Hi All. We’re trying to implement a rate limiting algorithm for accessing our Hybrid RTB platform.
We currently average around 50K QPS and for each of these we fire between 5 and 20 (more in the future) requests to publishers and these are the requests i need to implement throttling on.
Basically what we do is setup counters per publisher, these have a TTL of 1 second and we keep incrementing the counter with each outbound request.
It’s been working fine on low loads but when trying to run this on aerospike in memory on a cluster of 2 16cpu gcp instances it does not hold the load and eventually crashes.
(We are talking around 1 million tps).
I am new to AS and used the out of the box configuration with no performance tweaks.
I guess i’m looking for pointers on how to improve performance and maybe do it in a less resource heavy way as adding more servers gets expensive real fast
I’d appreciate any input.