dormando dormando at rydia.net
Wed Aug 22 09:46:00 UTC 2007


Tossing up a summary of my thoughts (since I'm up babysitting backups 
anyway). Thanks to Matt and Sun for hosting!

Some ... better preparation might've been good. I think memcached just 
doesn't really have any "very active" developers, and a lot of 
enthusiasts are at the hackathons. So they become a good source of 
discussion but are hard to focus into direction/code. Discussion ain't 
bad tho.

I wasn't in all of the discussions, but I have some takeaways:

- Stable fast C-backed libraries are a big deal. Most folks there 
(seemed to?) agree that their biggest performance boost are related to 
client fixes.

- pecl/memcached needs work, but no one's maintaining it? Are there 
alternatives? The PHP client frontier is mired. I tried to look into an 
error but found the same error message was copy/pasted in nine different 
places in the code. Not worth it.

- binary protocol is fuzzy. Dustin has example client/servers, but is 
anyone working on this? A lot of us had an "idea" that facebook is 
writing code to this effect, but no one knew for sure.

- Sun's interested in making memcached scale across multicpu machines 
more linearly, but this isn't a huge community interest yet. Not many 
users run out of CPU on their clusters, and fewer run threaded mode at 
all currently.

The biggest confusion here appears to be libevent and it not being very 
thread friendly. Someone from Sun will ramble about more details? :)

- No discussion of consistent hashing algorithms. It slipped my mind :( 
Not many client authors there anyway?

- Docs are good. Please add to the docs! Use case scenarios are very 
useful to people.

Paul Linder actually comitted some code :) Dustin/Sean/etc wrote code.

Personally, I've cooked some ideas in my head that I might try out soon.


More information about the memcached mailing list