That makes lot of sense. We used ascii protocol and synch operations. We also noted large variations between different EC2 instances and zones. Amazon is offering a product named ElastiCache which provides protocol compatibility with memcached. We didn't have the chance to test it though, but is something you probably would like to look at.
On Sep 16, 3:07 pm, Dustin <[email protected]> wrote: > On Friday, September 16, 2011 9:20:11 AM UTC-7, UserStupidException wrote: > > > We are using EC2 too. I found that the main problem is not bandwidth > > but latency, currently we are limited to around 50k ops/seg (being 20% > > writes) on the best case. > > If you made any benchmarks regarding memcached on EC2 I would really > > interested in your results. > > That makes sense. If you can pipeline more of your ops together into a > stream you can get more throughput at the same latency by sending more while > waiting. I think I've peaked around 160k ops/s on a fully async write load > with binary protocol on EC2. That seems to be around the limit.
