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, 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 | 49 ++++++++++++++++++++++++++++++++++++++++++++++++ migration/multifd-colo.h | 26 +++++++++++++++++++++++++ migration/multifd.c | 12 ++++++++++++ migration/multifd.h | 1 + 6 files changed, 90 insertions(+), 1 deletion(-) diff --git a/MAINTAINERS b/MAINTAINERS index 563804345fec68ee72793dbb7c1b7e5be4c32083..dbb217255c2cf35dc0ce971c2021b130fac5469b 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -3837,6 +3837,7 @@ 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 16909d54c5110fc5d8187fd3a68c4a5b08b59ea7..1e59fe4f1f0bbfffed90df38e8f39fa87bceb9b9 100644 --- a/migration/meson.build +++ b/migration/meson.build @@ -40,7 +40,7 @@ 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..d8d98e79b12ed52c41f341052a682d7786e221b5 --- /dev/null +++ b/migration/multifd-colo.c @@ -0,0 +1,49 @@ +/* + * 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) +{ + assert(p->block->colo_cache); + + /* + * 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()) { + colo_record_bitmap(p->block, p->normal, p->normal_num); + } + p->host = p->block->colo_cache; +} + +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()); + } + } + p->host = p->block->host; +} 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.c b/migration/multifd.c index 8e71171fb7a17726ba7eb0705e293c41e8aa32ec..6c85acec3bac134e85cfcee0d32057134f5af8d1 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 "threadinfo.h" #include "options.h" #include "qemu/yank.h" @@ -1269,7 +1270,18 @@ static int multifd_ram_state_recv(MultiFDRecvParams *p, Error **errp) { int ret; + if (migrate_colo()) { + multifd_colo_prepare_recv(p); + } + ret = multifd_recv_state->ops->recv(p, errp); + if (ret != 0) { + return ret; + } + + if (migrate_colo()) { + multifd_colo_process_recv(p); + } return ret; } diff --git a/migration/multifd.h b/migration/multifd.h index 9b6d81e7ede024f05d4cd235de95e73840d0bbc4..7036f438fade1baed2442bfdcf8b5d6397c4a448 100644 --- a/migration/multifd.h +++ b/migration/multifd.h @@ -280,6 +280,7 @@ typedef struct { /* ramblock */ RAMBlock *block; /* ramblock host address */ + /* or points to the corresponding address in the colo cache */ uint8_t *host; /* buffers to recv */ struct iovec *iov; -- 2.39.5
