Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Improve DeltaTree update routine #5200

Closed
SchrodingerZhu opened this issue Jun 23, 2022 · 2 comments · Fixed by #5199
Closed

Improve DeltaTree update routine #5200

SchrodingerZhu opened this issue Jun 23, 2022 · 2 comments · Fixed by #5199
Labels
type/enhancement The issue or PR belongs to an enhancement.

Comments

@SchrodingerZhu
Copy link
Contributor

Enhancement

On DeltaTree update routine (afterNodeUpdated), the call to getDelta() can be effectively translated by the compiler to SIMD memory scanning.

image

Since compiler has already decides that afterNodeUpdated is too large to be inlined, we can actually dispatch this function to multiple targets. Indeed, it enjoys different codegen for avx targets:

image

Experiments have shown that, it can provide up to 12% speed up at IDC environment:

2022-06-23T08:53:15+08:00
Running dbms/src/Storages/DeltaMerge/Benchmark/bench_deltatree
Run on (40 X 2394.97 MHz CPU s)
CPU Caches:
  L1 Data 32 KiB (x20)
  L1 Instruction 32 KiB (x20)
  L2 Unified 256 KiB (x20)
  L3 Unified 25600 KiB (x2)
Load Average: 8.91, 23.44, 19.41
------------------------------------------------------------------------------------------------
Benchmark                                                      Time             CPU   Iterations
------------------------------------------------------------------------------------------------
BM_deltatree_dispatch<true>/iterations:50/threads:40     1764554 ns     59946595 ns         2000
BM_deltatree_dispatch<false>/iterations:50/threads:40    1553962 ns     55583999 ns         2000

@SchrodingerZhu SchrodingerZhu added the type/enhancement The issue or PR belongs to an enhancement. label Jun 23, 2022
@SchrodingerZhu
Copy link
Contributor Author

SchrodingerZhu commented Jun 23, 2022

Test cases:

template <bool use_generic>
static void BM_deltatree_dispatch(benchmark::State & state)
{
    if (use_generic)
    {
        DB::DM::Impl::DELTA_TREE_VARIANT = DB::DM::Impl::DeltaTreeVariant::Generic;
    } else {
        DB::DM::Impl::DELTA_TREE_VARIANT = DB::DM::Impl::resolveDeltaTreeVariant();
    }

    SystemTree tree;
    for (auto _ : state)
    {
        for (int i = 0; i < 32768; ++i)
        {
            tree.addInsert(i, i);
        }
        for (int i = 0; i < 32768; ++i)
        {
            tree.addDelete(i);
        }

    }
}

BENCHMARK_TEMPLATE(BM_deltatree_dispatch, true)->Iterations(50);
BENCHMARK_TEMPLATE(BM_deltatree_dispatch, false)->Iterations(50); 

@flowbehappy
Copy link
Contributor

Looks like it only works for x86. Is there anyway to port to arm?

ti-chi-bot pushed a commit that referenced this issue Jun 24, 2022
Lloyd-Pottiger pushed a commit to Lloyd-Pottiger/tiflash that referenced this issue Jul 12, 2022
…s in README (pingcap#5182)

close pingcap#5172, ref pingcap#5178

Enhancement: add a integrated test on DDL module (pingcap#5130)

ref pingcap#5129

Revert "Revise default background threads size" (pingcap#5176)

close pingcap#5177

chore: remove extra dyn cast (pingcap#5186)

close pingcap#5185

Add MPPReceiverSet, which includes ExchangeReceiver and CoprocessorReader (pingcap#5175)

ref pingcap#5095

DDL: Use Column Name Instead of Offset to Find the common handle cluster index (pingcap#5166)

close pingcap#5154

Add random failpoint in critical paths (pingcap#4876)

close pingcap#4807

Segment test framework (pingcap#5150)

close pingcap#5151

optimize ps v3 restore (pingcap#5163)

ref pingcap#4914

Fix build failed (pingcap#5196)

close pingcap#5195

feat: delta tree dispatching (pingcap#5199)

close pingcap#5200

feat: introduce specialized API to write fixed length data rapidly (pingcap#5181)

close pingcap#5183

Add gtest for Limit, TopN, Projection (pingcap#5187) (pingcap#5188)

close pingcap#5187

add `MPPTask::handleError()` (pingcap#5202)

