Hi,
The time numbers from "openssl speed -multi <num processes>" can be
easily misunderstood as the elapsed time (latency) for each crypto
operation. However, they are simply 1 / <total throughput>. While it
makes sense in single-processor cases, they are roughly n times lower
than they should be in multi-processor cases. Having more CPU cores
does not reduce the latency of individual operations. Those operations
are not parallelized, but just replicated for each core. Multi-core
latency numbers should be same (linear scalability) with or higher
(sublinear scalablity) than single-core numbers.
I believe this bug(?) exists on all recent versions.
On a single core:
sign verify sign/s verify/s
rsa 512 bits 0.000106s 0.000010s 9394.8 100644.5
rsa 1024 bits 0.000534s 0.000029s 1872.3 34610.7
rsa 2048 bits 0.003293s 0.000099s 303.7 10121.8
rsa 4096 bits 0.023287s 0.000369s 42.9 2710.7
On eight cores:
sign verify sign/s verify/s
rsa 512 bits 0.000013s 0.000001s 74596.6 800000.0
rsa 1024 bits 0.000066s 0.000004s 15055.7 275862.1
rsa 2048 bits 0.000412s 0.000012s 2427.7 80911.2
rsa 4096 bits 0.002904s 0.000046s 344.3 21717.0
Regards,
Sangjin Han
______________________________________________________________________
OpenSSL Project http://www.openssl.org
Development Mailing List openssl-dev@openssl.org
Automated List Manager majord...@openssl.org
