Lukas Straub <[email protected]> writes: > Like in the normal ram_load() path, put the received pages into the > colo cache and mark the pages in the bitmap so that they will be > flushed to the guest later. >
> Multifd with COLO is useful to reduce the VM pause time during checkpointing > for latency sensitive workloads. In such workloads the worst-case latency > is especially important. > > Also, this is already worth it for the precopy phase as it helps with > converging. Moreover, multifd migration is the preferred way to do migration > nowadays and this allows to use multifd compression with COLO. > > Benchmark: > Cluster nodes > - Intel Xenon E5-2630 v3 > - 48Gb RAM > - 10G Ethernet > Guest > - Windows Server 2016 > - 6Gb RAM > - 4 cores > Workload > - Upload a file to the guest with SMB to simulate moderate > memory dirtying > - Measure the memory transfer time portion of each checkpoint > - 600ms COLO checkpoint interval > > Results > Plain > idle mean: 4.50ms 99per: 10.33ms > load mean: 24.30ms 99per: 78.05ms > Multifd-4 > idle mean: 6.48ms 99per: 10.41ms > load mean: 14.12ms 99per: 31.27ms > > Evaluation > While multifd has slightly higher latency when the guest idles, it is > 10ms faster under load and more importantly it's worst case latency is > less than 1/2 of plain under load as can be seen in the 99. Percentile. > > Signed-off-by: Juan Quintela <[email protected]> > Signed-off-by: Lukas Straub <[email protected]> > --- > MAINTAINERS | 1 + > migration/meson.build | 2 +- > migration/multifd-colo.c | 50 > ++++++++++++++++++++++++++++++++++++++++++++++ > migration/multifd-colo.h | 26 ++++++++++++++++++++++++ > migration/multifd-nocomp.c | 10 +++++++++- > migration/multifd.c | 8 ++++++++ > migration/multifd.h | 5 ++++- > 7 files changed, 99 insertions(+), 3 deletions(-) > > diff --git a/MAINTAINERS b/MAINTAINERS > index > 1e9bdd87c3a2f84f3abfc56986cd793976810fdd..883f0a8f4eb92d0bf0f89fcab4674ccc4aed1cc1 > 100644 > --- a/MAINTAINERS > +++ b/MAINTAINERS > @@ -3853,6 +3853,7 @@ COLO Framework > M: Lukas Straub <[email protected]> > S: Maintained > F: migration/colo* > +F: migration/multifd-colo.* > F: include/migration/colo.h > F: include/migration/failover.h > F: docs/COLO-FT.txt > diff --git a/migration/meson.build b/migration/meson.build > index > c7f39bdb55239ecb0e775c77b90a1aa9e6a4a9ce..c9f0f5f9f2137536497e53e960ce70654ad1b394 > 100644 > --- a/migration/meson.build > +++ b/migration/meson.build > @@ -39,7 +39,7 @@ system_ss.add(files( > ), gnutls, zlib) > > if get_option('replication').allowed() > - system_ss.add(files('colo-failover.c', 'colo.c')) > + system_ss.add(files('colo-failover.c', 'colo.c', 'multifd-colo.c')) > else > system_ss.add(files('colo-stubs.c')) > endif > diff --git a/migration/multifd-colo.c b/migration/multifd-colo.c > new file mode 100644 > index > 0000000000000000000000000000000000000000..c47f5044663969e0c9af56da5ec34902d635810a > --- /dev/null > +++ b/migration/multifd-colo.c > @@ -0,0 +1,50 @@ > +/* > + * SPDX-License-Identifier: GPL-2.0-or-later > + * > + * multifd colo implementation > + * > + * Copyright (c) Lukas Straub <[email protected]> > + * > + * This work is licensed under the terms of the GNU GPL, version 2 or later. > + * See the COPYING file in the top-level directory. > + */ > + > +#include "qemu/osdep.h" > +#include "exec/target_page.h" > +#include "qemu/error-report.h" > +#include "qapi/error.h" > +#include "ram.h" > +#include "multifd.h" > +#include "options.h" > +#include "io/channel-socket.h" > +#include "migration/colo.h" > +#include "multifd-colo.h" > +#include "system/ramblock.h" > + > +void multifd_colo_prepare_recv(MultiFDRecvParams *p) > +{ > + /* > + * While we're still in precopy state (not yet in colo state), we copy > + * received pages to both guest and cache. No need to set dirty bits, > + * since guest and cache memory