Summary:
Since DynamicBloom is now only used in-memory, we're free to
change it without schema compatibility issues. The new implementation
is drawn from (with manifest permission)
303542a767/bloom_simulation_tests/foo.cc (L613)
This has several speed advantages over the prior implementation:
* Uses fastrange instead of %
* Minimum logic to determine first (and all) probed memory addresses
* (Major) Two probes per 64-bit memory fetch/write.
* Very fast and effective (murmur-like) hash expansion/re-mixing. (At
least on recent CPUs, integer multiplication is very cheap.)
While a Bloom filter with 512-bit cache locality has about a 1.15x FP
rate penalty (e.g. 0.84% to 0.97%), further restricting to two probes
per 64 bits incurs an additional 1.12x FP rate penalty (e.g. 0.97% to
1.09%). Nevertheless, the unit tests show no "mediocre" FP rate samples,
unlike the old implementation with more erratic FP rates.
Especially for the memtable, we expect speed to outweigh somewhat higher
FP rates. For example, a negative table query would have to be 1000x
slower than a BF query to justify doubling BF query time to shave 10% off
FP rate (working assumption around 1% FP rate). While that seems likely
for SSTs, my data suggests a speed factor of roughly 50x for the memtable
(vs. BF; ~1.5% lower write throughput when enabling memtable Bloom
filter, after this change). Thus, it's probably not worth even 5% more
time in the Bloom filter to shave off 1/10th of the Bloom FP rate, or 0.1%
in absolute terms, and it's probably at least 20% slower to recoup that
much FP rate from this new implementation. Because of this, we do not see
a need for a 'locality' option that affects the MemTable Bloom filter
and have decoupled the MemTable Bloom filter from Options::bloom_locality.
Note that just 3% more memory to the Bloom filter (10.3 bits per key vs.
just 10) is able to make up for the ~12% FP rate drop in the new
implementation:
[] # Nearly "ideal" FP-wise but reasonably fast cache-local implementation
[~/wormhashing/bloom_simulation_tests] ./foo_gcc_IMPL_CACHE_WORM64_FROM32_any.out 10000000 6 10 $RANDOM 100000000
./foo_gcc_IMPL_CACHE_WORM64_FROM32_any.out time: 3.29372 sampled_fp_rate: 0.00985956 ...
[] # Close match to this new implementation
[~/wormhashing/bloom_simulation_tests] ./foo_gcc_IMPL_CACHE_MUL64_BLOCK_FROM32_any.out 10000000 6 10.3 $RANDOM 100000000
./foo_gcc_IMPL_CACHE_MUL64_BLOCK_FROM32_any.out time: 2.10072 sampled_fp_rate: 0.00985655 ...
[] # Old locality=1 implementation
[~/wormhashing/bloom_simulation_tests] ./foo_gcc_IMPL_CACHE_ROCKSDB_DYNAMIC_any.out 10000000 6 10 $RANDOM 100000000
./foo_gcc_IMPL_CACHE_ROCKSDB_DYNAMIC_any.out time: 3.95472 sampled_fp_rate: 0.00988943 ...
Also note the dramatic speed improvement vs. alternatives.
--
Performance unit test: DynamicBloomTest.concurrent_with_perf is updated
to report more precise timing data. (Measure running time of each
thread, not just longest running thread, etc.) Results averaged over
various sizes enabled with --enable_perf and 20 runs each; old dynamic
bloom refers to locality=1, the faster of the old:
old dynamic bloom, avg add latency = 65.6468
new dynamic bloom, avg add latency = 44.3809
old dynamic bloom, avg query latency = 50.6485
new dynamic bloom, avg query latency = 43.2186
old avg parallel add latency = 41.678
new avg parallel add latency = 24.5238
old avg parallel hit latency = 14.6322
new avg parallel hit latency = 12.3939
old avg parallel miss latency = 16.7289
new avg parallel miss latency = 12.2134
Tested on a dedicated 64-bit production machine at Facebook. Significant
improvement all around.
Despite now using std::atomic<uint64_t>, quick before-and-after test on
a 32-bit machine (Intel Atom N270, released 2008) shows no regression in
performance, in some cases modest improvement.
