Re: [Newbie] best strategy for querying multiple servers at a time

[Dan Winship <danw gnome org>]

On 03/23/2010 09:14 AM, Guilhem Bonnefille wrote:
> The question is: what design should I select:
> - a single SoupSessionAsync, queuing all requests to all servers?
> - a SoupSessionAsync per server?

A single session. SoupSession already has the logic for dealing with
requests to multiple servers. Even WebKitGTK only uses a single
SoupSession. (In general, the only reason you'd need to create multiple
sessions is if you need the two session to behave differently. Eg,
having two different sets of passwords or cookies for a single server.)

> For a concrete example, imagine I need to download tiles from servers
> s1 and s2. At a selected zoom level, I need to download 9 tiles on
> each servers. So, I queue all the 18 requests. But, s1 is faster than
> s2 (for example twice faster). So, I wished to retrieve 2 tiles from
> s1 and 1 tile from s2, then 2 tiles from s1 and 1 tile from s2... and
> so on. And I don't want to download all tiles from s1 BEFORE
> downloading tiles from s2.

That's basically what would happen, except that by default libsoup will
send 2 requests at a time per server (as configured by SoupSession's
"max-conns-per-host" property). So at time=0 it would send out 2
requests to each server. At time=1 it would get back responses from s1,
and send out 2 more requests. At time=2 it would get back responses to
the second set of s1 requests and the first set of s2 requests, and then
send out 2 new requests to each server, etc. So the s1 requests will
finish first (at time=5), but the s2 requests will be halfway done at
that point.

-- Dan