are in sync. > + */ > + if (migration_incoming_in_colo_state()) { What's the relationship between migration_incoming_in_colo_state() and migration_incoming_colo_enabled()? ram_load_precopy() checks both. Would migration_incoming_colo_enabled affect multifd as well? The multifd recv threads will be running until after process_incoming_migration_bh(), which is when migration_incoming_disable_colo() runs. Also, is the colo_cache guaranteed to be around until multifd threads exit? > + colo_record_bitmap(p->block, p->normal, p->normal_num); > + colo_record_bitmap(p->block, p->zero, p->zero_num); > + } > +} > + > +void multifd_colo_process_recv(MultiFDRecvParams *p) > +{ > + if (!migration_incoming_in_colo_state()) { > + for (int i = 0; i < p->normal_num; i++) { > + void *guest = p->block->host + p->normal[i]; > + void *cache = p->host + p->normal[i]; > + memcpy(guest, cache, multifd_ram_page_size()); > + } I see some differences between what ram.c does and what multifd will do after this patch regarding which flags are checked and order of copies (code below): ram.c: - migration_incoming_colo_enabled && migration_incoming_in_colo_state: Reads from stream into colo_cache. - migration_incoming_colo_enabled && !migration_incoming_in_colo_state: Reads from stream into guest and then memcpy into colo_cache. - !migration_incoming_colo_enabled Reads from stream into guest. multifd.c: - migrate_colo: Reads from stream into colo_cache. - !migration_incoming_in_colo_state: memcpy from colo_cache into guest. - !migration_incoming_colo_enabled ??? The resulting state should be the same, but I wonder if we want to i) use the same checks in multifd and ii) when not in colo state, copy first into guest (using readv) and later memcpy into the colo_cache. --- ram.c: host = host_from_ram_block_offset(block, addr); if (migration_incoming_colo_enabled()) { if (migration_incoming_in_colo_state()) { host = colo_cache_from_block_offset(block, addr, true); } else { host_bak = colo_cache_from_block_offset(block, addr, false); } } qemu_get_buffer(f, host, TARGET_PAGE_SIZE); if (host_bak) { memcpy(host_bak, host, TARGET_PAGE_SIZE); } multifd: if (migrate_colo()) { p->host = p->block->colo_cache; } for (int i = 0; i < p->normal_num; i++) { p->iov[i].iov_base = p->host + p->normal[i]; } return qio_channel_readv_all(p->c, p->iov, p->normal_num, errp); if (!migration_incoming_in_colo_state()) { for (int i = 0; i < p->normal_num; i++) { void *guest = p->block->host + p->normal[i]; void *cache = p->host + p->normal[i]; memcpy(guest, cache, multifd_ram_page_size()); } } --- > + for (int i = 0; i < p->zero_num; i++) { > + void *guest = p->block->host + p->zero[i]; > + memset(guest, 0, multifd_ram_page_size()); > + } At multifd_nocomp_recv, there will be a call to multifd_recv_zero_page_process(), which by that point will have p->host == p->block->colo_cache, so it looks like that function will do some zero page processing in the colo_cache, setting the rb->receivedmap for pages in the colo_cache and potentially also doing a memcpy. Is this intended? I'm thinking that maybe it would overall be better to hook colo directly in to multifd_nocomp_recv: static int multifd_nocomp_recv(MultiFDRecvParams *p, Error **errp) { uint32_t flags; if (migrate_mapped_ram()) { return multifd_file_recv_data(p, errp); } flags = p->flags & MULTIFD_FLAG_COMPRESSION_MASK; if (flags != MULTIFD_FLAG_NOCOMP) { error_setg(errp, "multifd %u: flags received %x flags expected %x", p->id, flags, MULTIFD_FLAG_NOCOMP); return -1; } + if (migration_incoming_colo_enabled() && migration_incoming_in_colo_state()) { + p->host = p->block->colo_cache; + } // or else{}, depending on how deal with zero pages in the cache multifd_recv_zero_page_process(p); if (!p->normal_num) { return 0; } for (int i = 0; i < p->normal_num; i++) { p->iov[i].iov_base = p->host + p->normal[i]; p->iov[i].iov_len = multifd_ram_page_size(); ramblock_recv_bitmap_set_offset(p->block, p->normal[i]); } + ret = qio_channel_readv_all(p->c, p->iov, p->normal_num, errp); + if (ret != 0) { + return ret; + } + + if (migration_incoming_colo_enabled()) { + multifd_colo_process_recv(); + } return ret; } > + } > +} > diff --git a/migration/multifd-colo.h b/migration/multifd-colo.h > new file mode 100644 > index > 0000000000000000000000000000000000000000..82eaf3f48c47de2f090f9de52f9d57a337d4754a > --- /dev/null > +++ b/migration/multifd-colo.h > @@ -0,0 +1,26 @@ > +/* > + * SPDX-License-Identifier: GPL-2.0-or-later > + * > + * multifd colo header > + * > + * Copyright (c) Lukas Straub <[email protected]> > + * > + * This work is licensed under the terms of the GNU GPL, version 2 or later. > + * See the COPYING file in the top-level directory. > + */ > + > +#ifndef QEMU_MIGRATION_MULTIFD_COLO_H > +#define QEMU_MIGRATION_MULTIFD_COLO_H > + > +#ifdef CONFIG_REPLICATION > + > +void multifd_colo_prepare_recv(MultiFDRecvParams *p); > +void multifd_colo_process_recv(MultiFDRecvParams *p); > + > +#else > + > +static inline void multifd_colo_prepare_recv(MultiFDRecvParams *p) {} > +static inline void multifd_colo_process_recv(MultiFDRecvParams *p) {} > + > +#endif > +#endif > diff --git a/migration/multifd-nocomp.c b/migration/multifd-nocomp.c > index > 9be79b3b8e00371ebff9e112766c225bec260bf7..9f7a792fa761b3bc30b971b35f464103a61787f0 > 100644 > --- a/migration/multifd-nocomp.c > +++ b/migration/multifd-nocomp.c > @@ -16,6 +16,7 @@ > #include "file.h" > #include "migration-stats.h" > #include "multifd.h" > +#include "multifd-colo.h" > #include "options.h" > #include "migration.h" > #include "qapi/error.h" > @@ -269,7 +270,6 @@ int multifd_ram_unfill_packet(MultiFDRecvParams *p, Error > **errp) > return -1; > } > > - p->host = p->block->host; > for (i = 0; i < p->normal_num; i++) { > uint64_t offset = be64_to_cpu(packet->offset[i]); > > @@ -294,6 +294,14 @@ int multifd_ram_unfill_packet(MultiFDRecvParams *p, > Error **errp) > p->zero[i] = offset; > } > > + if (migrate_colo()) { > + multifd_colo_prepare_recv(p); > + assert(p->block->colo_cache); > + p->host = p->block->colo_cache; Can't you just use p->block->colo_cache later? I don't see why p->host needs to be set beforehand even in the non-colo case. > + } else { > + p->host = p->block->host; > + } > + > return 0; > } > > diff --git a/migration/multifd.c b/migration/multifd.c > index > 332e6fc58053462419f3171f6c320ac37648ef7b..220ed8564960fdabc58e4baa069dd252c8ad293c > 100644 > --- a/migration/multifd.c > +++ b/migration/multifd.c > @@ -29,6 +29,7 @@ > #include "qemu-file.h" > #include "trace.h" > #include "multifd.h" > +#include "multifd-colo.h" > #include "options.h" > #include "qemu/yank.h" > #include "io/channel-file.h" > @@ -1258,6 +1259,13 @@ static int multifd_ram_state_recv(MultiFDRecvParams > *p, Error **errp) > int ret; > > ret = multifd_recv_state->ops->recv(p, errp); > + if (ret != 0) { > + return ret; > + } > + > + if (migrate_colo()) { > + multifd_colo_process_recv(p); > + } Either put all of colo hooks in multifd.c or multifd-nocomp.c. I think the latter is more appropriate as we have mapped_ram already in there. Let's drop patch 3 and put this in multifd_nocomp_recv(). > > return ret; > } > diff --git a/migration/multifd.h b/migration/multifd.h > index > 89a395aef2b09a6762c45b5361e0ab63256feff6..fbc35702b062fdc3213ce92baed35994f5967c2b > 100644 > --- a/migration/multifd.h > +++ b/migration/multifd.h > @@ -279,7 +279,10 @@ typedef struct { > uint64_t packets_recved; > /* ramblock */ > RAMBlock *block; > - /* ramblock host address */ > + /* > + * Normally, it points to ramblock's host address. When COLO > + * is enabled, it points to the mirror cache for the ramblock. > + */ > uint8_t *host; > /* buffers to recv */ > struct iovec *iov;