ref pingcap#5095

Check result of starting grpc server (pingcap#5257)

close pingcap#5255

feat: add optimized routines for aarch64 (pingcap#5231)

close pingcap#5240

fix: aarch64-quick-fix (pingcap#5259)

close pingcap#5260

Update client-c to support ipv6 (pingcap#5270)

close pingcap#5247

upgrade prometheus-cpp to v1.0.1 (pingcap#5279)

ref pingcap#2103, close pingcap#5278

Fix README type error (pingcap#5273)

ref pingcap#5178

fix(cmake): make sure libc++ is utilized by tiflash-proxy (pingcap#5281)

close pingcap#5282

fix the wrong order of execution summary for list based executors (pingcap#5242)

close pingcap#5241

Schema: allow loading empty schema diff when the version grows up. (pingcap#5245)

close pingcap#5244

Optimize apply speed under heavy write pressure (pingcap#4883)

ref pingcap#4728

update proxy to raftstore-proxy-6.2 (pingcap#5287)

ref pingcap#4982

Flush segment cache when doing the compaction (pingcap#5284)

close pingcap#5179

metrics: Fix incorrect metrics for delta_merge tasks (pingcap#5061)

close pingcap#5055

dep: upgrade jemalloc (pingcap#5197)

close pingcap#5258

*: TiFlash pagectl/dttool use only-decryption mode (pingcap#5271)

close pingcap#5122

suppresion false positive report from tsan (pingcap#5303)

close pingcap#5088

Refine test framework code and tests (pingcap#5261)

close pingcap#5262

feat: add logical cpu cores and memory into grafana (pingcap#5124)

close pingcap#3821

Implement TimeToSec function push down (pingcap#5235)

close pingcap#5116

feat: implement shiftRight function push down (pingcap#5156)

close pingcap#5100

schema : make update to partition tables when 'set tiflash replica' (pingcap#5267)

close pingcap#5266

Replace initializer_list with vector for planner test framework (pingcap#5307)

close pingcap#5295

KVStore: decouple flush region and CompactLog with a new FFI fn_try_flush_data (pingcap#5283)

ref pingcap#5170

refine error message in mpptask (pingcap#5304)

ref pingcap#5095

Implement ReverseUTF8/Reverse function push down (pingcap#5233)

close pingcap#5111

Optimize comparision for collation `UTF8_BIN` and `UTF8MB4_BIN` (pingcap#5299)

ref pingcap#5294

feat : support set tiflash mode ddl action (pingcap#5256)

ref pingcap#5252

Add non-blocking functions for MPMCQueue (pingcap#5311)

close pingcap#5310

add random segment test for CI weekly (pingcap#5300)

close pingcap#5301

*: tidy FunctionString.cpp (pingcap#5312)

close pingcap#5313

ci: fix check-license github action (pingcap#5318)

close pingcap#5317

update proxy to raftstore-proxy-6.2 (pingcap#5316)

ref pingcap#4982

Change one `additional_input_at_end` to many streams in `ParallelInputsProcessor`  (pingcap#5274)

close pingcap#4856, close pingcap#5263

support fine grained shuffle for window function (pingcap#5048)

close pingcap#5142

feat: pushdown get_format into TiFlash (pingcap#5269)

close pingcap#5115

fix: format throw data truncated error (pingcap#5272)

close pingcap#4891

Print content of columns for gtest (pingcap#5243)

close pingcap#5203

*: also enable O3 for aarch64 (pingcap#5338)

close pingcap#5342

Add debug image build target for CentOS7 (pingcap#5344)

close pingcap#5343

*: mini refactor (pingcap#5326)

close pingcap#4739

Refactor initialize of background pool (pingcap#5190)

close pingcap#5189

delete copy/move ctor of MPMCQueue explicitly (pingcap#5328)

close pingcap#5329

Introduce proxy_server and new-mock-engine-store (pingcap#5319)

ref pingcap#5170

fix: incorrect uptime in grafana panel

Signed-off-by: Lloyd-Pottiger <yan1579196623@gmail.com>
Lloyd-Pottiger pushed a commit to Lloyd-Pottiger/tiflash that referenced this issue Jul 19, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
type/enhancement The issue or PR belongs to an enhancement.
Projects
None yet
Development

Successfully merging a pull request may close this issue.

2 participants