On Fri, Jun 01, 2018 at 03:29:45PM +0800, Wei Wang wrote: > On 06/01/2018 01:07 PM, Peter Xu wrote: > > On Fri, Jun 01, 2018 at 12:58:24PM +0800, Peter Xu wrote: > > > On Tue, Apr 24, 2018 at 02:13:43PM +0800, Wei Wang wrote: > > > > This is the deivce part implementation to add a new feature, > > > > VIRTIO_BALLOON_F_FREE_PAGE_HINT to the virtio-balloon device. The device > > > > receives the guest free page hints from the driver and clears the > > > > corresponding bits in the dirty bitmap, so that those free pages are > > > > not transferred by the migration thread to the destination. > > > > > > > > - Test Environment > > > > Host: Intel(R) Xeon(R) CPU E5-2699 v4 @ 2.20GHz > > > > Guest: 8G RAM, 4 vCPU > > > > Migration setup: migrate_set_speed 100G, migrate_set_downtime 2 > > > > second > > > > > > > > - Test Results > > > > - Idle Guest Live Migration Time (results are averaged over 10 > > > > runs): > > > > - Optimization v.s. Legacy = 271ms vs 1769ms --> ~86% reduction > > > > - Guest with Linux Compilation Workload (make bzImage -j4): > > > > - Live Migration Time (average) > > > > Optimization v.s. Legacy = 1265ms v.s. 2634ms --> ~51% > > > > reduction > > > > - Linux Compilation Time > > > > Optimization v.s. Legacy = 4min56s v.s. 5min3s > > > > --> no obvious difference > > > > > > > > - Source Code > > > > - QEMU: https://github.com/wei-w-wang/qemu-free-page-lm.git > > > > - Linux: https://github.com/wei-w-wang/linux-free-page-lm.git > > > Hi, Wei, > > > > > > I have a very high-level question to the series. > > > > > > IIUC the core idea for this series is that we can avoid sending some > > > of the pages if we know that we don't need to send them. I think this > > > is based on the fact that on the destination side all the pages are by > > > default zero after they are malloced. While before this series, IIUC > > > any migration will send every single page to destination, no matter > > > whether it's zeroed or not. So I'm uncertain about whether this will > > > affect the received bitmap on the destination side. Say, before this > > > series, the received bitmap will directly cover the whole RAM bitmap > > > after migration is finished, now it's won't. Will there be any side > > > effect? I don't see obvious issue now, but just raise this question > > > up. > > > > > > Meanwhile, this reminds me about a more funny idea: whether we can > > > just avoid sending the zero pages directly from QEMU's perspective. > > > In other words, can we just do nothing if save_zero_page() detected > > > that the page is zero (I guess the is_zero_range() can be fast too, > > > but I don't know exactly how fast it is)? And how that would be > > > differed from this page hinting way in either performance and other > > > aspects. > > I noticed a problem (after I wrote the above paragraph 5 minutes > > ago...): when a page was valid and sent to the destination (with > > non-zero data), however after a while that page was zeroed. Then if > > we don't send zero pages at all, we won't send the page after it's > > zeroed. Then on the destination side we'll have a stale non-zero > > page. Is my understanding correct? Will that be a problem to this > > series too where a valid page can be possibly freed and hinted? > > I think that won't be an issue either for zero page optimization or this > free page optimization. > > For the zero page optimization, QEMU always sends compressed 0s to the > destination. The zero page is detected at the time QEMU checks it (before > sending the page). if it is a 0 page, QEMU compresses all 0s (actually just > a flag) and send it.
what I meant is, can we just do not even send that ZERO flag at all? :) > > For the free page optimization, we skip free pages (could be thought of as 0 > pages in this context). The zero pages are detected at the time guest > reports it QEMU. The page won't be reported if it is non-zero (i.e. used). Sorry I must have not explained myself well. Let's assume the page hint is used. I meant this: - start precopy, page P is non-zero (let's say, page has content P1, which is non-zero) - we send page P with content P1 on src, then latest destination cache of page P is P1 - page P is freed by the guest, then it becomes zero, dirty bitmap of P is set since it's changed (from P1 to zeroed page) - page P is provided as hint that we can skip it since it's zeroed, then the dirty bit of P is cleared - ... (page P is never used until migration completes) After migration completes, page P should be an zeroed page on the source, while IIUC on the destination side it's still with stale data P1. Did I miss anything important? Thanks, -- Peter Xu