Supported swap space increased.
The swap/paging code is now able to configure up to 1TB x 4 = 4TB of swap,
up from 8G x 4 = 32G of swap. I have tested 256G of configured swap on
a 64-bit kernel.
However, because the kernel must allocate tracking structures for each
block of paged-out swap the practical limit is going to be somewhat
lower. Approximately 1MB of kernel memory must be wired for every
1GB of swap which is in-use. So, for example, supporting 64G of in-use
swap requires 64M of wired physical memory. Just having that much
swap will not eat much memory, its using it that eats memory.
On i386 about 32M of KVM is reserved by default, giving us a practical
limit of 32G of swap.
On 64-bit about 512M of KVM is reserved by default, giving us a
practical limit of 512G of swap. Of course, actually using that much
will wire 512M of system ram so don't get carried away :-)
The KVM allocation can be changed by setting kern.maxswzone in
/boot/loader.conf, e.g.
kern.maxswzone=64m
Note that kernel virtual memory (KVM) is limited on 32 bit kernels
and you really can't specify a value much larger then 64m. 64-bit
kernels have a 512G KVM space and do not have that problem, hence
the higher default reservation.
There are some intriguing possibilities with the use of SSD-backed
swap space. I will be looking into how such a space might be used
for caching in the cluster work.
-Matt
Matthew Dillon
<dillon@backplane.com>