Skip to content

Command and memory comparison against Redis

Per-command throughput and memory growth for 60 commands and 12 data structures, Ruvio against Redis 8.10.2. The raw data behind the table is benchmark/records/x86-azure-2026-10-11.json; the performance page shows the same run with all samples.

Conditions (read before quoting any number)

Section titled “Conditions (read before quoting any number)”
  • Host: Azure VM, AMD EPYC 7763 (virtualized), 8 vCPU = 4 cores x 2 threads, 31 GiB, Ubuntu 24.04, kernel 6.17.0-azure; data disk Premium SSD v2 32 GiB (16,000 IOPS, 400 MB/s, host caching none) with ext4; OS disk Standard SSD. Native processes only (no Docker).
  • Ruvio commit 5520164; Redis 8.10.2 built from source (jemalloc 5.3.0); redis-benchmark is the 8.10.2 binary from the same build.
  • 60 commands from scripts/redis_compare_features.py run through scripts/pinned_run.py: redis-benchmark 8.10.2, 50 clients, pipeline 16, 64-byte values, five rounds with alternating order, median requests/s. Ruvio and Redis each run alone on cpu 0 (core 0, its hyper-thread idle), the load generator on cpus 2 and 4 (two other cores), one shard, memory durability.
  • one VM, one CPU model (virtualized AMD EPYC 7763), native processes only, loopback, 64-byte values, short runs, five rounds. The data disk is network-attached (an fdatasync costs about 0.8 ms); a local NVMe disk would change every always row for both servers. Server and load generator share the VM, so absolute numbers describe this host; compare the two servers with each other only.

Reproduce

Section titled “Reproduce”
Terminal window
cargo build --profile release-fast --bin ruvio
python3 scripts/redis_compare_features.py --ruvio target/release-fast/ruvio

Throughput

Section titled “Throughput”
Command Ruvio Redis Ruvio / Redis Note
GET (baseline, for scale) 1.20M/s 1.12M/s 1.08× not a new command
SET (baseline, for scale) 1.21M/s 917k/s 1.32× not a new command
LPOS (100 elements, hit in the middle) 817k/s 548k/s 1.49×
LSET (100 elements) 920k/s 826k/s 1.11×
LINSERT (pivot missing, scans 100) 904k/s 480k/s 1.88×
LREM (value missing, scans 100) 896k/s 500k/s 1.79×
LMOVE (rotate a 100-element list) 1.03M/s 730k/s 1.41×
BITFIELD INCRBY u8 (always writes) 952k/s 809k/s 1.18× Redis refuses zero-padded offsets, so the offset is fixed
BITFIELD GET u8 1.11M/s 929k/s 1.19×
SETBIT (1 KiB bitmap, same bit) 1.24M/s 1.01M/s 1.24× no-op after the first write
GETBIT (1 KiB bitmap) 1.25M/s 1.08M/s 1.16×
BITCOUNT (1 KiB bitmap) 1.18M/s 1.08M/s 1.09×
PFADD 1.22M/s 1.05M/s 1.17×
PFCOUNT 1.27M/s 1.14M/s 1.11×
COPY … REPLACE (64 B string) 1.15M/s 708k/s 1.63×
DUMP (64 B string) 898k/s 418k/s 2.15× formats differ: Ruvio’s is not RDB
EXPIRETIME 1.26M/s 1.15M/s 1.10×
TOUCH 1.27M/s 1.15M/s 1.10×
OBJECT ENCODING 1.23M/s 1.05M/s 1.18×
MEMORY USAGE 1.25M/s 1.08M/s 1.15×
GETRANGE 1.23M/s 1000k/s 1.23×
SETRANGE 1.27M/s 987k/s 1.28×
SETEX 1.19M/s 688k/s 1.73×
MSET (3 keys) 1.06M/s 610k/s 1.74×
MGET (3 keys) 1.03M/s 813k/s 1.27×
DEL (missing key) 1.23M/s 1.19M/s 1.04×
EXISTS 1.28M/s 1.14M/s 1.13×
TYPE 1.25M/s 1.11M/s 1.13×
TTL 1.22M/s 1.13M/s 1.08×
PTTL 1.21M/s 1.13M/s 1.08×
PERSIST (key without TTL) 1.29M/s 1.09M/s 1.18×
EXPIRE (key with TTL) 1.06M/s 872k/s 1.21×
APPEND (1 byte) 1.01M/s 997k/s 1.02× value grows
INCRBYFLOAT 694k/s 505k/s 1.37× Ruvio prints 15 digits
HSET (existing field) 1.08M/s 890k/s 1.21×
HGET 1.26M/s 1.02M/s 1.23×
HMGET (2 fields) 1.09M/s 901k/s 1.22×
HSTRLEN 1.28M/s 1.05M/s 1.22×
LPUSH 943k/s 901k/s 1.05× list grows
LLEN 1.27M/s 1.12M/s 1.14×
LRANGE 0 2 1.05M/s 898k/s 1.17×
SADD (existing member) 1.08M/s 1.02M/s 1.06×
SISMEMBER 1.27M/s 1.01M/s 1.26×
SRANDMEMBER 1.02M/s 997k/s 1.03×
ZADD (existing member, same score) 1.12M/s 917k/s 1.22×
ZINCRBY 829k/s 662k/s 1.25×
ZSCORE 1.20M/s 1.01M/s 1.19×
RENAME (same key) 1.26M/s 1.01M/s 1.25×
PUBLISH (no subscribers) 1.28M/s 1.28M/s 1.00×
HGET (10-field hash) 1.20M/s 968k/s 1.24×
HSET (10-field hash, existing field) 1.02M/s 852k/s 1.19×
HDEL (10-field hash, missing field) 1.04M/s 990k/s 1.05×
HGETALL (10-field hash) 524k/s 540k/s 0.97×
LINDEX (20-item list) 1.19M/s 968k/s 1.23×
LRANGE 0 9 (20-item list) 721k/s 743k/s 0.97×
LSET (20-item list) 1.07M/s 904k/s 1.19×
GEOADD 909k/s 656k/s 1.39×
GEOSEARCH (radius 50 km, COUNT 10) 14k/s 10k/s 1.31× all points share one coordinate: worst case
GETEX (set a 100 s deadline) 971k/s 804k/s 1.21× Ruvio’s GETEX takes milliseconds only; each side uses its own syntax
FILLGET (value present) 1.11M/s 1.16M/s 0.96× Redis column: plain GET

