Sina weibo (like twitter) use redis to do a lot of things

[redis]

Sina weibo (like twitter) use redis to do a lot of things

[floydlin]

FYI, we have tens of billions counters to store, so we want more memory efficient

[pietern]

Thanks, added to the list. As suggested on the mailing list, you can use hashes if you want to use the atomic increment operation (HINCRBY) and have space efficiency at the same time. You should group multiple counters under a single hash key, and take a look at the zipmap configuration settings in redis.conf on how to optimally use them. This reduces memory usage a lot.

[floydlin]

but we have mget request, which is not support by hashes by now