Hacker News Reader

Comment by lelf | Hacker News Reader

InnoDB API on top of memcached. Read-only. On the same “biggest HW box we have in the LAB” machine.

over 1,000,000 Query/sec

Cool. Carry on guys.

Huh? First off, the memcached api is talking to InnoDB directly not the other way around. This means that you can dramatically reduce the latency of queries that would previously require you to maintain a bunch of cache and the complexity that comes with it and you would incur the overhead of going to both the memcached server and the database on a cache miss(For complicated queries the you might have to continually hit many cache servers and then the database multiple times, this means only a single connection is needed even in the worsst case that the data isn't in RAM). This also has the advantage that a cache miss doesn't force the SQL layer to come into play just to fetch an item from disk and then have the application stick it into the cache(so your cache is just the page cache in InnoDB; this is much more efficient than application managed memcached).

So yes, this is very cool.

To be clear, it's the memcached API on top of InnoDB - not the other way around. This is cool because it allows fast access to data in InnoDB, while exposing the same underlying data through SQL, as well as for a bunch of other reasons.

> InnoDB API on top of memcached

It's actually the other way round: memcached API on top of InnoDB storage.

Well, the performance under the same conditions was 500KQPS before, so definitely there was an improvement!

Right but if someone showed a similar improvement by adding enough RAM to prevent swapping, would you be impressed? This is what is going on here. They have very much successfully proven that RAM is faster than HDD.

They have proven that SQL-based storage is faster than NOSQL databases.

The poster of the top comment doesn't understand what this topic is about.

This is actually a benchmark of InnoDB table access. You can query it using usual SQL-commands. But if you want faster access, you can use memcache API.

"Read-only" is also misleading here. This is just a benchmark of get command.

"They have proven that SQL-based storage is faster than NOSQL databases."

Too broad of a statement. 'NoSQL' is an umbrella term, and thus likely has several contenders for a raw QPS metric.

The additional step of parsing the query is obviated by statement caching. You're right though, it's simply a benchmark of how fast the query path is to the InnoDB buffer pool. The work is to optimize internal contention.

Very impressive results though.

I thought it was a benchmark of the same hardware with 5.6 vs 5.7. I see now that they changed more than that.

Reply on news.ycombinator.com