Introduction

Each ramzswap device has its own compression buffer, memory pools and other metadata. All these buffers are protected by a per-device lock which are serious bottlenecks on multi-core machines.

Starting with compcache-0.6, we have support for creating multiple ramzswap devices, primarily to overcome this scalability problem. Divide and Conquer!

Details

If you have multiple cores, then using a single ramzswap can be a bottleneck. To avoid this problem, following configuration is recommended: * Check no. of cores you have and create equal number of ramzswap devices during module load. For quad core machine, you would do: modprobe ramzswap num_devices=4 * Initialize all of these devices. For example above: ```

Assuming you need total disksize of 256MB.

So, per-device size would be 256/4=64MB.

rzscontrol /dev/ramzswap0 -i --disksize_kb=65536 rzscontrol /dev/ramzswap1 -i --disksize_kb=65536 rzscontrol /dev/ramzswap2 -i --disksize_kb=65536 rzscontrol /dev/ramzswap3 -i --disksize_kb=65536 (you can add backing swap disk/files to any/all of above devices. See [rzscontrol manpage](http://compcache.googlecode.com/hg/sub-projects/rzscontrol/man/rzscontrol.html) for more information). * swapon all these devices with the **SAME PRIORITY**. This priority must be higher than any of the disk based swap devices, if present. For example above: swapon /dev/ramzswap0 -p 10 swapon /dev/ramzswap1 -p 10 swapon /dev/ramzswap2 -p 10 swapon /dev/ramzswap3 -p 10 ```

This kind of split setup should give you much better performance on multi-cores than a single giant ramzswap device.