3 ms·
I skipped a step where you break the frontend servers out into separate machines when you add the memcache. iouring or dpdk on a many core machine might be nee
by hedora 5y ago
I skipped a step where you break the frontend servers out into separate machines when you add the memcache.
iouring or dpdk on a many core machine might be needed to get to 1M TPS when scaling up. (HTTP would probably also be non-trivial / impossible)
Anyway, with connection pooling / sharing, a database server should be able to hit it without those tricks.