Geometric mean of the 60 rows: 1.22×; 59 rows at or above 0.97×; lowest 0.96×.

Memory

Section titled “Memory”
Data Ruvio growth Redis growth Ruvio total RSS Redis total RSS Total ratio
20,000 lists of 20 elements 5.9 MiB 5.3 MiB 15.7 MiB 18.5 MiB 0.85×
5,000 HyperLogLogs of 100 elements 2.2 MiB 2.0 MiB 12.3 MiB 15.0 MiB 0.82×
20,000 bitmaps of 128 B (SETBIT) 3.6 MiB 3.3 MiB 13.4 MiB 16.4 MiB 0.82×
100,000 BITFIELD counters (u16) 3.5 MiB 5.1 MiB 13.6 MiB 18.3 MiB 0.75×
100,000 geo points in one key 6.3 MiB 5.7 MiB 16.2 MiB 18.9 MiB 0.86×
100,000 strings of 16 B 5.2 MiB 6.5 MiB 15.1 MiB 19.6 MiB 0.77×
100,000 strings of 128 B 17.4 MiB 19.2 MiB 27.4 MiB 32.6 MiB 0.84×
100,000 strings of 16 B with TTL 9.9 MiB 9.9 MiB 19.8 MiB 23.1 MiB 0.86×
100,000 one-element lists 4.7 MiB 6.5 MiB 14.7 MiB 19.9 MiB 0.74×
50,000 small hashes (3 fields) 3.0 MiB 3.4 MiB 12.8 MiB 16.6 MiB 0.77×
50,000 small sets (4 members) 2.6 MiB 3.1 MiB 12.7 MiB 16.3 MiB 0.78×
50,000 small sorted sets (3 members) 2.7 MiB 2.8 MiB 12.8 MiB 17.9 MiB 0.71×

Reading the memory table

Section titled “Reading the memory table”

Growth is Ruvio’s RSS increase from loading against Redis’s. Redis’s idle RSS on this machine was 13.2 MiB (8.4 MiB on the previous one), which pre-faults pages and makes its growth look smaller, so the total RSS after loading is the steadier number: Ruvio is lower in all 12 rows (0.71–0.86×). By growth Ruvio is lower in 7 rows, equal in 1 and higher in 4 (lists of 20 elements, HyperLogLogs, 128-byte bitmaps, geo points). Ruvio’s own growth values are unchanged from the previous run (within 0.2 MiB), so the difference comes from the Redis side.

Compatibility notes

Section titled “Compatibility notes”
  • GETEX key EX seconds (Redis syntax) is refused by Ruvio, whose GETEX takes milliseconds only; the compatibility matrix lists it.
  • BITFIELD and SETBIT in Redis refuse zero-padded numbers such as 000000000123; Ruvio accepts them.
  • DUMP is not comparable in content: Ruvio’s payload is its own checksummed WAL record, not RDB.

Server AGPL-3.0 · Clients MIT