On Thu, 2008-10-02 at 19:11 +0200, Stefan de Konink wrote:
> This probably scales well. But now; what if you a priori know:
>
> 1) One Brave Chief cannot handle the load due to heavy SSL
> 2) The amount of traffic to one machine is just too much
> 3) If failing = big problems
>
>
> I wonder what the users here would choose. Would you go with something
> like LVS, or would you take another balancing approach?
It really depends. For most situations, I try to make sure that the site
operator has to make only minimal (if any) changes to their code.
Sometimes sessions can be problematic, other times the application
itself must have at least one shared mount for certain configuration
files or other temporary data.
This works well for me:
[ CVIP ]
/ \
[ crossroads] [ crossroads ]
/ \ / \
[ cherokee ] [ cherokee ] [ cherokee ] [ cherokee ]
/ \ / \ / | \ / \
[ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ]
[ fcgi-php as you noted prior ]
It is relatively easy to scale and / or package as virtual appliances to
be deployed on a farm. There are many other options other than
crossroads for load balancing too.
Also note, each LB should be balancing over all 4 cherokee nodes ..
which is not depicted because its just an ad hoc ascii drawing in an
e-mail. The same goes for cherokee -> fcgi servers.
Also notable, in my experience PHP / Cherokee is not the usual culprit
when things get 'hot' .. its usually MySQL being sluggish causing PHP to
live much longer than it should resulting in prolonged connections all
the way down the pipe.
Cheers,
--Tim
_______________________________________________
Cherokee mailing list
[email protected]
http://lists.octality.com/listinfo/cherokee
