From patchwork Thu Dec 13 07:57:26 2018 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Xiao Guangrong X-Patchwork-Id: 10728133 Return-Path: Received: from mail.wl.linuxfoundation.org (pdx-wl-mail.web.codeaurora.org [172.30.200.125]) by pdx-korg-patchwork-2.web.codeaurora.org (Postfix) with ESMTP id E053D13AF for ; Thu, 13 Dec 2018 08:00:38 +0000 (UTC) Received: from mail.wl.linuxfoundation.org (localhost [127.0.0.1]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id D2EBD2A0B7 for ; Thu, 13 Dec 2018 08:00:38 +0000 (UTC) Received: by mail.wl.linuxfoundation.org (Postfix, from userid 486) id C72282A1E2; Thu, 13 Dec 2018 08:00:38 +0000 (UTC) X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on pdx-wl-mail.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-7.7 required=2.0 tests=BAYES_00,DKIM_ADSP_CUSTOM_MED, DKIM_INVALID,DKIM_SIGNED,FREEMAIL_FROM,MAILING_LIST_MULTI,RCVD_IN_DNSWL_HI autolearn=unavailable version=3.3.1 Received: from lists.gnu.org (lists.gnu.org [208.118.235.17]) (using TLSv1 with cipher AES256-SHA (256/256 bits)) (No client certificate requested) by mail.wl.linuxfoundation.org (Postfix) with ESMTPS id BE44B2A0B7 for ; Thu, 13 Dec 2018 08:00:37 +0000 (UTC) Received: from localhost ([::1]:51052 helo=lists.gnu.org) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1gXLvB-0005Yl-1w for patchwork-qemu-devel@patchwork.kernel.org; Thu, 13 Dec 2018 03:00:37 -0500 Received: from eggs.gnu.org ([2001:4830:134:3::10]:39177) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1gXLsS-0003jr-DM for qemu-devel@nongnu.org; Thu, 13 Dec 2018 02:57:50 -0500 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1gXLsO-0001qk-Ol for qemu-devel@nongnu.org; Thu, 13 Dec 2018 02:57:48 -0500 Received: from mail-pf1-x444.google.com ([2607:f8b0:4864:20::444]:38865) by eggs.gnu.org with esmtps (TLS1.0:RSA_AES_128_CBC_SHA1:16) (Exim 4.71) (envelope-from ) id 1gXLsO-0001lq-2t for qemu-devel@nongnu.org; Thu, 13 Dec 2018 02:57:44 -0500 Received: by mail-pf1-x444.google.com with SMTP id q1so704856pfi.5 for ; Wed, 12 Dec 2018 23:57:42 -0800 (PST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20161025; h=from:to:cc:subject:date:message-id:in-reply-to:references; bh=IbwTS/pVa7QqyjfSBY28dlTYLRu4YCmeh4Z/NKx5tbk=; b=uNiCR8jHqAD5gUyj6KzEQoqY+FafBkSuWcxCgPk7StP5xCeF7HptjQ4kUubEftxFg9 uciWoiNT4J9WJIbW0R4ZmTVAAPrq0J9l80D7ZQaYCiJTKkJX9xEsXHYautjssc/4ggBx 1W7kc4MBthp26SNGrvuaRht1FrpdMECHwQtkAFv9Kkqa27gT7rBsZXDC42+SgeW9G/QB eWE0bTNX+5Qc2eDtf+HKSKXupIDc5ZhAojoe6jf/xHdqlvJ6UGgJmyLH9i8rY20nLlvM BptUteWZgSgKQOxq68OwUjcFZWtVHK/3uzOHaSpCNjmm2QZyctHJNTWOGsWO2IraWTUH AWBg== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:cc:subject:date:message-id:in-reply-to :references; bh=IbwTS/pVa7QqyjfSBY28dlTYLRu4YCmeh4Z/NKx5tbk=; b=jA0Il+TC1Lvu0iAKI4c7PbGs2jMUEt8rCyl/ypQFgHvg/XiTaCa1VDzfkNW0/RjhDG dcoL/M423FdJ9CfI9iV5lZxvG477uItzpSvp0fo+noA6DSFxx0Xcp+kjY1KJjCGYq464 0YZENCwnxceDGrRcYXIgsyvBcaL+xDiXa99HubZpdRi7OBM/wiTcnWBCDlMKCLWfVXwW FFlBNTC11MVSvgI3h4jZLpUSkMWV+AaMQDQLZjsVCgJyxGUV+Owy7m00ZWyFXBajJ74s bk2EypSzm2+hDsFxnnjBYJ10/5Cmq4kWASGXjpilGr50RSDAGvf1917vOMhhYqbibDQh h9qw== X-Gm-Message-State: AA+aEWb4OHGR4LVs5AEeQSEsg6yWcg0aUZvptunVlSvfyBYMK7ohdo0L iTXacii4Tl7Y9MRBk4w1+Sg= X-Google-Smtp-Source: AFSGD/U1RwYBQyfvL721aGucVD/en/NZfWqJM1XWBMrt5im8KBeB6TxZgaqvSZx7CmUReLBtbil6EQ== X-Received: