Mac: 60 commands and memory after the hot-path changes

The same Mac comparison as the 2026-10-14 record, run again on the dev head after the hot-path pull requests (#281 to #299), next to the dev commit just before them, with Redis in every run. Raw data, every round: mac-2026-10-28.json (after), mac-2026-10-28-pre-hotpath.json (before), mac-2026-10-14.json (earlier record). Script: scripts/redis_compare_features.py.

• Host: Apple silicon, 12 cores, 24 GiB, macOS 26.7.1. Not pinned, not quiet: two Astro dev servers and one idle Ruvio process were running. The earlier record was taken on macOS 26.6.2.
• Ruvio: plain release build, memory durability, 1 shard. After is 93eb7084 (#299), before is 37bf8a5d, the dev head ahead of #281. Redis 8.10.2, same binary in all runs.
• Load: redis-benchmark, 50 clients, pipeline 16, 64 B values, one command per run, request counts scaled by 0.3, three rounds with alternating server order; the table shows the median. Three seconds between rows because of macOS ephemeral port exhaustion.
• Order: the before run and the after run were back to back (about 4 minutes each), each with its own Redis; the earlier column is the 2026-10-14 record.
• Many rows stop at 3.46M to 3.75M requests per second. That is the resolution and client limit of redis-benchmark on this host, so a gain on those rows cannot be seen here.
• Not measured: latency (p50, p99), durability on, more than one shard, large data sets, x86 or Linux. A Mac number says nothing about x86.

Geometric mean over the 60 commands: before against the 2026-10-14 record 1.28×, after against before 1.03×, after against Redis 1.46×. Most of the gain since 14 October was already in the dev head before #281; the hot-path changes add about 3.5%, with a few commands clearly ahead (EXISTS, GETBIT, ZINCRBY, APPEND, TTL, MEMORY USAGE, LPUSH) and a few slightly behind that are inside the round-to-round noise of one run (ZSCORE, SRANDMEMBER, PFCOUNT).

Requests per second in thousands. “After / before” and “After / Redis” are ratios.

