Redis optimal hash set entry size

Question

I have some questions regarding the optimal entry size setting for Redis hash sets.

1. In this example memory-optimization they use 100 hash entries per key but use hash-max-zipmap-entries 256 ? Why not hash-max-zipmap-entries 100 or 128?

2. On the redis website (above link) they used max hash entry size of 100, but in this post instagram, they mention 1000 entries. So does this mean the optimal setting is a function of the product of hash-max-zipmap-entries & hash-max-zipmap-value ?(ie in this case Instagram has smaller hash-values than memory optimization example?)

Your comments/clarifications are much appreciated.

Answer 1

The key is, from here:

manipulating the compact versions of these [ziplist] structures can become slow as they grow longer

and

[as ziplists grow longer] fetching/updating individual fields of a HASH, Redis will have to decode many individual entries, and CPU caches won’t be as effective

So to your questions

1. This page just shows an example and I doubt the author gave much thought to the exact values. In real life, IF you wanted to take advantage of ziplists, and you knew your number of entries per hash was <100, then setting it at 100, 128 or 256 would make no difference. hash-max-zipmap-entries is only the LIMIT over which you're telling Redis to change the encoding from ziplist to hash.

2. There may be some truth in your "product of hash-max-zipmap-entries & hash-max-zipmap-value" idea, but I'm speculating. More importantly, first you have to define "optimal" based on what you want to do. If you want to do lots of HSET/HGETs in a large ziplist, it will be slower than if you used a hash. But if you never get/update single fields only ever do HMSET/HGETALL on a key, large ziplists wouldn't slow you down. The Instagram 1000 was THEIR optimal number based on THEIR specific data, use cases, and Redis function call frequencies.