Summary:
* JemallocNodumpAllocator was passing a size_t to FastRange32, which could cause compilation errors or warnings (seen with clang)
* Fixed the order of arguments to match what would be used with modulo operator (%), for clarity.
Fixes https://github.com/facebook/rocksdb/issues/11006
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11707
Test Plan: no functional change, existing tests
Reviewed By: ajkr
Differential Revision: D48435149
Pulled By: pdillinger
fbshipit-source-id: e6e8b107ded4eceda37db20df59985c846a2546b
Summary:
**Background** - runtime detection of certain x86 CPU features was added for optimizing CRC32c checksums, where performance is dramatically affected by the availability of certain CPU instructions and code using intrinsics for those instructions. And Java builds with native library try to be broadly compatible but performant.
What has changed is that CRC32c is no longer the most efficient cheecksum on contemporary x86_64 hardware, nor the default checksum. XXH3 is generally faster and not as dramatically impacted by the availability of certain CPU instructions. For example, on my Skylake system using db_bench (similar on an older Skylake system without AVX512):
PORTABLE=1 empty USE_SSE : xxh3->8 GB/s crc32c->0.8 GB/s (no SSE4.2 nor AVX2 instructions)
PORTABLE=1 USE_SSE=1 : xxh3->19 GB/s crc32c->16 GB/s (with SSE4.2 and AVX2)
PORTABLE=0 USE_SSE ignored: xxh3->28 GB/s crc32c->16 GB/s (also some AVX512)
Testing a ~10 year old system, with SSE4.2 but without AVX2, crc32c is a similar speed to the new systems but xxh3 is only about half that speed, also 8GB/s like the non-AVX2 compile above. Given that xxh3 has specific optimization for AVX2, I think we can infer that that crc32c is only fastest for that ~2008-2013 period when SSE4.2 was included but not AVX2. And given that xxh3 is only about 2x slower on these systems (not like >10x slower for unoptimized crc32c), I don't think we need to invest too much in optimally adapting to these old cases.
x86 hardware that doesn't support fast CRC32c is now extremely rare, so requiring a custom build to support such hardware is fine IMHO.
**This change** does two related things:
* Remove runtime CPU detection for optimizing CRC32c on x86. Maintaining this code is non-zero work, and compiling special code that doesn't work on the configured target instruction set for code generation is always dubious. (On the one hand we have to ensure the CRC32c code uses SSE4.2 but on the other hand we have to ensure nothing else does.)
* Detect CPU features in source code, not in build scripts. Although there are some hypothetical advantages to detectiong in build scripts (compiler generality), RocksDB supports at least three build systems: make, cmake, and buck. It's not practical to support feature detection on all three, and we have suffered from missed optimization opportunities by relying on missing or incomplete detection in cmake and buck. We also depend on some components like xxhash that do source code detection anyway.
**In more detail:**
* `HAVE_SSE42`, `HAVE_AVX2`, and `HAVE_PCLMUL` replaced by standard macros `__SSE4_2__`, `__AVX2__`, and `__PCLMUL__`.
* MSVC does not provide high fidelity defines for SSE, PCLMUL, or POPCNT, but we can infer those from `__AVX__` or `__AVX2__` in a compatibility header. In rare cases of false negative or false positive feature detection, a build engineer should be able to set defines to work around the issue.
* `__POPCNT__` is another standard define, but we happen to only need it on MSVC, where it is set by that compatibility header, or can be set by the build engineer.
* `PORTABLE` can be set to a CPU type, e.g. "haswell", to compile for that CPU type.
* `USE_SSE` is deprecated, now equivalent to PORTABLE=haswell, which roughly approximates its old behavior.
Notably, this change should enable more builds to use the AVX2-optimized Bloom filter implementation.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11419
Test Plan:
existing tests, CI
Manual performance tests after the change match the before above (none expected with make build).
We also see AVX2 optimized Bloom filter code enabled when expected, by injecting a compiler error. (Performance difference is not big on my current CPU.)
Reviewed By: ajkr
Differential Revision: D45489041
Pulled By: pdillinger
fbshipit-source-id: 60ceb0dd2aa3b365c99ed08a8b2a087a9abb6a70
Summary:
* Inefficient block-based filter is no longer customizable in the public
API, though (for now) can still be enabled.
* Removed deprecated FilterPolicy::CreateFilter() and
FilterPolicy::KeyMayMatch()
* Removed `rocksdb_filterpolicy_create()` from C API
* Change meaning of nullptr return from GetBuilderWithContext() from "use
block-based filter" to "generate no filter in this case." This is a
cleaner solution to the proposal in https://github.com/facebook/rocksdb/issues/8250.