--
Performance integration test (synthetic): with DEBUG_LEVEL=0, used
TEST_TMPDIR=/dev/shm ./db_bench --benchmarks=fillrandom,readmissing,readrandom,stats --num=2000000
and optionally with -memtable_whole_key_filtering -memtable_bloom_size_ratio=0.01
300 runs each configuration.
Write throughput change by enabling memtable bloom:
Old locality=0: -3.06%
Old locality=1: -2.37%
New: -1.50%
conclusion -> seems to substantially close the gap
Readmissing throughput change by enabling memtable bloom:
Old locality=0: +34.47%
Old locality=1: +34.80%
New: +33.25%
conclusion -> maybe a small new penalty from FP rate
Readrandom throughput change by enabling memtable bloom:
Old locality=0: +31.54%
Old locality=1: +31.13%
New: +30.60%
conclusion -> maybe also from FP rate (after memtable flush)
--
Another conclusion we can draw from this new implementation is that the
existing 32-bit hash function is not inherently crippling the Bloom
filter speed or accuracy, below about 5 million keys. For speed, the
implementation is essentially the same whether starting with 32-bits or
64-bits of hash; it just determines whether the first multiplication
after fastrange is a pseudorandom expansion or needed re-mix. Note that
this multiplication can occur while memory is fetching.
For accuracy, in a standard configuration, you need about 5 million
keys before you have about a 1.1x FP penalty due to using a
32-bit hash vs. 64-bit:
[~/wormhashing/bloom_simulation_tests] ./foo_gcc_IMPL_CACHE_MUL64_BLOCK_FROM32_any.out $((5 * 1000 * 1000 * 10)) 6 10 $RANDOM 100000000
./foo_gcc_IMPL_CACHE_MUL64_BLOCK_FROM32_any.out time: 2.52069 sampled_fp_rate: 0.0118267 ...
[~/wormhashing/bloom_simulation_tests] ./foo_gcc_IMPL_CACHE_MUL64_BLOCK_any.out $((5 * 1000 * 1000 * 10)) 6 10 $RANDOM 100000000
./foo_gcc_IMPL_CACHE_MUL64_BLOCK_any.out time: 2.43871 sampled_fp_rate: 0.0109059
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5762
Differential Revision: D17214194
Pulled By: pdillinger
fbshipit-source-id: ad9da031772e985fd6b62a0e1db8e81892520595
Summary:
DynamicBloom was being used both for memory-only and for on-disk filters, as part of the PlainTable format. To set up enhancements to the memtable Bloom filter, this splits the code into two copies and removes unused features from each copy. Adds test PlainTableDBTest.BloomSchema to ensure no accidental change to that format.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5767
Differential Revision: D17206963
Pulled By: pdillinger
fbshipit-source-id: 6cce8d55305ed0df051b4c58bdc98c8ad81d0553
Summary:
VersionSet::ApproximateSize doesn't need to create two separate index iterators and do binary search for each in BlockBasedTable. So BlockBasedTable::ApproximateSize was added that creates the iterator once and uses it to calculate the data size between start and end keys.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5693
Differential Revision: D16774056
Pulled By: elipoz
fbshipit-source-id: 53ce262e1a057788243bf30cd9b8aa6581df1a18
Summary:
This PR adds more callers for table readers. These information are only used for block cache analysis so that we can know which caller accesses a block.
1. It renames the BlockCacheLookupCaller to TableReaderCaller as passing the caller from upstream requires changes to table_reader.h and TableReaderCaller is a more appropriate name.
2. It adds more table reader callers in table/table_reader_caller.h, e.g., kCompactionRefill, kExternalSSTIngestion, and kBuildTable.
This PR is long as it requires modification of interfaces in table_reader.h, e.g., NewIterator.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5454
Test Plan: make clean && COMPILE_WITH_ASAN=1 make check -j32.
Differential Revision: D15819451
Pulled By: HaoyuHuang
fbshipit-source-id: b6caa704c8fb96ddd15b9a934b7e7ea87f88092d
Summary:
Currently the read-ahead logic for user reads and compaction reads go through different code paths where compaction reads create new table readers and use `ReadaheadRandomAccessFile`. This change is to unify read-ahead logic to use read-ahead in BlockBasedTableReader::InitDataBlock(). As a result of the change `ReadAheadRandomAccessFile` class and `new_table_reader_for_compaction_inputs` option will no longer be used.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5431
Test Plan:
make check
Here is the benchmarking - https://gist.github.com/vjnadimpalli/083cf423f7b6aa12dcdb14c858bc18a5
Differential Revision: D15772533
Pulled By: vjnadimpalli
fbshipit-source-id: b71dca710590471ede6fb37553388654e2e479b9
Summary:
The patch brings the semantics of per-block-type read performance
context counters in sync with the generic block_read_count by only
incrementing the counter if the block was actually read from the file.