by 2002:a62:c683:: with SMTP id x3mr22646894pfk.10.1544687861140; Wed, 12 Dec 2018 23:57:41 -0800 (PST) Received: from localhost.localdomain ([203.205.141.36]) by smtp.gmail.com with ESMTPSA id h19sm1423618pfn.114.2018.12.12.23.57.37 (version=TLS1_2 cipher=ECDHE-RSA-CHACHA20-POLY1305 bits=256/256); Wed, 12 Dec 2018 23:57:40 -0800 (PST) From: guangrong.xiao@gmail.com X-Google-Original-From: xiaoguangrong@tencent.com To: pbonzini@redhat.com, mst@redhat.com, mtosatti@redhat.com Date: Thu, 13 Dec 2018 15:57:26 +0800 Message-Id: <20181213075727.23540-2-xiaoguangrong@tencent.com> X-Mailer: git-send-email 2.14.5 In-Reply-To: <20181213075727.23540-1-xiaoguangrong@tencent.com> References: <20181213075727.23540-1-xiaoguangrong@tencent.com> X-detected-operating-system: by eggs.gnu.org: Genre and OS details not recognized. X-Received-From: 2607:f8b0:4864:20::444 Subject: [Qemu-devel] [PATCH 1/2] migration: introduce compress-wait-thread-adaptive X-BeenThere: qemu-devel@nongnu.org X-Mailman-Version: 2.1.21 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: kvm@vger.kernel.org, quintela@redhat.com, Xiao Guangrong , qemu-devel@nongnu.org, peterx@redhat.com, dgilbert@redhat.com, wei.w.wang@intel.com, cota@braap.org Errors-To: qemu-devel-bounces+patchwork-qemu-devel=patchwork.kernel.org@nongnu.org Sender: "Qemu-devel" X-Virus-Scanned: ClamAV using ClamSMTP From: Xiao Guangrong Currently we have two behaviors if all threads are busy to do compression, the main thread mush wait one of them becoming free if @compress-wait-thread set to on or the main thread can directly return without wait and post the page out as normal one Both of them have its profits and short-comes, however, if the bandwidth is not limited extremely so that compression can not use out all of it bandwidth, at the same time, the migration process is easily throttled if we posted too may pages as normal pages. None of them can work properly under this case In order to use the bandwidth more effectively, we introduce the third behavior, compress-wait-thread-adaptive, which make the main thread wait if there is no bandwidth left or let the page go out as normal page if there has enough bandwidth to make sure the migration process will not be throttled Signed-off-by: Xiao Guangrong --- hmp.c | 11 +++++ migration/migration.c | 37 ++++++++++++++++ migration/migration.h | 4 ++ migration/ram.c | 118 +++++++++++++++++++++++++++++++++++++++++++++----- qapi/migration.json | 26 ++++++++++- 5 files changed, 184 insertions(+), 12 deletions(-) diff --git a/hmp.c b/hmp.c index 43ae9ec61a..2c5bb504d4 100644 --- a/hmp.c +++ b/hmp.c @@ -282,6 +282,8 @@ void hmp_info_migrate(Monitor *mon, const QDict *qdict) info->compression->compressed_size); monitor_printf(mon, "compression rate: %0.2f\n", info->compression->compression_rate); + monitor_printf(mon, "compress-no-wait-weight: %"PRIu64"\n", + info->compression->compress_no_wait_weight); } if (info->has_cpu_throttle_percentage) { @@ -344,6 +346,11 @@ void hmp_info_migrate_parameters(Monitor *mon, const QDict *qdict) monitor_printf(mon, "%s: %s\n", MigrationParameter_str(MIGRATION_PARAMETER_COMPRESS_WAIT_THREAD), params->compress_wait_thread ? "on" : "off"); + assert(params->has_compress_wait_thread_adaptive); + monitor_printf(mon, "%s: %s\n", + MigrationParameter_str( + MIGRATION_PARAMETER_COMPRESS_WAIT_THREAD_ADAPTIVE), + params->compress_wait_thread_adaptive ? "on" : "off"); assert(params->has_decompress_threads); monitor_printf(mon, "%s: %u\n", MigrationParameter_str(MIGRATION_PARAMETER_DECOMPRESS_THREADS), @@ -1676,6 +1683,10 @@ void hmp_migrate_set_parameter(Monitor *mon, const QDict *qdict) p->has_compress_wait_thread = true; visit_type_bool(v, param, &p->compress_wait_thread, &err); break; + case MIGRATION_PARAMETER_COMPRESS_WAIT_THREAD_ADAPTIVE: + p->has_compress_wait_thread_adaptive = true; + visit_type_bool(v, param, &p->compress_wait_thread_adaptive, &err); + break; case MIGRATION_PARAMETER_DECOMPRESS_THREADS: p->has_decompress_threads = true; visit_type_int(v, param, &p->decompress_threads, &err); diff --git a/migration/migration.c b/migration/migration.c index 49ffb9997a..d19935b529 100644 --- a/migration/migration.c +++ b/migration/migration.c @@ -706,6 +706,9 @@ MigrationParameters *qmp_query_migrate_parameters(Error **errp) params->compress_threads = s->parameters.compress_threads; params->has_compress_wait_thread = true; params->compress_wait_thread = s->parameters.compress_wait_thread; + params->has_compress_wait_thread_adaptive = true; + params->compress_wait_thread_adaptive = + s->parameters.compress_wait_thread_adaptive; params->has_decompress_threads = true; params->decompress_threads = s->parameters.decompress_threads; params->has_cpu_throttle_initial = true; @@ -799,6 +802,8 @@ static void populate_ram_info(MigrationInfo *info, MigrationState *s) compression_counters.compressed_size; info->compression->compression_rate = compression_counters.compression_rate; + info->compression->compress_no_wait_weight = + compression_counters.compress_no_wait_weight; } if (cpu_throttle_active()) { @@ -1132,6 +1137,11 @@ static void migrate_params_test_apply(MigrateSetParameters *params, dest->compress_wait_thread = params->compress_wait_thread; } + if (params->has_compress_wait_thread_adaptive) { + dest->compress_wait_thread_adaptive = + params->compress_wait_thread_adaptive; + } + if (params->has_decompress_threads) { dest->decompress_threads = params->decompress_threads; } @@ -1204,6 +1214,11 @@ static void migrate_params_apply(MigrateSetParameters *params, Error **errp) s->parameters.compress_wait_thread = params->compress_wait_thread; } + if (params->has_compress_wait_thread_adaptive) { + s->parameters.compress_wait_thread_adaptive = + params->compress_wait_thread_adaptive; + } + if (params->has_decompress_threads) { s->parameters.decompress_threads = params->decompress_threads; } @@ -1925,6 +1940,15 @@ bool migrate_postcopy_blocktime(void) return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME]; } +int64_t migration_max_bandwidth(void) +{ + MigrationState *s; + + s = migrate_get_current(); + + return s->parameters.max_bandwidth; +} + bool migrate_use_compression(void) { MigrationState *s; @@ -1961,6 +1985,15 @@ int migrate_compress_wait_thread(void) return s->parameters.compress_wait_thread; } +int migrate_compress_wait_thread_adaptive(void) +{ + MigrationState *s; + + s = migrate_get_current(); + + return s->parameters.compress_wait_thread_adaptive; +} + int migrate_decompress_threads(void) { MigrationState *s; @@ -2898,6 +2931,8 @@ static void migration_update_counters(MigrationState *s, s->mbps = (((double) transferred * 8.0) / ((double) time_spent / 1000.0)) / 1000.0 / 1000.0; + compress_adaptive_update(s->mbps); + /* * if we haven't sent anything, we don't want to * recalculate. 