performance_multi_engines.md

Performance

This document mainly introduces the performance of main algorithms in FATE-v2.0 running on different computing engines

Testing configuration:

• 2 hosts with same machine configuration
  • System: Centos 7.2 64bit
  • cpu: Intel(R) Xeon(R) Platinum 8255C CPU @ 2.50GHz, 16 cores
  • Memory: 32GB
  • hard disk: 1T
  • Network: Lan, 1Gb
• use 16 cores on each host
• Dataset:
  • PSI: PSI algorithm, input data is 10 million, result size is 10 million
  • Other Algorithms:
    • guest site: 10w * 30 dimensions, host size: 10w * 300 dimensions
• Job Configuration:
  • computing_partitions: 16

Algorithm	EggRoll	Spark-Local	Standalone
PSI	7m43s	7m28s	7m10s
Hetero-SSHE-LR	5m17s/epoch	6m4s/epoch	5m34s/epoch
Hetero-Coordinated-LR	2m22s/epoch	2m21s/epoch	2m19s/epoch
Hetero-SecureBoost	2m11s/epoch	3m41s/epoch	1m54s/tree

Base on the same testing configuration，we have also tested on different encryption schemas.

• algorithm parameters:
  • he_params: tested on he_param(kind="paillier") and he_param(kind="mock"), mock stands for plaintext.
• other configurations: same with Testing configuration list above

Algorithm	EggRoll(Paillier)	EggRoll(Mock)	Spark-Local(Paillier)	Spark-Local(Mock)	Standalone(Paillier)	Standalone(Mock)
Hetero-Coordinated-LR	2m22s/epoch	19s/epoch	2m21s/epoch	20s/epoch	2m19s/epoch	19s/epoch
Hetero-SecureBoost	2m11s/epoch	1m36s/epoch	3m41s/epoch	2m57s/epoch	1m54s/epoch	1m17s/epoch