Implement XXH3 block checksum type #9069

pdillinger · 2021-10-23T21:23:08Z

Summary: XXH3 - latest hash function that is extremely fast on large
data, easily faster than crc32c on most any x86_64 hardware. In
integrating this hash function, I have handled the compression type byte
in a non-standard way to avoid using the streaming API (extra data
movement and active code size because of hash function complexity). This
approach got a thumbs-up from Yann Collet.

Existing functionality change:

reject bad ChecksumType in options with InvalidArgument

This change split off from #9058 because context-aware checksum is
likely to be handled through different configuration than ChecksumType.

Test Plan: tests updated, and substantially expanded. Unit tests now check
that we don't accidentally change the values generated by the checksum
algorithms ("schema test") and that we properly handle
invalid/unrecognized checksum types in options or in file footer.

DBTestBase::ChangeOptions (etc.) updated from two to one configuration
changing from default CRC32c ChecksumType. The point of this test code
is to detect possible interactions among features, and the likelihood of
some bad interaction being detected by including configurations other
than XXH3 and CRC32c--and then not detected by stress/crash test--is
extremely low.

Stress/crash test also updated (manual run long enough to see it accepts
new checksum type). db_bench also updated for microbenchmarking
checksums.

Performance microbenchmark (PORTABLE=0 DEBUG_LEVEL=0, Broadwell processor)

./db_bench -benchmarks=crc32c,xxhash,xxhash64,xxh3,crc32c,xxhash,xxhash64,xxh3,crc32c,xxhash,xxhash64,xxh3
crc32c : 0.200 micros/op 5005220 ops/sec; 19551.6 MB/s (4096 per op)
xxhash : 0.807 micros/op 1238408 ops/sec; 4837.5 MB/s (4096 per op)
xxhash64 : 0.421 micros/op 2376514 ops/sec; 9283.3 MB/s (4096 per op)
xxh3 : 0.171 micros/op 5858391 ops/sec; 22884.3 MB/s (4096 per op)
crc32c : 0.206 micros/op 4859566 ops/sec; 18982.7 MB/s (4096 per op)
xxhash : 0.793 micros/op 1260850 ops/sec; 4925.2 MB/s (4096 per op)
xxhash64 : 0.410 micros/op 2439182 ops/sec; 9528.1 MB/s (4096 per op)
xxh3 : 0.161 micros/op 6202872 ops/sec; 24230.0 MB/s (4096 per op)
crc32c : 0.203 micros/op 4924686 ops/sec; 19237.1 MB/s (4096 per op)
xxhash : 0.839 micros/op 1192388 ops/sec; 4657.8 MB/s (4096 per op)
xxhash64 : 0.424 micros/op 2357391 ops/sec; 9208.6 MB/s (4096 per op)
xxh3 : 0.162 micros/op 6182678 ops/sec; 24151.1 MB/s (4096 per op)

As you can see, especially once warmed up, xxh3 is fastest.

Performance macrobenchmark (PORTABLE=0 DEBUG_LEVEL=0, Broadwell processor)

Test

for I in `seq 1 50`; do for CHK in 0 1 2 3 4; do TEST_TMPDIR=/dev/shm/rocksdb$CHK ./db_bench -benchmarks=fillseq -memtablerep=vector -allow_concurrent_memtable_write=false -num=30000000 -checksum_type=$CHK 2>&1 | grep 'micros/op' | tee -a results-$CHK & done; wait; done

Results (ops/sec)

for FILE in results*; do echo -n "$FILE "; awk '{ s += $5; c++; } END { print 1.0 * s / c; }' < $FILE; done

results-0 252118 # kNoChecksum
results-1 251588 # kCRC32c
results-2 251863 # kxxHash
results-3 252016 # kxxHash64
results-4 252038 # kXXH3

