Twemcache Mentor email addresses

81 views
Skip to first unread message

Vinay K

unread,
Apr 19, 2013, 1:54:36 AM4/19/13
to twitte...@googlegroups.com
Hi,

Can someone please share Twemcache project mentors' email addresses. I'm interested in working on Lock revamp project, and I would like to discuss with them about the project.

Thanks,
Vinay

Vinay K

unread,
Apr 23, 2013, 2:33:06 AM4/23/13
to twitte...@googlegroups.com
Since I did not get any responses on my previous post, I'm posting my queries here:

i) From the description of Lock revamp project, I understand that the locking design will have to be changed to reduce locking contention. Does this mean that this will involve development of a new customized locking library or mechanism, using basic constructs such as futexes, which can be used along with the pthread libraries? Or is it fine if we identify parts of the code which may not necessarily need locks (not sure if there are any), and move them into non-critical section part of the code or may be even incorporate thread local data (to get around this problem)?
ii) I haven't gone through the entire code yet, but I did see linked lists being used in a few places. Does the implementation of a lock free linked lists sound feasible for the project or is it too complicated?

Any responses to my queries would be much appreciated.

Thanks!

Yao Yue

unread,
Apr 23, 2013, 6:34:10 PM4/23/13
to twitte...@googlegroups.com
Please take a look at https://github.com/twitter/twemcache/wiki/GSoC-Information

Anything is possible- a new design that uses pthread's locking primitives, a lockless design with cas support from the hardware, etc. There is no right answer here, we'll need to find out with prototyping and measurement.

-Yao

Vinay K

unread,
Apr 24, 2013, 3:29:32 AM4/24/13
to twitte...@googlegroups.com
Thanks for the link!

Vinay K

unread,
May 1, 2013, 3:15:05 AM5/1/13
to twitte...@googlegroups.com
Hi,

I went thru the project wiki link, and I might need clarifications on the following points:

i) The investigation talks only about GET requests, which would be more like read requests on the KV store. What is the common use-case, do we a lot of reads, or would there be a lot of writes/updates, or an almost equal number of reads and writes on the store?
ii) Also, the wiki link talks about breaking the global lock on the hash table and making it lock free. But the tests that were run,  were mostly GET requests, and it does not clearly state if the write locks have been done with too. 

Thanks!
Message has been deleted

Chris Aniszczyk (zx)

unread,
May 2, 2013, 10:09:45 AM5/2/13
to twitte...@googlegroups.com
On Wednesday, May 1, 2013 2:15:05 AM UTC-5, Vinay K wrote:
Hi,

I went thru the project wiki link, and I might need clarifications on the following points:

i) The investigation talks only about GET requests, which would be more like read requests on the KV store. What is the common use-case, do we a lot of reads, or would there be a lot of writes/updates, or an almost equal number of reads and writes on the store?
ii) Also, the wiki link talks about breaking the global lock on the hash table and making it lock free. But the tests that were run,  were mostly GET requests, and it does not clearly state if the write locks have been done with too. 

I would followup with y...@twitter.com directly who is the mentor. 
Reply all
Reply to author
Forward
0 new messages