Version française
Home     About     Download     Resources     Contact us    
Browse thread
Re: OCaml is broken
[ Home ] [ Index: by date | by threads ]
[ Search: ]

[ Message by date: previous | next ] [ Message in thread: previous | next ] [ Thread: previous | next ]
Date: -- (:)
From: Dario Teixeira <darioteixeira@y...>
Subject: Re: [Caml-list] Re: multicore wish
Hi,

> I am beginning using Ocsigen, for a growing web project:
> Is multicore support useless for scaling on Ocsigen?

Categorically, yes.  In fact, I would say that the model used by Ocsigen
is close to being optimal performance-wise as far as web applications are
concerned.  The Ocsigen server and Eliom applications use Lwt for concurrency,
ensuring that the CPU is always busy and will not idle waiting for I/O.
Moreover, the green threads offered by Lwt are much lighter than system
threads, and avoid the context switching penalty incurred by the latter.

And what about those multiple cores?  Simple, if you have n cores, then
simply fire up n instances of the Ocsigen server, and put a dispatching
server like HAProxy or Ocsigen itself as frontend (there are some simple
tricks to ensure that the same client is always directed to same server).
This solution takes advantage of the fact that serving web requests from
multiple clients is embarrasingly parallel from the web application
standpoint. (Sure, you'll have contention on the database side, but
most DBMSs handle that reasonably well).

(And yes, I am aware that Lwt's performance could be improved further
by using syscalls like epoll/kqeueue/etc instead of select.  That is
however an implementation issue, not an architectural flaw).

Best regards,
Dario Teixeira