Summary: XXH3 - latest hash function that is extremely fast on large data, easily faster than crc32c on most any x86_64 hardware. In integrating this hash function, I have handled the compression type byte in a non-standard way to avoid using the streaming API (extra data movement and active code size because of hash function complexity). This approach got a thumbs-up from Yann Collet. This change split off from facebook#9058 because context-aware checksum is likely to be handled through different configuration than ChecksumType. Test Plan: tests updated, including stress test. db_bench also updated for microbenchmarking checksums. ### Performance microbenchmark (PORTABLE=0 DEBUG_LEVEL=0, Broadwell processor) ./db_bench -benchmarks=crc32c,xxhash,xxhash64,xxh3,crc32c,xxhash,xxhash64,xxh3,crc32c,xxhash,xxhash64,xxh3 crc32c : 0.200 micros/op 5005220 ops/sec; 19551.6 MB/s (4096 per op) xxhash : 0.807 micros/op 1238408 ops/sec; 4837.5 MB/s (4096 per op) xxhash64 : 0.421 micros/op 2376514 ops/sec; 9283.3 MB/s (4096 per op) xxh3 : 0.171 micros/op 5858391 ops/sec; 22884.3 MB/s (4096 per op) crc32c : 0.206 micros/op 4859566 ops/sec; 18982.7 MB/s (4096 per op) xxhash : 0.793 micros/op 1260850 ops/sec; 4925.2 MB/s (4096 per op) xxhash64 : 0.410 micros/op 2439182 ops/sec; 9528.1 MB/s (4096 per op) xxh3 : 0.161 micros/op 6202872 ops/sec; 24230.0 MB/s (4096 per op) crc32c : 0.203 micros/op 4924686 ops/sec; 19237.1 MB/s (4096 per op) xxhash : 0.839 micros/op 1192388 ops/sec; 4657.8 MB/s (4096 per op) xxhash64 : 0.424 micros/op 2357391 ops/sec; 9208.6 MB/s (4096 per op) xxh3 : 0.162 micros/op 6182678 ops/sec; 24151.1 MB/s (4096 per op) As you can see, especially once warmed up, xxh3 is fastest. ### Performance macrobenchmark (PORTABLE=0 DEBUG_LEVEL=0, Broadwell processor) Test for I in `seq 1 50`; do for CHK in 0 1 2 3 4; do TEST_TMPDIR=/dev/shm/rocksdb$CHK ./db_bench -benchmarks=fillseq -memtablerep=vector -allow_concurrent_memtable_write=false -num=30000000 -checksum_type=$CHK 2>&1 | grep 'micros/op' | tee -a results-$CHK & done; wait; done Results (ops/sec) for FILE in results*; do echo -n "$FILE "; awk '{ s += $5; c++; } END { print 1.0 * s / c; }' < $FILE; done results-0 252118 # kNoChecksum results-1 251588 # kCRC32c results-2 251863 # kxxHash results-3 252016 # kxxHash64 results-4 252038 # kXXH3

mrambacher · 2021-10-25T13:06:28Z

db/db_test_util.cc

-      break;
-    }
-    case kxxHash64Checksum: {
-      table_options.checksum = kxxHash64;


Is there a reason these cases are no longer part of the test? Can you add a comment somewhere (either in the code or the PR message) as to why this change was made?

Description updated (in test plan)

mrambacher · 2021-10-25T13:14:24Z

tools/db_bench_tool.cc

+      block_based_options.checksum =
+          static_cast<ChecksumType>(FLAGS_checksum_type);


Should there be (either here or using a Validator) some check that the checksum_type is a valid type?

No, that should be in BlockBasedTableFactory but is not.

mrambacher · 2021-10-25T15:21:49Z

options/options_helper.cc

+std::vector<ChecksumType> GetSupportedChecksums() {
+  std::vector<ChecksumType> checksum_types;


Just asking if this should be a set or vector? Wonder if at some point the same enum might have two names...

