ToplingDB vs RocksDB bench results

dcompact bench → — offloads most CPU and memory cost.

Updated (UTC): 2026-10-08 09:56:31+00:00

Latest: plain

run_id=37752667561 | result table | 2026-10-08 09:56:30+00:00

raw logs: ToplingDB zipkeyonly | ToplingDB zipkeyvalue | RocksDB v8.10 | RocksDB master (dec67990) | DB INFO LOGs + bench yamls (Actions artifact)

ToplingDB bench yaml: zipkeyonly db_bench-fillrandom.yaml | zipkeyonly db_bench-fillseq.yaml | zipkeyonly db_bench-fillseq-cspp.yaml | zipkeyvalue db_bench-fillrandom.yaml | zipkeyvalue db_bench-fillseq.yaml | zipkeyvalue db_bench-fillseq-cspp.yaml

Runner hardware/software

OSUbuntu 24.04.5 LTS
Kernel6.17.0-1022-azure
CPUAMD EPYC 7763 64-Core Processor
Logical CPUs4
Memory15.6GiB
/dev/shm11.7GiB
RocksDB block cache7.8GiB

num=50000000 | key_size=8 | value_size=50

On-disk DB size (1.9GiB) ≤ block cache (7.8GiB) — the dataset fits in cache, so this bench is a CPU and memory comparison.

/dev/shm usage (disk space; after db_bench)

Allocated disk usage (IEC blocks). zipkeyonly does not compress values (speed-optimized). RocksDB uses per-level compression (L0 none, L1-L5 Snappy, L6 Zstd), corresponding to the ToplingDB zipkeyvalue variant's level_writers (lightweight upper levels, heavyweight L6). Space ratio = engine / v8.10. color sign: ToplingDB is Better/Worse.

workloadToplingDB zipkeyonlyToplingDB zipkeyvalueRocksDB v8.10RocksDB master (dec67990)zipkeyonly / v8.10 (space)zipkeyvalue / v8.10 (space)
fillrandom suite2.0GiB736.0MiB1.9GiB1.9GiB1.07x0.39x
fillseq suite3.2GiB1.4GiB1.6GiB1.6GiB1.98x0.87x
fillseq suite (CSPP)2.9GiB1.1GiBn/an/a——

Peak RSS (RAM; during db_bench)

RSS is Resident Set Size. RocksDB block cache = half physical memory (7.8GiB). Ratio = engine / v8.10. color sign: ToplingDB is Better/Worse. RocksDB omit cells are =readseq (same scan; no omit). scan-omit-value: restart process, reuse fill* data, scan without access value, benefited by lazy load value (ToplingDB feature).

workloadToplingDB zipkeyonlyToplingDB zipkeyvalueRocksDB v8.10RocksDB master (dec67990)zipkeyonly / v8.10 (RSS)zipkeyvalue / v8.10 (RSS)
fillrandom suite peak2.3GiB2.5GiB2.1GiB2.1GiB1.11x1.18x
fillrandom suite readrandom138.7MiB391.8MiB2.1GiB2.1GiB0.07x0.19x
fillrandom scan-omit-value127.7MiB139.5MiB2.1GiB (=readseq)2.1GiB (=readseq)0.06x0.06x
fillseq suite peak1.9GiB1.9GiB3.2GiB3.3GiB0.59x0.60x
fillseq suite readrandom140.9MiB751.6MiB3.2GiB3.3GiB0.04x0.23x
fillseq scan-omit-value119.7MiB142.7MiB3.2GiB (=readseq)3.3GiB (=readseq)0.04x0.04x
fillseq suite (CSPP) peak2.1GiB2.6GiBn/an/a——
fillseq suite (CSPP) readrandom147.3MiB707.0MiBn/an/a——
fillseq scan-omit-value (CSPP)119.9MiB143.2MiBn/an/a——

RAM usage over time

Sampled once per second from /proc/statm plus open-file page cache. The bar below each plot shows stage name + duration (consecutive repeats merge, e.g. readseq×3); colored bands mark the same intervals. Stage names may overflow into a neighbor cell; a label under the bar is used only when names would overlap.

