On Fri, Mar 10, 2017 at 06:02:31PM +0800, Wei Wang wrote: > On 03/08/2017 12:01 PM, Michael S. Tsirkin wrote: > > On Fri, Mar 03, 2017 at 01:40:28PM +0800, Wei Wang wrote: > > > From: Liang Li <liang.z.li@xxxxxxxxx> > > > > > > The implementation of the current virtio-balloon is not very > > > efficient, because the pages are transferred to the host one by one. > > > Here is the breakdown of the time in percentage spent on each > > > step of the balloon inflating process (inflating 7GB of an 8GB > > > idle guest). > > > > > > 1) allocating pages (6.5%) > > > 2) sending PFNs to host (68.3%) > > > 3) address translation (6.1%) > > > 4) madvise (19%) > > > > > > It takes about 4126ms for the inflating process to complete. > > > The above profiling shows that the bottlenecks are stage 2) > > > and stage 4). > > > > > > This patch optimizes step 2) by transfering pages to the host in > > > chunks. A chunk consists of guest physically continuous pages, and > > > it is offered to the host via a base PFN (i.e. the start PFN of > > > those physically continuous pages) and the size (i.e. the total > > > number of the pages). A normal chunk is formated as below: > > > ----------------------------------------------- > > > | Base (52 bit) | Size (12 bit)| > > > ----------------------------------------------- > > > For large size chunks, an extended chunk format is used: > > > ----------------------------------------------- > > > | Base (64 bit) | > > > ----------------------------------------------- > > > ----------------------------------------------- > > > | Size (64 bit) | > > > ----------------------------------------------- > > > > > > By doing so, step 4) can also be optimized by doing address > > > translation and madvise() in chunks rather than page by page. > > > > > > This optimization requires the negotation of a new feature bit, > > > VIRTIO_BALLOON_F_CHUNK_TRANSFER. > > > > > > With this new feature, the above ballooning process takes ~590ms > > > resulting in an improvement of ~85%. > > > > > > TODO: optimize stage 1) by allocating/freeing a chunk of pages > > > instead of a single page each time. > > > > > > Signed-off-by: Liang Li <liang.z.li@xxxxxxxxx> > > > Signed-off-by: Wei Wang <wei.w.wang@xxxxxxxxx> > > > Suggested-by: Michael S. Tsirkin <mst@xxxxxxxxxx> > > > Cc: Michael S. Tsirkin <mst@xxxxxxxxxx> > > > Cc: Paolo Bonzini <pbonzini@xxxxxxxxxx> > > > Cc: Cornelia Huck <cornelia.huck@xxxxxxxxxx> > > > Cc: Amit Shah <amit.shah@xxxxxxxxxx> > > > Cc: Dave Hansen <dave.hansen@xxxxxxxxx> > > > Cc: Andrea Arcangeli <aarcange@xxxxxxxxxx> > > > Cc: David Hildenbrand <david@xxxxxxxxxx> > > > Cc: Liang Li <liliang324@xxxxxxxxx> > > > Cc: Wei Wang <wei.w.wang@xxxxxxxxx> > > Does this pass sparse? I see some endian-ness issues here. > > "pass sparse"- what does that mean? > I didn't see any complaints from "make" on my machine. Run with make C=1 (or C=2 to check all source). Generally there's a ton of useful info you will find if you run make help. -- MST
