Ken Wei's Library tagged KISS → View Popular

05 Mar 09

Keeping working set in memory is not the only reason for sharding but one of the most frequent ones. The examples I like to use is YouTube - they did not shard until after Google bought them (though they were in pain) and 37Signals
For me simple is best. Complex architectures are more error prone harder to maintain (upgrades etc) and troubleshoot.
5 more annotations...
- Unless you’re Google scale with failures happening daily you can’t really be sure you’re handling “wild” failures, not the test ones well.
- Google guys tells us single MySQL server on a good hardware has MTBF somewhere between 1000 and 2000 days.
- So am I denying all MySQL industry practices (which we also covered in a great depth in our book) ? Not really. I’m just suggesting do not just grab advice from the Internet or friends tip and do not complicate beyond the need.
- It may be boring but boring systems often have highest uptime
- where startups spend lots of time building systems that will scale to the billions of records they’ll never have to store.