10000 is a small enough number for our purposes @@ -3232,6 +3267,8 @@ static Property migration_properties[] = { DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT), DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState, parameters.compress_wait_thread, true), + DEFINE_PROP_BOOL("x-ccompress-wait-thread-adaptive", MigrationState, + parameters.compress_wait_thread_adaptive, false), DEFINE_PROP_UINT8("x-decompress-threads", MigrationState, parameters.decompress_threads, DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT), diff --git a/migration/migration.h b/migration/migration.h index e413d4d8b6..d631776230 100644 --- a/migration/migration.h +++ b/migration/migration.h @@ -270,11 +270,15 @@ bool migrate_use_block(void); bool migrate_use_block_incremental(void); int migrate_max_cpu_throttle(void); bool migrate_use_return_path(void); +int64_t migration_max_bandwidth(void); bool migrate_use_compression(void); int migrate_compress_level(void); int migrate_compress_threads(void); int migrate_compress_wait_thread(void); +int migrate_compress_wait_thread_adaptive(void); +void compress_adaptive_update(double mbps); + int migrate_decompress_threads(void); bool migrate_use_events(void); bool migrate_postcopy_blocktime(void); diff --git a/migration/ram.c b/migration/ram.c index 7e7deec4d8..3b08a605e4 100644 --- a/migration/ram.c +++ b/migration/ram.c @@ -276,6 +276,8 @@ struct RAMSrcPageRequest { QSIMPLEQ_ENTRY(RAMSrcPageRequest) next_req; }; +#define COMPRESS_BUSY_COUNT_PERIOD 100 + /* State of RAM for migration */ struct RAMState { /* QEMUFile used for this migration */ @@ -292,6 +294,19 @@ struct RAMState { bool ram_bulk_stage; /* How many times we have dirty too many pages */ int dirty_rate_high_cnt; + + /* used by by compress-wait-thread-adaptive */ + /* + * the count for the case that all compress threads are busy to + * handle a page in a period + */ + uint8_t compress_busy_count; + /* + * the number of pages that can be directly posted as normal page when + * all compress threads are busy in a period + */ + uint8_t compress_no_wait_left; + /* these variables are used for bitmap sync */ /* last time we did a full bitmap_sync */ int64_t time_last_bitmap_sync; @@ -470,6 +485,8 @@ static void compress_threads_save_cleanup(void) comp_param = NULL; } +static void compress_adaptive_init(void); + static int compress_threads_save_setup(void) { int i, thread_count; @@ -477,6 +494,9 @@ static int compress_threads_save_setup(void) if (!migrate_use_compression()) { return 0; } + + compress_adaptive_init(); + thread_count = migrate_compress_threads(); compress_threads = g_new0(QemuThread, thread_count); comp_param = g_new0(CompressParam, thread_count); @@ -1593,6 +1613,67 @@ uint64_t ram_pagesize_summary(void) return summary; } +static void compress_adaptive_init(void) +{ + /* fully wait on default. */ + compression_counters.compress_no_wait_weight = 0; + ram_state->compress_no_wait_left = 0; + ram_state->compress_busy_count = 0; +} + +void