Startup Architecture: Scale up LAMP architecture using HAProxy, PHP, Redis, Memcache and MySQL to Handle 1 Billion Requests A Week!!
Now mostly startup companies are using LAMP architecture to build product. We can create quite simple architecture based on HAProxy, PHP, Redis, Memcachce and MySQL that seamlessly handles approx 1 billion requests every week .
Stats:
-
Servers:
-
3x application nodes
-
2x MySQL + 1x for backup
- 2x memcache nodes
-
2x Redis
-
-
Application:
-
Application handles 1,000,000,000 requests every week
-
Average response time - 30 milliseconds
-
Varnish - more than 12,000 req/s (achieved during stress test)
-
-
Data store:
-
Redis - 160,000,000 records, 100 GB of data (our primary data store!),
-
MySQL - 300,000,000 records - 300 GB (third cache layer)
-
- Logical Architecture:
- Application Architecture:
Scalability
Database is always the hardest bottleneck of an application. Currently, there weren’t needed any scaling-out operations - to this time we’re scaling vertically by moving our Redis and MySQL to bigger boxes. There’s still a space for it as e.g. Redis is running on a server with 128 GB memory - it’s possible to migrate them to nodes with 256 GB. Of course such heavy boxes also come with disadvantages in operations like snapshot or just running up the server - it’ll take much longer to start up Redis server.
After scaling-up (vertically), goes scaling-out (horizontally). Happily, we’ve prepared easy to shard structure of our data:
We’ve got 4 “heavy” types of records in Redis. They can be sharded on 4 servers based on data type. We’d avoid partitioning based on hashing in favor to dividing the data by a records type. That way we’ll be still able to run MGET which is always performed on an one type of keys.
In a MySQL, tables are structured the way, that let easy migration of some of them to the different server - also basing on the records type (tables).
After exceeding possibility of further partitioning by data types, we’ll go into hashing :-)
