We are using the new Amazon ec2 (h1.4xlarge) SSD boosted Machines to test and maybe production this week.


The copy velocity and the full resync is amazing. Renaming fields went a lot faster, but not as much as we´d like.

Map-reduce of about 20 million documents took less than 10 minutes. 

Renaming a field on 10 Million documents took 20 minutes. Renaming on 320 Million, however, took more than 19 hours....

Indexing is still a problem because mongodb uses only one core to do everything and can take as long as a common machine with attached EBS.

That said, in real life situations the %lock on the DB goes down to half of the normal and the background flush avg is less than 200 ms.  I think that on the lung run it will be great once the paging starts to push the limits on our RAM.

I´m eager to put it into production to see how it goes. They are already in production since last week.
Our app have never been faster, the paging are about 50/sec, but are almost unnoticed.  We are expecting to reach 10 MM impressions this Sunday!

A few gotchas: we cannot save them as part of an AMI, so everytime you have to rebuild it, you have to make it with a full resync. And as of today Amazon only allows you to have two of those babies, so if you want more, you have to ask them...


Here´s how it compares in terms of %lock:



And here´s the SWEET IOP measures at the primary:


This weekend, I´ll put the newer iop provisioned EBS (http://aws.typepad.com/aws/2012/08/fast-forward-provisioned-iops-ebs.html)  to work on a secondary, as the older ones are suffering at 50% lock and they could get into a state where they will not be able to catch up the primary.