* Also, when user specifies bits_per_key < 0.5, we now round this down
to "no filter" because we expect a filter with >= 80% FP rate is
unlikely to be worth the CPU cost of accessing it (esp with
cache_index_and_filter_blocks=1 or partition_filters=1).
* bits_per_key >= 0.5 and < 1.0 is still rounded up to 1.0 (for 62% FP
rate)
* This also gives us some support for configuring filters from OPTIONS
file as currently saved: `filter_policy=rocksdb.BuiltinBloomFilter`.
Opening from such an options file will enable reading filters (an
improvement) but not writing new ones. (See Customizable follow-up
below.)
* Also removed deprecated functions
* FilterBitsBuilder::CalculateNumEntry()
* FilterPolicy::GetFilterBitsBuilder()
* NewExperimentalRibbonFilterPolicy()
* Remove default implementations of
* FilterBitsBuilder::EstimateEntriesAdded()
* FilterBitsBuilder::ApproximateNumEntries()
* FilterPolicy::GetBuilderWithContext()
* Remove support for "filter_policy=experimental_ribbon" configuration
string.
* Allow "filter_policy=bloomfilter:n" without bool to discourage use of
block-based filter.
Some pieces for https://github.com/facebook/rocksdb/issues/9389
Likely follow-up (later PRs):
* Refactoring toward FilterPolicy Customizable, so that we can generate
filters with same configuration as before when configuring from options
file.
* Remove support for user enabling block-based filter (ignore `bool
use_block_based_builder`)
* Some months after this change, we could even remove read support for
block-based filter, because it is not critical to DB data
preservation.
* Make FilterBitsBuilder::FinishV2 to avoid `using
FilterBitsBuilder::Finish` mess and add support for specifying a
MemoryAllocator (for cache warming)
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9501
Test Plan:
A number of obsolete tests deleted and new tests or test
cases added or updated.
Reviewed By: hx235
Differential Revision: D34008011
Pulled By: pdillinger
fbshipit-source-id: a39a720457c354e00d5b59166b686f7f59e392aa
Summary:
The core algorithms for InterleavedSolutionStorage and the
implementation SerializableInterleavedSolution make Ribbon fast for
filter queries. Example output from new unit test:
Simple outside query, hot, incl hashing, ns/key: 117.796
Interleaved outside query, hot, incl hashing, ns/key: 42.2655
Bloom outside query, hot, incl hashing, ns/key: 24.0071
Also includes misc cleanup of previous Ribbon code and comments.
Some TODOs and FIXMEs remain for futher work / investigation.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/7598
Test Plan: unit tests included (integration work and tests coming later)
Reviewed By: jay-zhuang
Differential Revision: D24559209
Pulled By: pdillinger
fbshipit-source-id: fea483cd354ba782aea3e806f2bc96e183d59441
Summary:
A generic algorithm in progress depends on a templatized
version of fastrange, so this change generalizes it and renames
it to fit our style guidelines, FastRange32, FastRange64, and now
FastRangeGeneric.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/7436
Test Plan: added a few more test cases
Reviewed By: jay-zhuang
Differential Revision: D23958153
Pulled By: pdillinger
fbshipit-source-id: 8c3b76101653417804997e5f076623a25586f3e8
Summary:
When dynamically linking two binaries together, different builds of RocksDB from two sources might cause errors. To provide a tool for user to solve the problem, the RocksDB namespace is changed to a flag which can be overridden in build time.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6433
Test Plan: Build release, all and jtest. Try to build with ROCKSDB_NAMESPACE with another flag.
Differential Revision: D19977691
fbshipit-source-id: aa7f2d0972e1c31d75339ac48478f34f6cfcfb3e
Summary:
With many millions of keys, the old Bloom filter implementation
for the block-based table (format_version <= 4) would have excessive FP
rate due to the limitations of feeding the Bloom filter with a 32-bit hash.
This change computes an estimated inflated FP rate due to this effect
and warns in the log whenever an SST filter is constructed (almost
certainly a "full" not "partitioned" filter) that exceeds 1.5x FP rate
due to this effect. The detailed condition is only checked if 3 million
keys or more have been added to a filter, as this should be a lower
bound for common bits/key settings (< 20).
Recommended remedies include smaller SST file size, using
format_version >= 5 (for new Bloom filter), or using partitioned
filters.
This does not change behavior other than generating warnings for some
constructed filters using the old implementation.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6317
Test Plan:
Example with warning, 15M keys @ 15 bits / key: (working_mem_size_mb is just to stop after building one filter if it's large)
$ ./filter_bench -quick -impl=0 -working_mem_size_mb=1 -bits_per_key=15 -average_keys_per_filter=15000000 2>&1 | grep 'FP rate'
[WARN] [/block_based/filter_policy.cc:292] Using legacy SST/BBT Bloom filter with excessive key count (15.0M @ 15bpk), causing estimated 1.8x higher filter FP rate. Consider using new Bloom with format_version>=5, smaller SST file size, or partitioned filters.
