[Freeciv-Dev] Re: (PR#5081) huge send buffer

["kersten@xxxxxxxxxx" <kersten@xxxxxxxxxx>]

On Thu, 14 Aug 2003, Christian Knoke wrote:

> Just some remarks:
> 
> On Wed, Aug 13, 2003 at 11:08:05PM -0700, kersten@xxxxxxxxxx wrote:
> > On Wed, 13 Aug 2003, Vasco Alexandre da Silva Costa wrote:
> > > On Wed, 13 Aug 2003, Christian Knoke wrote:
> > > 
> > > >
> > > > CVS 13 AUG 2003 GTK 1
> > > >
> > > > big game, slow machine, one player client, two observer clients.
> > > >
> > > > at turn done, I kill both observer clients:
> > > >
> > > > Spiel gespeichert als 'carthago+1730.sav.gz'
> > > > >
> > > > 2: Verbindung verloren: peter von localhost (Spieler John A. MacDonald)
> > > > (Beobachter).
> > > > >
> > > > 2: Verbindung verloren: paul von localhost (Spieler Charlemagne)
> > > > (Beobachter).
> 
> I killed the client here.
> But has the server really closed the connection at this point?
> 
> > > > >
> > > > 2: Verbindung verloren: paul von localhost (Verbindung unvollständig).
> > > > >
> > > > 1: cut connection paul von localhost (Verbindung unvollständig) due to 
> > > > huge
> > > > send buffer (2)
> 
> It looks as it didn't, couldn't flush the buffer ...
> 
> > > Basically, for whatever reason, that client was not receiving fast enough.
> > > We limit the size of the send buffers in order to not waste too much
> > > memory in the server. No, they are not statically allocated.
> > > 
> > > ---
> > > Vasco Alexandre da Silva Costa @ Instituto Superior Tecnico, Lisboa
> > > 
> > 
> > Same with me here: large game, no observers, fast machine, server running 
> > on localhost; client takes forever to display the movements of all units. 
> 
> How fast is your machine?
>

2.66Ghz PIV.
 
> > In the first part of the move they crawl, then suddenly they speed up. The 
> > machine takes 100% CPU. That's a lot. The server disconnects the client 
> > almost every turn due to ping timeouts.
> 
> Anybody else seeing ping timeouts? I didn't. You refer to CVS version I
> suppose.

Yep.

> 
> > Restarting the client from scratch 
> > speeds things up, reconnecting from the existing client makes it worse. 
> 
> *Maybe* related to PR#3769, what i consider a serious bug, and the problem
> also emerges only after 2 or 3 turns?
>

It occurs only in the endgame, with lots of units and a (alomost) complete 
vision. Yes, after 2 or 3 tuns I'm back where I started (disconnected).

Just an observation: the AI moves *lots* of units accross its continent in 
the crawling stage, and after popup of a still-to-be-filled message 
window, things speed up dramaticaly and de AI moves almost all its units 
in other directions than before. Seems odd to me.
 
> > After a few turns the cycle restarts. Memory consumage rises as well. I'm 
> > using the GTK 2.0 client.
> 
> Memory consumage I see here, too, occasionally, but this is hard to
> quantisize due to virtual memory management.
> 

I agree, but the rise is significant (swap 70%, mem 90%). There were no 
other applications to address this to.

> > Any clues?
> > -- 
> > Pieter J. Kersten
> > 
> 
> Christian

HTH,
-- 
Pieter J. Kersten