We hacked up a quick https client and server using libev and were shocked to see openssl performing ~ 10% faster! And openssl is more memory greedy too… Has anybody else made a different experience?
Are you using yaSSL or CyaSSL? CyaSSL will preform faster on most compiler/processor/systems. If you're using CyaSSL are you using fastmath? Compiling in 64bit mode will also help if possible.
You say openssl is more memory greddy too. Do you mean disk space, or runtime memory?
It was yaSSL (because CyaSSL had problems compiling on Debian Lenny… which we've now tracked down to a not compiling fastmath). Without fastmath then CyaSSL is a little bit faster than yaSSL…
Greedy: I mean runtime memory. yaSSL appears to use very little per-connection runtime memory - I'm guessing it allocates most buffers dynamically? CyaSSL appears to be sligthly more greedy than openssl.
I'd like to help you get fastmath working on Debian Lenny. Are you compiling in 32 or 64 bit mode? I'm guessing you're using GCC. In 32 bit mode fastmath needs all of the ia32 registers to be available so we may need to tweak the build settings to get it to work. Is there a box you can give me access to? You can email me directly at "todd (at) yassl.com".
We're using 32 bit mode on that box. Thanks for the offer to help get it working. Unfortunately all the boxes are behind firewalls and difficult to access. We'll figure it out. At this point we're more interested in memory consumption (= more concurrent clients) rather than speed.
Sign up for the SourceForge newsletter:
You seem to have CSS turned off.
Please don't fill out this field.