涉及SQL时，使用Amazon EC2自动扩展

Question

I'm building a whiteboard web app with self-contained "rooms" of clients that runs off Amazon EC2 instances (a single one for now). Commands are sent via websockets to a PHP server, which stores all commands in a SQL database.

Up until now I was using Google Cloud SQL. My plan was to learn how to scale with EC2 and have all instances use the same remote database. I've learned this won't work due to the 200 ms write latency of a remote SQL server vs. the 0.5 ms write latency of a local SQL server. The server makes a write every time a command arrives.

I'm new to scalability and distributed systems. My intuition tells me I either need to use Amazon RDS and hope for millisecond latencies if my EC2 and RDS instances are in the same region, or work with SQL locally on EC2 instances. I'm leaning toward the latter. Here's my issue: EC2 is elastic . What happens when I need to get rid of an instance?

All I can think of right now is somehow replicating the SQL data from each EC2 instance to a master instance (maybe even Google Cloud SQL!). In other words, all reads/writes for each "room" happen locally, and are eventually replicated to the master server for long-term storage. If a "room" is re-opened a week later, a different EC2 instance can grab data from the master server, work with it locally, and replicate changes back before being destroyed.

Does my approach sound correct--is replication the right concept here? If so, how much support for what I'm trying to do already exists? That is, do I need to set up a master server that manages EC2 instances and distributes/collects the SQL data manually (100% custom implementation), or is there are there existing libraries/mechanisms for SQL and maybe even EC2 instance replication/management? And if my approach is wrong, what are some better approaches? This is one of those times where I don't know what to research on my own. Thanks!

Answer 1

我同意user02525或许看看使用Elasticache redis，听起来更符合你正在做的事情。