Predicted FP rate %: 0.766702
Average FP rate %: 0.66846
Example without warning (150K keys):
$ ./filter_bench -quick -impl=0 -working_mem_size_mb=1 -bits_per_key=15 -average_keys_per_filter=150000 2>&1 | grep 'FP rate'
Predicted FP rate %: 0.422857
Average FP rate %: 0.379301
$
With more samples at 15 bits/key:
150K keys -> no warning; actual: 0.379% FP rate (baseline)
1M keys -> no warning; actual: 0.396% FP rate, 1.045x
9M keys -> no warning; actual: 0.563% FP rate, 1.485x
10M keys -> warning (1.5x); actual: 0.564% FP rate, 1.488x
15M keys -> warning (1.8x); actual: 0.668% FP rate, 1.76x
25M keys -> warning (2.4x); actual: 0.880% FP rate, 2.32x
At 10 bits/key:
150K keys -> no warning; actual: 1.17% FP rate (baseline)
1M keys -> no warning; actual: 1.16% FP rate
10M keys -> no warning; actual: 1.32% FP rate, 1.13x
25M keys -> no warning; actual: 1.63% FP rate, 1.39x
35M keys -> warning (1.6x); actual: 1.81% FP rate, 1.55x
At 5 bits/key:
150K keys -> no warning; actual: 9.32% FP rate (baseline)
25M keys -> no warning; actual: 9.62% FP rate, 1.03x
200M keys -> no warning; actual: 12.2% FP rate, 1.31x
250M keys -> warning (1.5x); actual: 12.8% FP rate, 1.37x
300M keys -> warning (1.6x); actual: 13.4% FP rate, 1.43x
The reason for the modest inaccuracy at low bits/key is that the assumption of independence between a collision between 32-hash values feeding the filter and an FP in the filter is not quite true for implementations using "simple" logic to compute indices from the stock hash result. There's math on this in my dissertation, but I don't think it's worth the effort just for these extreme cases (> 100 million keys and low-ish bits/key).
Differential Revision: D19471715
Pulled By: pdillinger
fbshipit-source-id: f80c96893a09bf1152630ff0b964e5cdd7e35c68
Summary:
There's no technological impediment to allowing the Bloom
filter bits/key to be non-integer (fractional/decimal) values, and it
provides finer control over the memory vs. accuracy trade-off. This is
especially handy in using the format_version=5 Bloom filter in place
of the old one, because bits_per_key=9.55 provides the same accuracy as
the old bits_per_key=10.
This change not only requires refining the logic for choosing the best
num_probes for a given bits/key setting, it revealed a flaw in that logic.
As bits/key gets higher, the best num_probes for a cache-local Bloom
filter is closer to bpk / 2 than to bpk * 0.69, the best choice for a
standard Bloom filter. For example, at 16 bits per key, the best
num_probes is 9 (FP rate = 0.0843%) not 11 (FP rate = 0.0884%).
This change fixes and refines that logic (for the format_version=5
Bloom filter only, just in case) based on empirical tests to find
accuracy inflection points between each num_probes.
Although bits_per_key is now specified as a double, the new Bloom
filter converts/rounds this to "millibits / key" for predictable/precise
internal computations. Just in case of unforeseen compatibility
issues, we round to the nearest whole number bits / key for the
legacy Bloom filter, so as not to unlock new behaviors for it.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6092
Test Plan: unit tests included
Differential Revision: D18711313
Pulled By: pdillinger
fbshipit-source-id: 1aa73295f152a995328cb846ef9157ae8a05522a
Summary:
Adds an improved, replacement Bloom filter implementation (FastLocalBloom) for full and partitioned filters in the block-based table. This replacement is faster and more accurate, especially for high bits per key or millions of keys in a single filter.
Speed
The improved speed, at least on recent x86_64, comes from
* Using fastrange instead of modulo (%)
* Using our new hash function (XXH3 preview, added in a previous commit), which is much faster for large keys and only *slightly* slower on keys around 12 bytes if hashing the same size many thousands of times in a row.
* Optimizing the Bloom filter queries with AVX2 SIMD operations. (Added AVX2 to the USE_SSE=1 build.) Careful design was required to support (a) SIMD-optimized queries, (b) compatible non-SIMD code that's simple and efficient, (c) flexible choice of number of probes, and (d) essentially maximized accuracy for a cache-local Bloom filter. Probes are made eight at a time, so any number of probes up to 8 is the same speed, then up to 16, etc.
