Re: Proposed COSS patch

On Tue, Aug 01, 2006, Steven wrote:
> As mentioned yesterday, I've been doing a fair amount of work on COSS over
> the past few weeks. Attached is a patch that I've come up with that fixes
> a couple of bugs, and adds a couple of configuration options to COSS.

Looks fine to me. Its good to see someone else taking an interest in the
COSS code.

The "don't relocate object unless we're X stripes away from it" is a good
idea.

I also like the 'memory only' stripe idea. The 'clean' way would be to
have two types of object reads - one that reads, another that relocates
and reads - but your patch is simpler and gets the job done. Which is good.
Its good that you've dropped the write IO and worked on fixing the load
balancing across stores. The selection algorithm is pretty low-tech;
I remember hacking it up thinking "someone will come up with something
better as a better store is written". 5 years later.. :)

So good job. I'd like to get some more data back from the COSS method
of disk IO and look at some optimisations. But I think, for now,
the COSS throughput should be sufficient to keep squid users happy.
(Again, commercial caches are far far ahead at the present time
and there doesn't seem to be much money available for open-source
cache software work.)

Adrian
Received on Tue Aug 01 2006 - 06:15:22 MDT