We have encountered some issues with swapping when all 128G is filled (we do this a lot); the whole system can hang up completely, this may be a disk controller issue.

The big advantage of the Dell is that it will take 128G in 4G sticks. Very few other machines will do this currently. 8G sticks more than double the total cost. I believe it is based on a chassis that Intel produce.

If there was a machine that took 256G in 4G sticks I would buy one tomorrow.

Regarding Scale Out – true though Scale Out should allow you to pick the optimal size of the box. For most application this optimal size is pretty large because there is operations/monitoring etc overhead involved with each not to mention space/power concerns which are the often the reason for consolidation.

It on large scale app I would rather have 10 128GB boxes than 50 of 16GB boxes to deal with

The only thing we’ve found is that it really will only use 2 – 3 of those 8 cores.. this is probably because of innodb’s threading issues. However.. having that much RAM means *zero* IO bound load. And even when we do need to hit the disks, having that much write cache means that the system never shows more than about 2% “io wait” (and thats 2% of a possible 800%…). If it weren’t for next-key locking in big selects, the system would appear nearly transparent.

Its definitely not ideal. Apps should be designed to scale out. However, in this case, being able to quickly insert many thousands of records per second into one DB, and then have them fan out (via various methods) to the parts of the application that must scale out, means the critical parts of the app can be simpler and (theoretically) have less bugs.

Sure. In world of commodity if Dell has systems everybody has similar systems too. I note Dell in particular because we deal a lot with Dell hardware and the fact they tend to ship relatively mature platforms. My experience certain new systems from SuperMicro and Tyan was not always positive.