and the web-app I'm working on.
Re:Next year's computers same speed but more cores/cpus: dev...
by jsankey on Thursday October 01, 2009 @10:17AM
(#854935)
I apologise for not being clear: certainly multi-core makes a difference to the performance of web apps. My point is that most web apps don't need a drastic change in the way they are written to take advantage of this: they are already multithread or multiprocess to handle concurrent users. Contrast this to single-user desktop apps which have simpler threading models and are less likely to be "naturally" parallelised to take advantage of multi-core.