[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH] xenpaging:add a new array to speed up page-in in xenpaging

Sorry, this mail is sent out in hurry. In fact we send out 2 patches , and this 
is only one patch. The second patch is the key.

The second patch is about adding a fast way to trigger page-in. 
Titile is [PATCH 2 of 2] xenpaging:change page-in process to speed up page-in 
in xenpaging.


In this patch, we get the return value from p2m_mem_paging_populate(). 
Because we want to known whether this p2mt is changed by balloon or others.

After this patch is published, we have received much advice, so a more perfect 
patch will be sent out later 

> -----Original Message-----
> From: Olaf Hering [mailto:olaf@xxxxxxxxx]
> Sent: Wednesday, January 11, 2012 1:40 AM
> To: Hongkaixing
> Cc: 'Ian Jackson'; xen-devel@xxxxxxxxxxxxxxxxxxx; 'bicky.'
> Subject: Re: [Xen-devel] [PATCH] xenpaging:add a new array to speed up 
> page-in in xenpaging
> On Fri, Jan 06, Hongkaixing wrote:
> > > Why wrap this up in a struct ?
> >
> > We want to keep the same style with
> >
> > typedef struct xenpaging_victim {
> >     /* the gfn of the page to evict */
> >     unsigned long gfn;
> > } xenpaging_victim_t;
> This dates back to the initial implementation of xenpaging.
> In my testing I started a guest paused, paged it all out, and paged all
> back into memory. By monitoring nr_pages I noticed that page-in got
> slightly slower over time, so your suggestion to use two indexes will
> speed things up a bit.
> Looking through xenpaging.c, I think its best to have two flat arrays:
> unsigned long slot_to_gfn[paging->max_pages]; /* was victims */
> int gfn_to_slot[paging->max_pages];
> I will prepare a patch to implement this.
> Olaf

Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.