mirror of
https://github.com/nalgeon/redka.git
synced 2025-12-24 12:38:00 +08:00
818954d339
Shared cache is discouraged. The supported alternative (and increasingly recommended by SQLite developers), is the memdb VFS (which I believe is supported by all drivers). I'm sure modernc has it, and mattn also does if compiled with SQLITE_ENABLE_DESERIALIZE, which became the default in 2021. https://github.com/ncruces/go-sqlite3/issues/94#issuecomment-2157679766
1.3 KiB
1.3 KiB
Performance
I've compared Redka with Redis using redis-benchmark with the following parameters:
- 10 parallel connections
- 1000000 requests
- 10000 randomized keys
- GET/SET commands
SQLite settings:
pragma journal_mode = wal;
pragma synchronous = normal;
pragma temp_store = memory;
pragma mmap_size = 268435456;
pragma foreign_keys = on;
Hardware: Apple M1 8-core CPU, 16GB RAM
Redis:
redis-server --appendonly no
redis-benchmark -p 6379 -q -c 10 -n 1000000 -r 10000 -t get,set
SET: 133262.25 requests per second, p50=0.055 msec
GET: 139217.59 requests per second, p50=0.055 msec
Redka (in-memory):
./redka -p 6380
redis-benchmark -p 6380 -q -c 10 -n 1000000 -r 10000 -t get,set
SET: 36188.62 requests per second, p50=0.167 msec
GET: 104405.93 requests per second, p50=0.063 msec
Redka (persisted to disk):
./redka -p 6380 data.db
redis-benchmark -p 6380 -q -c 10 -n 1000000 -r 10000 -t get,set
SET: 26773.76 requests per second, p50=0.215 msec
GET: 103092.78 requests per second, p50=0.063 msec
So while Redka is 2-5 times slower than Redis (not surprising, since we are comparing a relational database to a key-value data store), it can still do 26K writes/sec and 94K reads/sec, which is pretty good if you ask me.