I’m hoping they’ll be clever enough to back port the zram snappy compression patches at some stage

That sounds handy/cool, what is it?

Zram is what’s being used instead of a storage-based swap partition, and I’d suggest is why there is so much less (read none) soft blocking compared to the n900 (to be fair the much larger amount of ram helps a lot too). Instead of putting swapped out memory to a swap partition, it is instead compressed but still kept in memory. It tends to use a whole pile less space (than just being kept in ram), and means that swap operations no longer have any i/o requirement, at the cost of a little more cpu usage.

The snappy compression patches were about using the google snappy compression format (which is more efficient that the LZO algorithm that zram uses) to improve zram performance.
Unfortunately, I since realised that it’s quite x86_64 specific so it’s considered unlikely to be of much benefit on arm :-(
There is apparently though some recent improvements to the LZO algorithm that improve it’s performance greatly which would be great to see adopted on the n9 :-)

Originally Posted by thedead1440

I'm not sure of that but you may want to check this out by rainisto which uses the spare mtd block for swap instead of taking away 256MB of RAM as swap and keeping RAM at 1GB...

Originally Posted by Hurrian

If you want to build a kernel and benchmark, the tiny patches are available here.