|
Hi Thomas,
Have you thought about the algorithm that I posted? It aims for both
locality and load balancing for cache cluster. You can read the source
code for detailed information directly.
I don't think consistent hashing scheduler can work well. Static mapping
will overload a cache server, and you never know which IP address is
"hot" when packets are redirected.
LBLC's strategy is to assign a new destionation to least loaded server,
and keep sending packets for the same destination to the same server,
unless the server is overloaded. Then, allocate the least loaded server
for this destination.
On Wed, 1 Nov 2000, Thomas Proell wrote:
> Hi!
>
> > As for the one in WCCP and your consistent hashing scheduler, they may
> > have overloading problem, for example, requests for several big sites
> > (several destination IP addresses) may be directed to one cache node, it
> > can overload this cache node, while other cache nodes may stay idle.
>
> Not any more. I have already implemented a solution for these "hot spots".
> I'll do some testing and it'll hopefully released next week.
>
> > As for the code of consistent hashing scheduler, it still needs some
> > tuning. Please don't be angry. :)
>
> :-)
> Yes, I agree. I just released it for the people to play with it.
> But I'm interested in improving it, so that it may become an official
> part of LVS. The problem is that I have to start writing the thesis
> now, and I'll be very busy till end of the year.
>
> But maybe I have some spare time once, so I may tune it a bit.
> What are you main critics?
>
1. I don't think the algorithm is good. :)
2. The scheduler can only work for one service. Scheduler are usually
bound to several services.
3. Hash method is not good.
4. code is not neat. :)
Cheers,
Wensong
>
>
> Thomas
>
|
