Installed DragonFly on n2-standard-8 (8 vCPU, 32GB Memory) Arch: x86/64 GCP Machine using binary.
Generating load from another GCP instance 62 vCPU, 64GB Memory using memtier_benchmark benchmarking tool.
Installed the Redis(Version: 7.2) on similar GCP Machine.
After tuning threads(–threads) and clients(–clients), I didn’t get significant DragonFly performance over Redis.
Redis:
Type Ops/sec Hits/sec Misses/sec Avg. Latency p50 Latency p99 Latency p99.9 Latency KB/sec
----------------------------------------------------------------------------------------------------------------------------
Sets 129430.41 --- --- 0.61780 0.59900 1.27100 1.82300 38410.60
DragonFly:
Result: 1
**CPUAffinity=0-3 ** (Using 4 vCPU)
Type Ops/sec Hits/sec Misses/sec Avg. Latency p50 Latency p99 Latency p99.9 Latency KB/sec
----------------------------------------------------------------------------------------------------------------------------
Sets 266886.94 --- --- 0.29965 0.29500 0.51900 0.76700 79203.099
Result: 2
CPUAffinity=0-7 (Using only 8 vCPU)
Type Ops/sec Hits/sec Misses/sec Avg. Latency p50 Latency p99 Latency p99.9 Latency KB/sec
----------------------------------------------------------------------------------------------------------------------------
Sets 262463.29 --- --- 0.30471 0.26300 0.76700 1.41500 77890.29
Observation Performance of dragonfly got saturated after 4CPU (Ran dragonfly with CPUAffinity) and performance didn’t scale with increased CPU.
Please help with the steps to achieve the claimed performance of Dragonfly and point out if anything wrong with above steps??
What should be max CPU% or load at which we should run DF and Redis in Prod?