* Prefetching cache lines when building the filter. Although this optimization could be applied to the old structure as well, it seems to balance out the small added cost of accumulating 64 bit hashes for adding to the filter rather than 32 bit hashes.
Here's nominal speed data from filter_bench (200MB in filters, about 10k keys each, 10 bits filter data / key, 6 probes, avg key size 24 bytes, includes hashing time) on Skylake DE (relatively low clock speed):
$ ./filter_bench -quick -impl=2 -net_includes_hashing # New Bloom filter
Build avg ns/key: 47.7135
Mixed inside/outside queries...
Single filter net ns/op: 26.2825
Random filter net ns/op: 150.459
Average FP rate %: 0.954651
$ ./filter_bench -quick -impl=0 -net_includes_hashing # Old Bloom filter
Build avg ns/key: 47.2245
Mixed inside/outside queries...
Single filter net ns/op: 63.2978
Random filter net ns/op: 188.038
Average FP rate %: 1.13823
Similar build time but dramatically faster query times on hot data (63 ns to 26 ns), and somewhat faster on stale data (188 ns to 150 ns). Performance differences on batched and skewed query loads are between these extremes as expected.
The only other interesting thing about speed is "inside" (query key was added to filter) vs. "outside" (query key was not added to filter) query times. The non-SIMD implementations are substantially slower when most queries are "outside" vs. "inside". This goes against what one might expect or would have observed years ago, as "outside" queries only need about two probes on average, due to short-circuiting, while "inside" always have num_probes (say 6). The problem is probably the nastily unpredictable branch. The SIMD implementation has few branches (very predictable) and has pretty consistent running time regardless of query outcome.
Accuracy
The generally improved accuracy (re: Issue https://github.com/facebook/rocksdb/issues/5857) comes from a better design for probing indices
within a cache line (re: Issue https://github.com/facebook/rocksdb/issues/4120) and improved accuracy for millions of keys in a single filter from using a 64-bit hash function (XXH3p). Design details in code comments.
Accuracy data (generalizes, except old impl gets worse with millions of keys):
Memory bits per key: FP rate percent old impl -> FP rate percent new impl
6: 5.70953 -> 5.69888
8: 2.45766 -> 2.29709
10: 1.13977 -> 0.959254
12: 0.662498 -> 0.411593
16: 0.353023 -> 0.0873754
24: 0.261552 -> 0.0060971
50: 0.225453 -> ~0.00003 (less than 1 in a million queries are FP)
Fixes https://github.com/facebook/rocksdb/issues/5857
Fixes https://github.com/facebook/rocksdb/issues/4120
Unlike the old implementation, this implementation has a fixed cache line size (64 bytes). At 10 bits per key, the accuracy of this new implementation is very close to the old implementation with 128-byte cache line size. If there's sufficient demand, this implementation could be generalized.
Compatibility
Although old releases would see the new structure as corrupt filter data and read the table as if there's no filter, we've decided only to enable the new Bloom filter with new format_version=5. This provides a smooth path for automatic adoption over time, with an option for early opt-in.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6007
Test Plan: filter_bench has been used thoroughly to validate speed, accuracy, and correctness. Unit tests have been carefully updated to exercise new and old implementations, as well as the logic to select an implementation based on context (format_version).
Differential Revision: D18294749
Pulled By: pdillinger
fbshipit-source-id: d44c9db3696e4d0a17caaec47075b7755c262c5f
Summary:
Some recent commits might not have passed through the formatter. I formatted recent 45 commits. The script hangs for more commits so I stopped there.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5827
Test Plan: Run all existing tests.
Differential Revision: D17483727
fbshipit-source-id: af23113ee63015d8a43d89a3bc2c1056189afe8f
Summary:
Refactoring to consolidate implementation details of legacy
Bloom filters. This helps to organize and document some related,
obscure code.
Also added make/cpp var TEST_CACHE_LINE_SIZE so that it's easy to
compile and run unit tests for non-native cache line size. (Fixed a
related test failure in db_properties_test.)
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5784
Test Plan:
make check, including Recently added Bloom schema unit tests
(in ./plain_table_db_test && ./bloom_test), and including with
TEST_CACHE_LINE_SIZE=128U and TEST_CACHE_LINE_SIZE=256U. Tested the
schema tests with temporary fault injection into new implementations.
Some performance testing with modified unit tests suggest a small to moderate
improvement in speed.
Differential Revision: D17381384
Pulled By: pdillinger
fbshipit-source-id: ee42586da996798910fc45ac0b6289147f16d8df
Summary:
This will allow us to fix history by having the code changes for PR#5784 properly attributed to it.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5810
Differential Revision: D17400231
Pulled By: pdillinger
fbshipit-source-id: 2da8b1cdf2533cfedb35b5526eadefb38c291f09
