ToplingDB vs RocksDB bench results

dcompact bench → — offloads most CPU and memory cost.

Updated (UTC): 2026-09-15 07:18:26+00:00

Latest: plain

run_id=34935695406 | result table | 2026-09-15 07:18:23+00:00

raw logs: ToplingDB zipkeyonly | ToplingDB zipkeyvalue | RocksDB v8.10 | RocksDB master (db114777) | 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

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.

• On readrandom (the fairest comparison), ToplingDB zipkeyvalue is 18% memory / 4.98x speed vs RocksDB v8.10.
• In production the working set typically does not fit, so cache misses dominate. The same RAM holds more ToplingDB data, which cuts I/O and widens the gap.
• ToplingDB dcompact can offload most of that CPU and memory cost.

/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.

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).

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.

• rss: resident set size (pages currently in RAM for the process); rss = shared + anony.
• shared: shared resident pages; mostly readonly (cheap; OS prefers reclaiming these, no swap needed).
• anony: rss - shared; mostly readwrite anonymous pages (costly, needs swap).
• pagecache: kernel file page cache for regular files this process currently has open (cachestat(2) on /proc/pid/fd, deduped by inode; covers buffered read/write and mmap). Pages brought in only via buffered I/O are not charged to process RSS; mmap'd file pages may appear in both pagecache and RSS/shared, so the series can overlap. When reads hit this cache, they avoid disk I/O, and the DB benefits in its own way.
• anony+pc: anony+pagecache (sum of process anonymous RSS and open-file page cache; not a disjoint partition).

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.

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.

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.

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

scan-omit-value on data from fillseq

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

memtablerep_bench: OffsetSkipList and CSPP vs skiplist

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

History

• 2026-09-15 07:18:23+00:00 — plain run_id=34935695406 — result table
• 2026-09-14 06:56:05+00:00 — plain run_id=34809394323 — result table
• 2026-09-13 18:28:04+00:00 — plain run_id=34770497158 — result table
• 2026-09-13 00:02:25+00:00 — plain run_id=34722719277 — result table
• 2026-09-12 19:11:46+00:00 — plain run_id=34708681171 — result table
• 2026-09-12 15:43:43+00:00 — plain run_id=34698861539 — result table
• 2026-08-31 15:48:30+00:00 — plain run_id=33404363930 — result table
• 2026-08-30 11:14:25+00:00 — plain run_id=33306036311 — result table
• 2026-08-23 16:20:05+00:00 — plain run_id=32643128750 — result table
• 2026-08-18 15:53:23+00:00 — plain run_id=32148249444 — result table
• 2026-08-16 17:28:08+00:00 — plain run_id=31957895530 — result table
• 2026-08-16 04:43:40+00:00 — plain run_id=31921183844 — result table
• 2026-08-16 03:27:00+00:00 — plain run_id=31918647193 — result table
• 2026-08-16 02:07:06+00:00 — plain run_id=31914258669 — result table
• 2026-08-16 00:41:41+00:00 — plain run_id=31914256719 — result table
• 2026-08-15 18:00:11+00:00 — plain run_id=31896896755 — result table
• 2026-08-15 14:01:09+00:00 — plain run_id=31884049900 — result table
• 2026-08-15 13:08:09+00:00 — dcompact run_id=31880805602 — result table
• 2026-08-15 11:25:39+00:00 — plain run_id=31873946138 — result table
• 2026-08-15 10:07:47+00:00 — dcompact run_id=31873910431 — result table
• 2026-08-15 06:19:56+00:00 — plain run_id=31866312589 — result table
• 2026-08-14 15:45:19+00:00 — dcompact run_id=31800981467 — result table
• 2026-08-14 14:00:45+00:00 — plain run_id=31800922314 — result table
• 2026-08-14 12:01:06+00:00 — dcompact run_id=31787128815 — result table
• 2026-08-14 10:14:47+00:00 — plain run_id=31787125871 — result table
• 2026-08-09 17:28:01+00:00 — dcompact run_id=31324418704 — result table
• 2026-08-09 10:21:07+00:00 — dcompact run_id=31303589765 — result table
• 2026-08-09 09:29:35+00:00 — dcompact run_id=31298143542 — result table
• 2026-08-09 07:40:13+00:00 — plain run_id=31298142068 — result table
• 2026-08-08 13:03:29+00:00 — plain run_id=31252003395 — result table
• 2026-08-08 11:48:09+00:00 — dcompact run_id=31251963803 — result table
• 2026-08-08 09:12:52+00:00 — plain run_id=31247301157 — result table
• 2026-08-08 03:43:31+00:00 — plain run_id=31234948599 — result table
• 2026-08-07 23:42:24+00:00 — plain run_id=31220758079 — result table
• 2026-08-07 22:21:52+00:00 — plain run_id=31218150702 — result table
• 2026-08-07 17:54:24+00:00 — plain run_id=31199586368 — result table
• 2026-08-07 16:10:39+00:00 — plain run_id=31191036555 — result table