some technical notes

WolfWings — Sat, 15 Feb 2025 18:15:19 +0000

Honestly the 'sidecar' proxy setup just makes development versus production so trivial as well. Run a localhost memcache instance in dev, but in production that's a proxy with the actual memcache cluster configuration.

Memories!

bradfitz — Sat, 15 Feb 2025 16:10:14 +0000

Whoa, blast from the past!

This makes me feel old. :)

I'm glad the project is still kicking!

some technical notes

dormando — Sat, 15 Feb 2025 00:38:33 +0000

Latest release as of today is 1.6.36, actually :) A bunch of proxy perf. improvements went into 1.6.35!

Some subtle technical notes to add to what's in this high level article:

- Redis is a different thing. All of MC's commands run in a _clustered context_. If anyone has had to scale one redis instance from one to many instances, you had to restrict operations on multiple keys, as those keys will now be spread across different instances. A big reason why MC is "reliable" is because its commands both cluster and run as close to O(1) as possible.

- The article seems dubious on extstore: "While not default Memcached behavior, using storage might be useful in specific situations, such as when a cache stores enormous data values." - We built the storage as hybrid RAM/disk structure, so many operations that normally cause a disk access don't (overwrite/delete/TTL updates/eviction/etc). Disk reads are always 1 disk op, since there is no tree to walk.

If you have a cluster of 3 servers with 20G of memory each, and wish to improve your hit ratio by adding more cache: you could add cache space by giving it 20G of disk. Disk gets used for cold data greater than a few hundred bytes in size. This can lead to huge cost savings without needing to have "enormous data values".

Extstore has become critical for orgs that adopt it. Large or small data.

LWN: Comments on "The evolution of Memcached"

some technical notes

Memories!

some technical notes