On Tue, 24 Apr 2001, Henrik Nordstrom wrote:
> Drazen Gemic wrote:
>
> > CARP load balancing does not seem to work. I have disabled ICP
> > communicating, and still get 50%/50% distribution for
> > two CARP parents.
>
> Squid version? Have a wague memory of this being a problem in some..
The latest stable 2.4.
>
> > I have noticed that people involved with Squid do not
> > like CARP very much. Does it mean that this problem
> > is not going to be corrected ?
>
> I like the theory of it, but have never had any reason to use it.
>
> What I do not like is that what we have in Squid is called CARP while it
> only implements a small subset of what actually makes CARP. If something
> on CARP it should be called CARP-hash distribution. From a coding
> perspective about 80% of CARP is missing from Squid (1/5 implemented).
> See http://squid.sourceforge.net/carp/
The part implemented so far is quite enough for my purpose,
and, I belive, for lot of other people neeeds. I think the calculation algorithm
is the problem. I have taken a look at the source, and run squid -k debug in the
same time.
Everything seems to be allright, except the fact that the load distribution is
very close to 50% - 50%, no matter what is specified with carp-load-factor.
Is there an algorithm specification somewhere ? The source module (carp.c)
seems to be relativly small, I'd like to try to find out is there a problem with
my compilation/instalation, or maybe something else.
TNX
DG
Received on Mon Apr 23 2001 - 16:58:38 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:59:34 MST