startup 2.0 fillrandom 28 flush 0.0 compact 26 readseq×3 5.9 readrandom 8.6 0 500 1000 1500 2000 2500 3000 3500 4000 MiB 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 Time (sec) ToplingDB zipkeyonly — fillrandom suite RAM rss shared anony pagecache anony+pc
startup 2.0 fillseq 14 flush 0.0 compact 21 readseq×3 4.9 readrandom 8.1 0 500 1000 1500 2000 2500 3000 3500 4000 MiB 0 5 10 15 20 25 30 35 40 45 50 55 60 Time (sec) ToplingDB zipkeyonly — fillseq suite RAM rss shared anony pagecache anony+pc
startup 2.0 fillseq 15 flush 0.1 compact 20 readseq×3 4.9 readrandom 8.1 0 500 1000 1500 2000 2500 3000 3500 4000 MiB 0 5 10 15 20 25 30 35 40 45 50 55 60 Time (sec) ToplingDB zipkeyonly — fillseq-cspp suite RAM rss shared anony pagecache anony+pc
startup 2.0 fillrandom 27 flush 0.0 compact 35 readseq×3 8.8 readrandom 21 0 500 1000 1500 2000 2500 3000 3500 4000 MiB 0 50 100 Time (sec) ToplingDB zipkeyvalue — fillrandom suite RAM rss shared anony pagecache anony+pc
startup 2.0 fillseq 13 flush 0.0 compact 63 readseq×3 7.3 readrandom 29 0 500 1000 1500 2000 2500 3000 3500 4000 MiB 0 50 100 150 200 Time (sec) ToplingDB zipkeyvalue — fillseq suite RAM rss shared anony pagecache anony+pc
startup 2.0 fillseq 15 flush 0.0 compact 62 readseq×3 7.3 readrandom 28 0 500 1000 1500 2000 2500 3000 3500 4000 MiB 0 50 100 150 200 Time (sec) ToplingDB zipkeyvalue — fillseq-cspp suite RAM rss shared anony pagecache anony+pc
fillrandom 62 flush 0.2 compact 30 readseq×3 10 readrandom 125 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 MiB 0 50 100 150 200 250 300 Time (sec) RocksDB v8.10 — fillrandom suite RAM rss shared anony pagecache anony+pc
fillseq 20 flush 0.2 compact 19 readseq×3 17 readrandom 135 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 MiB 0 50 100 150 200 Time (sec) RocksDB v8.10 — fillseq suite RAM rss shared anony pagecache anony+pc
fillrandom 71 flush 0.3 compact 37 readseq×3 16 readrandom 148 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 MiB 0 50 100 150 200 250 300 Time (sec) RocksDB master (dec67990) — fillrandom suite RAM rss shared anony pagecache anony+pc
fillseq 20 flush 0.2 compact 21 readseq×3 26 readrandom 139 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 MiB 0 50 100 150 200 250 300 Time (sec) RocksDB master (dec67990) — fillseq suite RAM rss shared anony pagecache anony+pc

Comparison: db_bench fillrandom suite (perf)

Benchmarks: fillrandom, flush, compact, readseq×3, readrandom. RocksDB uses per-level compression (L0 none, L1-L5 Snappy, L6 Zstd), corresponding to the ToplingDB zipkeyvalue variant's level_writers (lightweight upper levels, heavyweight L6). compact row shows operations/time. color sign: ToplingDB is Better/Worse.

benchmarkToplingDB zipkeyonly ops/secToplingDB zipkeyvalue ops/secRocksDB v8.10 ops/secRocksDB master (dec67990) ops/seczipkeyonly / zipkeyvaluezipkeyonly / v8.10zipkeyonly / master (dec67990)zipkeyvalue / v8.10zipkeyvalue / master (dec67990)
fillrandom178914018695438013137000540.96x2.23x2.56x2.33x2.67x
compact1/26.0s1/34.9s1/30.2s1/37.4s1.34x1.16x1.44x0.87x1.07x
readseq25635898171153861075154767290481.50x2.38x3.81x1.59x2.54x
readrandom583000623999844003163388912.43x14.56x17.20x6.00x7.08x

Comparison: db_bench fillseq suite (perf)

Same as fillrandom, except ToplingDB fillseq uses OffsetSkipList (fillrandom still uses CSPP). RocksDB fillseq benefits from shortcuts: trivial_move on non-overlapping SSTs; refit level skips zstd on L6: faster, larger size. Seqno-zeroing compact still runs.

benchmarkToplingDB zipkeyonly ops/secToplingDB zipkeyvalue ops/secRocksDB v8.10 ops/secRocksDB master (dec67990) ops/seczipkeyonly / zipkeyvaluezipkeyonly / v8.10zipkeyonly / master (dec67990)zipkeyvalue / v8.10zipkeyvalue / master (dec67990)
fillseq36788523810866254220825376130.97x1.45x1.45x1.50x1.50x
compact1/21.1s1/63.4s1/19.0s1/21.1s3.00x0.90x1.00x0.30x0.33x
readseq30530398203796811071993367545251.50x2.85x4.52x1.90x3.02x
readrandom621025317515463705593590953.55x16.76x17.29x4.73x4.88x

Comparison: db_bench fillseq suite (CSPP) (perf)

Same as fillrandom, except ToplingDB fillseq uses CSPP. RocksDB fillseq benefits from shortcuts: trivial_move on non-overlapping SSTs; refit level skips zstd on L6: faster, larger size. Seqno-zeroing compact still runs.

benchmarkToplingDB zipkeyonly ops/secToplingDB zipkeyvalue ops/secRocksDB v8.10 ops/secRocksDB master (dec67990) ops/seczipkeyonly / zipkeyvaluezipkeyonly / v8.10zipkeyonly / master (dec67990)zipkeyvalue / v8.10zipkeyvalue / master (dec67990)
fillseq33251093447608254220825376130.96x1.31x1.31x1.36x1.36x
compact1/19.8s1/62.0s1/19.0s1/21.1s3.13x0.96x1.06x0.31x0.34x
readseq30673943206027451071993367545251.49x2.86x4.54x1.92x3.05x
readrandom613529917641093705593590953.48x16.56x17.09x4.76x4.91x

Lazy load demo (scan; RocksDB v8.10 baseline)

zipkey* needs an extra omit pass: scan_omit_key/value enables lazy value load (no real value load). RocksDB has no lazy load, so the baseline is readseq×3 already present in the main fill* suite (no extra pass). RocksDB nextwithkey cells are =readseq. master omitted here (v8.10 is the stronger RocksDB baseline). color sign: ToplingDB is Better/Worse.

scan-omit-value on data from fillrandom

benchmarkToplingDB zipkeyonly ops/secToplingDB zipkeyvalue ops/secRocksDB v8.10 ops/seczipkeyonly / v8.10zipkeyvalue / v8.10
readseq3780372438606878107515473.52x3.59x
nextwithkey397465443992143410751547 (=readseq)3.70x3.71x

scan-omit-value on data from fillseq

benchmarkToplingDB zipkeyonly ops/secToplingDB zipkeyvalue ops/secRocksDB v8.10 ops/seczipkeyonly / v8.10zipkeyvalue / v8.10
readseq5133823350831553107199334.79x4.74x
nextwithkey523165785152769310719933 (=readseq)4.88x4.81x

scan-omit-value on data from fillseq (CSPP)

benchmarkToplingDB zipkeyonly ops/secToplingDB zipkeyvalue ops/secRocksDB v8.10 ops/seczipkeyonly / v8.10zipkeyvalue / v8.10
readseq5078173450779258107199334.74x4.74x
nextwithkey521572775172642110719933 (=readseq)4.87x4.83x

memtablerep_bench: OffsetSkipList and CSPP vs skiplist

Focus: OffsetSkipList / CSPP (ToplingDB) vs skiplist. Baseline = RocksDB v8.10 skiplist. color sign: ToplingDB is Better/Worse.

benchmarkmetricOffsetSkipList (ToplingDB)CSPP (ToplingDB)skiplist (ToplingDB)skiplist (RocksDB v8.10)OffsetSkipList / v8.10CSPP / v8.10
fillrandomElapsed time243443593 us (2.43444 us/op)25159609 us (0.251596 us/op)274530713 us (2.74531 us/op)300764319 us (3.00764 us/op)0.81x0.08x
fillrandomWrite throughput6.65964 MiB/s64.4385 MiB/s5.90552 MiB/s5.39042 MiB/s1.24x11.95x
readrandomElapsed time263408026 us (2.63408 us/op)33533229 us (0.335332 us/op)325868860 us (3.25869 us/op)325682295 us (3.25682 us/op)0.81x0.10x
readrandomRead throughput6.15489 MiB/s48.3475 MiB/s4.97515 MiB/s4.978 MiB/s1.24x9.71x

History