Re: [Xen-devel] Implementing working set tracker in xen

[Tim Deegan <tim@xxxxxxx>]

Hi,

At 19:56 +0100 on 18 Mar (1426708593), HANNAS YAYA Issa wrote:
> ESX Server uses a statistical sampling approach to ob-
> tain aggregate VM working set estimates directly, with-
> out any guest involvement. Each VM is sampled inde-
> pendently, using a configurable sampling period defined
> in units of VM execution time. At the start of each sam-
> pling period, a small number of the virtual machine?s
> ?physical? pages are selected randomly using a uniform
> distribution. Each sampled page is tracked by invalidat-
> ing any cached mappings associated with its PPN, such
> as hardware TLB entries and virtualized MMU state.
> The next guest access to a sampled page will be inter-
> cepted to re-establish these mappings, at which time a
> touched page count is incremented. At the end of the
> sampling period, a statistical estimate of the fraction
> of memory actively accessed by the VM is f = t/n.
> 
> The algorithm is easy to understand. but I have a real problem while 
> implementing it.
> I don't how I can invalidate all cache mapping with the select physical 
> number. it's simple to invalidate tlb mapping by flushing the tlb
> but I cannot invalidate page table entries that map the selected 
> physical number because I cannot get the virtual address of the physical 
> number.
> I have the virtual addresse of page table entries only during the 
> hypercall HYPERVISOR_mmu_update. but this give only the address of the 
> pages to be updated.
> 
> Please can someone help me to resole this issue.

Have a look at the log_dirty/mark_dirty mechanism: that does exactly
this, but to _every_ frame of guest memory, for tracking which ones
have changed during migrastion.

There's already a secondary version of this for tracking only the
framebuffer (look for 'dirty_vram') so that changes can be picked up
in a GUI.  It sounds like what yuop want to do could be another
variant of this, where only a random sample of pages are tracked,
and you don't need to keep the bitmap of which pages have changed.

Cheers,

Tim.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel