We ran the SEM daemon on a Linux PC equipped with an 800 Mhz Pentium III processor. Two different clients were used. The fast client was on another Linux PC with a 930 MHz Pentium III. Both SEM and fast client PC-s had 256M of RAM. The slow client was on a Linux PC with 466 MHz Pentium II and 128M of RAM. Although an 800 Mhz processor is not exactly state-of-the-art, we opted to err on the side of safety and assume a relatively conservative (i.e., slow) SEM platform. In practice, a SEM might reside on much faster hardware and is likely to be assisted by an RSA hardware acceleration card.
Each experiment involved one thousand iterations. All reported timings are in milliseconds (rounded to the nearest 0.1 ms). The SEM and client PCs were located in different sites interconnected by a high-speed regional network. All protocol messages are transmitted over UDP.
Client RSA key (modulus) sizes were varied among 512, 1024 and 2048 bits. (Though it is clear that 512 is not a realistic RSA key size any longer.) The timings are only for the mRSA sign operation since mRSA decrypt is operationally almost identical.