10

We are considering to use a caching system to our application to do data lookup. This data will change very infrequently. My findings came out with Redis, GemFire/Geode, VoltDB, Aerospike, Hazlecast and few others. I shortlisted Geode and Redis. Both are key-value in-memory stores.

Redis (C based) seems to be easier to use and supports data types such as lists, hashes, sets, sorted sets, etc. It also has some good client available in Java, NodeJS, C#, etc. It supports multiple nodes (master-slave).

Geode (Java based) seems to have more features considering its existence dates a long back. It has concepts of regions in which stores the data. It has clients available in Java, C#, C++. It has locator(s) and server(s) which make the system. It supports distributed servers/nodes (non-shared/no single point of failure).

I searched internet but could not find any comparison between these two in terms of Read/write speed, Memory usage, Disk I/O dumping, Scaling, Security. Has anyone come across such situation before ?

Brij
  • 11,731
  • 22
  • 78
  • 116
  • For those evaluating the solutions listed above, Apache Ignite is also worth looking at. Some key differentiators with GemFire/Geode are SQL and ACID transactions. It seems to have slightly higher adoption (support) and personally I think the docs are easier to read. https://ignite.apache.org/ – jocull Oct 12 '22 at 14:23

1 Answers1

12

I'm working primarily with Redis but got to know Geode (Open-Source variant of Pivotal Gemfire) throughout the last year as I work for Pivotal. Might be biased, though.

Redis is primarily a standalone Key-Value store, driven by Salvatore Sanfilippo, HA and sharding options (Sentinel, Cluster) were built at a much later time. Redis is blazing fast, and if you favor performance over consistency, Redis is a good choice. Redis only security feature is password-based authentication, but you would rather secure it on transport-level which requires additional effort. There's no support besides the community or if you opt for a commercial Redis product (such as RedisLabs).

Geode (Open-Sourced Pivotal GemFire about a year ago, nowadays GemFire is based on Geode) is a product driven by Pivotal for many years. It's built with consistency, HA, and data distribution in mind – call it an in-memory data grid. The security layer of Geode is extensible and pluggable so you can tailor it to your needs. If you want to get commercial support, then Pivotal can serve you with your needs.

Geode features a Redis protocol adapter which allows you to connect with Redis clients to Geode and to use a subset of Redis data structures and commands. It's not complete from a Redis 3.2 perspective but still a good choice if you want to integrate with a single key-value store product, and you don't want to use Redis and Geode.

mp911de
  • 17,546
  • 2
  • 55
  • 95
  • you mentioned `Redis is blazing fast, and if you favor performance over consistency, Redis is a good choice.` What consistency issues are there in Redis ? – Brij Jul 04 '17 at 14:05
  • 1
    Some commands (`SCAN`, `HSCAN`, …) have weak consistency guarantees if keys have changed between iterations. In HA scenarios you can lose data (see https://aphyr.com/posts/307-jepsen-redis-redux). Persistence is handled in the background and with keeping performance high you might lose up to 1 second of writes - alternatively, you can persist the data on each write which dramatically impacts performance. – mp911de Jul 04 '17 at 14:10
  • 1
    In redis, replication is also asynchronous by default. Even with synchronous replication the documentation says you can loose data: https://redis.io/topics/replication – Swapnil Jul 05 '17 at 18:12
  • Here's very deep dive into Redis consistency issues. Some of them are fundamental problems that don't have workarounds. https://towardsdatascience.com/redis-unsafe-at-any-speed-f2731f738a25 – jocull Oct 12 '22 at 14:20