Yes, we saw, the per-database write lock. It's soooo much better.
Come on, guys. They have real NoSQL datastores out there... with more sophisticated strategies than "keep it all in memory and let the operating system swap everything to disk" to boot.
not sure what you're going on about..
what you're characterizing as "punting" to the OS was a carefully thought through decision, along the same vein as the one phk of FreeBSD and Varnish fame made[1]
Varnish is a carefully tuned piece of machinery that consciously leverages the page cache because that happens to be a great fit for the access pattern that you'd expect in a cache.
Mongo is not a cache. It generates quite a different access pattern and it's widely documented what happens when your working-set exceeds RAM or when you put it under anything but the lightest write-load. A single bulk update literally halts the world.
Comparing Varnish to MongoDB is quite an insult, akin to comparing a Swiss precision Rolex to a plastic Mickey Mouse watch "with many advanced secret-agent features" from a gumball machine.
