I'm building a system with autoscaling servers and need a distributed cache

A distributed cache is a good fit for autoscaling servers because it lets any instance read/write shared data without depending on local memory.

What to look for

Key requirements usually are:

• Low latency for reads/writes
• Horizontal scalability
• High availability
• TTL / expiration support
• Atomic operations if you need counters, locks, rate limiting
• Eviction policies when memory fills up
• Replication / failover so cache loss doesn’t break the system

Common options

1. Redis

Best default choice for many systems.

• Very fast
• Widely supported
• Supports clustering, replication, persistence, TTL, pub/sub, streams
• Good for sessions, rate limiting, leaderboards, queues, feature flags

Use if you want a mature, general-purpose distributed cache.

2. Memcached

Simple and fast.

• Great for basic caching
• Easy to use
• No persistence, fewer features than Redis

Use if you only need pure ephemeral caching and want simplicity.

3. Managed cloud caches

If you’re on cloud, consider managed services:

• AWS ElastiCache for Redis/Memcached
• Azure Cache for Redis
• Google Cloud Memorystore

These reduce operational overhead and work well with autoscaling apps.

Architecture suggestion

For autoscaling servers:

• Keep app servers stateless
• Store shared session/state in Redis or Memcached
• Use cache-aside pattern:
  1. App checks cache
  2. If miss, fetch from DB
  3. Populate cache with TTL
• Set sensible TTLs and invalidation rules

Recommended starting point

If you’re unsure: use Redis (managed if possible).

It gives you:

• Cache
• Session store
• Distributed locks
• Counters
• Message/stream primitives

in one system.

Important caveats

• Don’t treat cache as primary storage
• Plan for cache misses and cache eviction
• Avoid very large objects
• Use consistent key naming and TTLs
• Watch for stampedes; consider request coalescing or locking

If you want, I can also help you choose between Redis vs Memcached, or design a cache strategy for your specific autoscaling setup.