Normally when the entire DB is in RAM, an announce takes ~0.2-0.3ms. (First third of graph)
Usually in high traffic times we observe redis swapped out to disk, I/O taking all the time and announces taking 10000ms+ as a result (Middle part of graph)
But now we are in this unique situation where some of redis is swapped out, but not all, so announces are taking ~50ms as a result.
So if we can intentionally swap out some redis to disk (but not all!), we can get pretty decent performance?!
Normally when the entire DB is in RAM, an announce takes ~0.2-0.3ms. (First third of graph)
Usually in high traffic times we observe redis swapped out to disk, I/O taking all the time and announces taking 10000ms+ as a result (Middle part of graph)
But now we are in this unique situation where some of redis is swapped out, but not all, so announces are taking ~50ms as a result.
So if we can intentionally swap out some redis to disk (but not all!), we can get pretty decent performance?!