compress_adaptive_update(double mbps) +{ + int64_t rate_limit, remain_bw, max_bw = migration_max_bandwidth(); + + if (!migrate_use_compression() || + !migrate_compress_wait_thread_adaptive()) { + return; + } + + /* no bandwith is set to the file then we can not do adaptive adjustment */ + rate_limit = qemu_file_get_rate_limit(migrate_get_current()->to_dst_file); + if (rate_limit == 0 || rate_limit == INT64_MAX) { + return; + } + + max_bw = (max_bw >> 20) * 8; + remain_bw = abs(max_bw - (int64_t)(mbps)); + if (remain_bw <= ((max_bw / 10) * 2)) { + /* if we have used all the bandwidth, let's compress more. */ + if (compression_counters.compress_no_wait_weight) { + compression_counters.compress_no_wait_weight--; + } + goto exit; + } + + /* have enough bandwidth left, do not need to compress so aggressively */ + if (compression_counters.compress_no_wait_weight != + COMPRESS_BUSY_COUNT_PERIOD) { + compression_counters.compress_no_wait_weight++; + } + +exit: + ram_state->compress_busy_count = 0; + ram_state->compress_no_wait_left = + compression_counters.compress_no_wait_weight; +} + +static bool compress_adaptive_need_wait(void) +{ + if (++ram_state->compress_busy_count == COMPRESS_BUSY_COUNT_PERIOD) { + ram_state->compress_busy_count = 0; + ram_state->compress_no_wait_left = + compression_counters.compress_no_wait_weight; + } + + if (ram_state->compress_no_wait_left) { + ram_state->compress_no_wait_left--; + return false; + } + + return true; +} + static void migration_update_rates(RAMState *rs, int64_t end_time) { uint64_t page_count = rs->target_page_count - rs->target_page_count_prev; @@ -1970,6 +2051,15 @@ static int compress_page_with_multi_thread(RAMState *rs, RAMBlock *block, { int idx, thread_count, bytes_xmit = -1, pages = -1; bool wait = migrate_compress_wait_thread(); + bool adaptive = migrate_compress_wait_thread_adaptive(); + + /* + * 'compress-wait-thread-adaptive' has higher priority than + * 'compress-wait-thread'. + */ + if (adaptive) { + wait = false; + } thread_count = migrate_compress_threads(); qemu_mutex_lock(&comp_done_lock); @@ -1984,20 +2074,29 @@ retry: qemu_mutex_unlock(&comp_param[idx].mutex); pages = 1; update_compress_thread_counts(&comp_param[idx], bytes_xmit); - break; + goto exit; } } + if (adaptive && !wait) { + /* it is the first time go to the loop */ + wait = compress_adaptive_need_wait(); + } + /* - * wait for the free thread if the user specifies 'compress-wait-thread', - * otherwise we will post the page out in the main thread as normal page. + * wait for the free thread if the user specifies + * 'compress-wait-thread-adaptive' that detected the bandwidth is + * not enough or compress-wait-thread', otherwise we will post the + * page out in the main thread as normal page. */ - if (pages < 0 && wait) { + if (wait) { qemu_cond_wait(&comp_done_cond, &comp_done_lock); goto retry; } - qemu_mutex_unlock(&comp_done_lock); + +exit: + qemu_mutex_unlock(&comp_done_lock); return pages; } @@ -3147,19 +3246,18 @@ static int ram_save_setup(QEMUFile *f, void *opaque) RAMState **rsp = opaque; RAMBlock *block; - if (compress_threads_save_setup()) { - return -1; - } - /* migration has already setup the bitmap, reuse it. */ if (!migration_in_colo_state()) { if (ram_init_all(rsp) != 0) { - compress_threads_save_cleanup(); return -1; } } (*rsp)->f = f; + if (compress_threads_save_setup()) { + return -1; + } + rcu_read_lock(); qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE); diff --git a/qapi/migration.json b/qapi/migration.json index 38d4c41d88..6d925c73fc 100644 --- a/qapi/migration.json +++ b/qapi/migration.json @@ -90,11 +90,16 @@ # # @compression-rate: rate of compressed size # +# @compress-no-wait-weight: it controls how many pages are directly posted +# out as normal page when all compression threads are currently busy. +# Only available if compress-wait-thread-adaptive is on. (Since 3.2) +# # Since: 3.1 ## { 'struct': 'CompressionStats', 'data': {'pages': 'int', 'busy': 'int', 'busy-rate': 'number', - 'compressed-size': 'int', 'compression-rate': 'number' } } + 'compressed-size': 'int', 'compression-rate': 'number', + 'compress-no-wait-weight': 'int'} } ## # @MigrationStatus: @@ -490,6 +495,11 @@ # compression thread to become available; otherwise, # send the page uncompressed. (Since 3.1) # +# @compress-wait-thread-adaptive: Adaptively controls compress-wait-thread +# based on the rate limit and it's off on default. When it is on +# and it has enough bandwidth, it acts as compress-wait-thread is +# off. (Since 3.2) +# # @decompress-threads: Set decompression thread count to be used in live # migration, the decompression thread count is an integer between 1 # and 255. Usually, decompression is at least 4 times as fast as @@ -558,7 +568,7 @@ ## { 'enum': 'MigrationParameter', 'data': ['compress-level', 'compress-threads', 'decompress-threads', - 'compress-wait-thread', + 'compress-wait-thread', 'compress-wait-thread-adaptive', 'cpu-throttle-initial', 'cpu-throttle-increment', 'tls-creds', 'tls-hostname', 'max-bandwidth', 'downtime-limit', 'x-checkpoint-delay', 'block-incremental', @@ -578,6 +588,11 @@ # compression thread to become available; otherwise, # send the page uncompressed. (Since 3.1) # +# @compress-wait-thread-adaptive: Adaptively controls compress-wait-thread +# based on the rate limit and it's off on default. When it is on +# and it has enough bandwidth, it acts as compress-wait-thread is +# off. (Since 3.2) +# # @decompress-threads: decompression thread count # # @cpu-throttle-initial: Initial percentage of time guest cpus are @@ -653,6 +668,7 @@ 'data': { '*compress-level': 'int', '*compress-threads': 'int', '*compress-wait-thread': 'bool', + '*compress-wait-thread-adaptive': 'bool', '*decompress-threads': 'int', '*cpu-throttle-initial': 'int', '*cpu-throttle-increment': 'int', @@ -698,6 +714,11 @@ # compression thread to become available; otherwise, # send the page uncompressed. (Since 3.1) # +# @compress-wait-thread-adaptive: Adaptively controls compress-wait-thread +# based on the rate limit and it's off on default. When it is on +# and it has enough bandwidth, it acts as compress-wait-thread is +# off. (Since 3.2) +# # @decompress-threads: decompression thread count # # @cpu-throttle-initial: Initial percentage of time guest cpus are @@ -769,6 +790,7 @@ 'data': { '*compress-level': 'uint8', '*compress-threads': 'uint8', '*compress-wait-thread': 'bool', + '*compress-wait-thread-adaptive': 'bool', '*decompress-threads': 'uint8', '*cpu-throttle-initial': 'uint8', '*cpu-throttle-increment': 'uint8',