Command	14 Oct	Before	After	After / before	Redis	After / Redis
GET	2,143	3,103	3,103	1.00	2,432	1.28
SET	2,812	3,214	3,103	0.97	2,045	1.52
LPOS (100 elements, hit in the middle)	1,579	1,800	1,765	0.98	1,125	1.57
LSET (100 elements)	1,875	2,143	2,195	1.02	1,765	1.24
LINSERT (pivot missing, scans 100)	1,286	1,385	1,406	1.02	882	1.59
LREM (value missing, scans 100)	1,304	1,385	1,429	1.03	909	1.57
LMOVE (rotate a 100-element list)	2,045	2,647	2,727	1.03	1,475	1.85
BITFIELD INCRBY u8 (always writes)	1,875	2,500	2,571	1.03	1,607	1.60
BITFIELD GET u8	2,195	2,903	2,903	1.00	1,800	1.61
SETBIT (1 KiB bitmap, same bit)	2,571	3,462	3,462	1.00	2,093	1.65
GETBIT (1 KiB bitmap)	2,250	3,000	3,600	1.20	2,368	1.52
BITCOUNT (1 KiB bitmap)	2,250	3,462	3,462	1.00	2,500	1.38
PFADD	2,432	3,462	3,462	1.00	2,432	1.42
PFCOUNT	2,432	3,462	3,214	0.93	2,647	1.21
COPY … REPLACE (64 B string)	2,093	3,214	3,214	1.00	1,525	2.11
DUMP (64 B string)	1,957	2,308	2,500	1.08	1,765	1.42
EXPIRETIME	2,500	3,600	3,750	1.04	2,727	1.37
TOUCH	2,308	3,750	3,750	1.00	2,812	1.33
OBJECT ENCODING	2,727	3,333	3,333	1.00	2,368	1.41
MEMORY USAGE	2,432	3,103	3,600	1.16	2,432	1.48
GETRANGE	2,571	3,103	3,103	1.00	2,093	1.48
SETRANGE	2,432	3,600	3,600	1.00	2,143	1.68
SETEX	2,500	3,000	3,103	1.03	1,304	2.38
MSET (3 keys)	2,195	2,812	2,903	1.03	1,154	2.52
MGET (3 keys)	1,837	2,045	2,093	1.02	1,667	1.26
DEL (missing key)	2,368	3,462	3,462	1.00	2,571	1.35
EXISTS	2,500	2,647	3,600	1.36	2,727	1.32
TYPE	2,368	3,214	3,214	1.00	2,727	1.18
TTL	2,368	3,103	3,600	1.16	2,727	1.32
PTTL	2,250	3,333	3,462	1.04	2,727	1.27
PERSIST (key without TTL)	2,500	3,750	3,600	0.96	2,727	1.32
EXPIRE (key with TTL)	2,308	3,103	3,214	1.04	1,875	1.71
APPEND (1 byte)	2,432	2,571	3,000	1.17	2,368	1.27
INCRBYFLOAT	1,343	1,385	1,500	1.08	1,268	1.18
HSET (existing field)	2,500	3,000	3,000	1.00	2,000	1.50
HGET	2,571	3,214	3,333	1.04	2,250	1.48
HMGET (2 fields)	2,000	2,432	2,368	0.97	1,837	1.29
HSTRLEN	2,500	3,214	3,600	1.12	2,308	1.56
LPUSH	2,093	2,308	2,647	1.15	2,000	1.32
LLEN	2,308	3,600	3,600	1.00	2,647	1.36
LRANGE 0 2	1,875	2,093	2,045	0.98	1,800	1.14
SADD (existing member)	2,368	3,462	3,462	1.00	2,500	1.38
SISMEMBER	2,500	3,333	3,600	1.08	2,432	1.48
SRANDMEMBER	2,432	3,214	3,000	0.93	2,571	1.17
ZADD (existing member, same score)	2,143	2,812	3,000	1.07	1,915	1.57
ZINCRBY	1,579	1,731	2,045	1.18	1,552	1.32
ZSCORE	2,250	3,103	2,812	0.91	2,250	1.25
RENAME (same key)	2,500	3,333	3,214	0.96	2,368	1.36
PUBLISH (no subscribers)	2,571	3,600	3,600	1.00	2,727	1.32
HGET (10-field hash)	2,308	3,103	3,214	1.04	1,915	1.68
HSET (10-field hash, existing field)	1,915	2,647	2,903	1.10	1,731	1.68
HDEL (10-field hash, missing field)	1,957	3,000	3,103	1.03	2,045	1.52
HGETALL (10-field hash)	763	744	763	1.03	763	1.00
LINDEX (20-item list)	2,143	2,903	3,214	1.11	2,250	1.43
LRANGE 0 9 (20-item list)	1,184	1,233	1,233	1.00	1,216	1.01
LSET (20-item list)	2,195	2,903	3,000	1.03	2,000	1.50
GEOADD	1,935	2,609	2,609	1.00	1,154	2.26
GEOSEARCH (radius 50 km, COUNT 10)	44	46	47	1.03	19	2.47
GETEX (set a 100 s deadline)	2,368	2,812	3,000	1.07	1,500	2.00
FILLGET (value present)	1,957	2,432	2,500	1.03	2,647	0.94

Resident memory above the idle process, in KiB, after loading the same data into each server. Idle resident memory of Ruvio: 8,896 KiB on 14 Oct, 9,312 KiB before, 9,296 KiB after (about 420 KiB more than on 14 Oct in both new runs; the cause was not looked into). Before and after are the same within a few percent, so the hot-path changes did not change memory.

Data	14 Oct	Before	After	Redis
20,000 lists of 20 elements	6,016	6,032	6,032	6,464
5,000 HyperLogLogs of 100 elements	2,288	2,288	2,272	2,720
20,000 bitmaps of 128 B (SETBIT)	3,808	3,808	3,776	4,512
100,000 BITFIELD counters (u16)	3,648	4,016	4,000	7,376
100,000 geo points in one key	6,384	7,616	6,912	7,936
100,000 strings of 16 B	5,200	5,136	5,120	8,928
100,000 strings of 128 B	18,176	18,144	18,128	21,536
100,000 strings of 16 B with TTL	10,176	10,480	10,448	11,504
100,000 one-element lists	4,992	4,992	4,992	8,944
50,000 small hashes (3 fields)	3,136	3,136	3,120	4,512
50,000 small sets (4 members)	2,704	2,736	2,752	4,560
50,000 small sorted sets (3 members)	2,688	2,784	2,832	4,496