I am a PhD student at EPFL Lausanne and I implemented, some time ago,
RSA 1024/2048 decryption on NVIDIA GPUs.
My software achieved a quite high throughput when decryption involves a
single private key or a few.
The latency is not very low unfortunately.
I would like to integrate my code in open SSL to allow GPU acceleration
of RSA decryption.
The problem is that to benefit from that, it should be possible to batch
decryptions.
To minimize confusion it's probably more appropriate to refer to
operation as "private key operation" or "sign" rather than "decryption".
Provided that it is realistic to assume that real SSL based applications
may have thousands of handshake requests at once (could you shed some
light on this?), would it be hard to allow batch decryption in SSL?.
I am working on reducing the latency of my code make worth offloading on
the GPU just a few decryptions, but even if I succeed in that I would
need some batching facility...
Do see discussion at http://marc.info/?t=118825449500017&r=1&w=2.
Personally I'm skeptical that it's feasible in general SSL case such as
web server in sense that it would be hard [if even possible] to justify
the effort and additional complexity. It probably would be more
appropriate to target specific cases. DNSSEC comes to mind...
______________________________________________________________________
OpenSSL Project http://www.openssl.org
Development Mailing List openssl-dev@openssl.org
Automated List Manager majord...@openssl.org