It also fixes index_block_read_count, which fell victim to the
refactoring in PR https://github.com/facebook/rocksdb/issues/5298.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5484
Test Plan: Extended the unit tests.
Differential Revision: D15887431
Pulled By: ltamasi
fbshipit-source-id: a3889759d0ac5759d56625d692cd828d1b9207a6
Summary:
BlockCacheLookupContext only contains the caller for now.
We will trace block accesses at five places:
1. BlockBasedTable::GetFilter.
2. BlockBasedTable::GetUncompressedDict.
3. BlockBasedTable::MaybeReadAndLoadToCache. (To trace access on data, index, and range deletion block.)
4. BlockBasedTable::Get. (To trace the referenced key and whether the referenced key exists in a fetched data block.)
5. BlockBasedTable::MultiGet. (To trace the referenced key and whether the referenced key exists in a fetched data block.)
We create the context at:
1. BlockBasedTable::Get. (kUserGet)
2. BlockBasedTable::MultiGet. (kUserMGet)
3. BlockBasedTable::NewIterator. (either kUserIterator, kCompaction, or external SST ingestion calls this function.)
4. BlockBasedTable::Open. (kPrefetch)
5. Index/Filter::CacheDependencies. (kPrefetch)
6. BlockBasedTable::ApproximateOffsetOf. (kCompaction or kUserApproximateSize).
I loaded 1 million key-value pairs into the database and ran the readrandom benchmark with a single thread. I gave the block cache 10 GB to make sure all reads hit the block cache after warmup. The throughput is comparable.
Throughput of this PR: 231334 ops/s.
Throughput of the master branch: 238428 ops/s.
Experiment setup:
RocksDB: version 6.2
Date: Mon Jun 10 10:42:51 2019
CPU: 24 * Intel Core Processor (Skylake)
CPUCache: 16384 KB
Keys: 20 bytes each
Values: 100 bytes each (100 bytes after compression)
Entries: 1000000
Prefix: 20 bytes
Keys per prefix: 0
RawSize: 114.4 MB (estimated)
FileSize: 114.4 MB (estimated)
Write rate: 0 bytes/second
Read rate: 0 ops/second
Compression: NoCompression
Compression sampling rate: 0
Memtablerep: skip_list
Perf Level: 1
Load command: ./db_bench --benchmarks="fillseq" --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --statistics --cache_index_and_filter_blocks --cache_size=10737418240 --disable_auto_compactions=1 --disable_wal=1 --compression_type=none --min_level_to_compress=-1 --compression_ratio=1 --num=1000000
Run command: ./db_bench --benchmarks="readrandom,stats" --use_existing_db --threads=1 --duration=120 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --statistics --cache_index_and_filter_blocks --cache_size=10737418240 --disable_auto_compactions=1 --disable_wal=1 --compression_type=none --min_level_to_compress=-1 --compression_ratio=1 --num=1000000 --duration=120
TODOs:
1. Create a caller for external SST file ingestion and differentiate the callers for iterator.
2. Integrate tracer to trace block cache accesses.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5421
Differential Revision: D15704258
Pulled By: HaoyuHuang
fbshipit-source-id: 4aa8a55f8cb1576ffb367bfa3186a91d8f06d93a
Summary:
When using `PRIu64` type of printf specifier, current code base does the following:
```
#ifndef __STDC_FORMAT_MACROS
#define __STDC_FORMAT_MACROS
#endif
#include <inttypes.h>
```
However, this can be simplified to
```
#include <cinttypes>
```
as long as flag `-std=c++11` is used.
This should solve issues like https://github.com/facebook/rocksdb/issues/5159
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5402
Differential Revision: D15701195
Pulled By: miasantreble
fbshipit-source-id: 6dac0a05f52aadb55e9728038599d3d2e4b59d03
Peter Dillinger