Future-proofing this and copied code

mrambacher

The db_stress_tool and fault_inject_fs also deal with checksum types. Do those classes also need to be updated for this PR?

pdillinger · 2021-10-25T18:57:22Z

The db_stress_tool and fault_inject_fs also deal with checksum types. Do those classes also need to be updated for this PR?

db_crashtest.py is updated; db_stress uses standard option parsing (which is updated). fault_injection_fs deals with "handoff checksums" which we will likely only use with CRCs because they are extendable (from block checksum to checksum of block+trailer). 32-bit xxhash is in the test to demonstrate theoretical capability, so XXH3 wouldn't add any material value to the test at this time.

facebook-github-bot · 2021-10-25T19:23:18Z

@pdillinger has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

mrambacher

Minor comments. LGTM!

mrambacher · 2021-10-28T21:36:17Z

table/table_test.cc

+  sink.contents_[len - Footer::kNewVersionsEncodedLength] = '\x7b';
+


Please add a comment that this is corrupting the checksum type to "123.*" for line 2231 below

mrambacher · 2021-10-28T21:38:51Z

tools/db_bench_tool.cc

+DEFINE_int32(checksum_type,
+             ROCKSDB_NAMESPACE::BlockBasedTableOptions().checksum,
+             "ChecksumType as an int");


nit: Should this be an int or one of the string values of the checksum enum?

Next person can add it if they want

facebook-github-bot · 2021-10-28T22:24:24Z

@pdillinger has updated the pull request. You must reimport the pull request before landing.

facebook-github-bot · 2021-10-28T22:31:20Z

@pdillinger has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

facebook-github-bot · 2021-10-29T05:15:24Z

@pdillinger merged this pull request in a7d4bea.

facebook-github-bot added the CLA Signed label Oct 23, 2021

mrambacher reviewed Oct 25, 2021

View reviewed changes

Significant test updates, supporting refactoring + minor tweak

aa15658

Add to HISTORY

492e9f2

mrambacher approved these changes Oct 28, 2021

View reviewed changes

pdillinger added 2 commits October 28, 2021 15:12

Address review feedback

bf65cb9

Merge remote-tracking branch 'origin/main' into xxh3_checksum

275af49

facebook-github-bot closed this in a7d4bea Oct 29, 2021

facebook-github-bot added the Merged label Oct 29, 2021

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Implement XXH3 block checksum type #9069

Implement XXH3 block checksum type #9069

pdillinger commented Oct 23, 2021 •

edited

Loading

mrambacher Oct 25, 2021

pdillinger Oct 25, 2021

mrambacher Oct 25, 2021

pdillinger Oct 25, 2021

mrambacher Oct 25, 2021

pdillinger Oct 25, 2021

mrambacher left a comment

pdillinger commented Oct 25, 2021

facebook-github-bot commented Oct 25, 2021

mrambacher left a comment

mrambacher Oct 28, 2021

mrambacher Oct 28, 2021

pdillinger Oct 28, 2021

facebook-github-bot commented Oct 28, 2021

facebook-github-bot commented Oct 28, 2021

facebook-github-bot commented Oct 29, 2021

		block_based_options.checksum =
		static_cast<ChecksumType>(FLAGS_checksum_type);

		std::vector<ChecksumType> GetSupportedChecksums() {
		std::vector<ChecksumType> checksum_types;

		sink.contents_[len - Footer::kNewVersionsEncodedLength] = '\x7b';

Implement XXH3 block checksum type #9069

Implement XXH3 block checksum type #9069

Conversation

pdillinger commented Oct 23, 2021 • edited Loading

Performance microbenchmark (PORTABLE=0 DEBUG_LEVEL=0, Broadwell processor)

Performance macrobenchmark (PORTABLE=0 DEBUG_LEVEL=0, Broadwell processor)

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mrambacher left a comment

Choose a reason for hiding this comment

pdillinger commented Oct 25, 2021

facebook-github-bot commented Oct 25, 2021

mrambacher left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

facebook-github-bot commented Oct 28, 2021

facebook-github-bot commented Oct 28, 2021

facebook-github-bot commented Oct 29, 2021

pdillinger commented Oct 23, 2021 •

edited

Loading