mRSA Measurements

The mRSA results are obtained by measuring the time starting with the message hash computation by the user (client) and ending with the verification of the signature by the user. The measurements are illustrated in Table 4.

Table 4: Timings for mRSA (in milliseconds).

Keysize	466 Mhz PII	930 Mhz PIII
(bits)	(slow client)	(fast client)
512	8.0	9.9
1024	45.6	31.2
2048	335.6	178.3

It comes as no surprise that the numbers for the slow client in Table 4 are very close to the unoptimized RSA measurements in Table 3. This is because the time for an mRSA operation is determined solely by the client for 1024- and 2048- bit keys. With a 512-bit key, the slow client is fast enough to compute its PS_u in 6.9ms. This is still under 8.0ms (the sum of 4ms round-trip delay and 4ms RSA operation at the SEM).

The situation is very different with a fast client. Here, for all key sizes, the timing is determined by the sum of the round-trip client-SEM packet delay and the service time at the SEM. For instance, 178.3ms (clocked for 2048-bit keys) is very close to 174.7ms which is the sum of 5.5ms communication delay and 169.2ms unoptimized RSA operation at the SEM.

All of the above measurements were taken with the SEM operating in a stateful mode. In a stateless mode, SEM incurs further overhead due to the processing of the SEM bundle for each incoming request. This includes decryption of the bundle and verification of the CA's signature found inside. To get an idea of the mRSA overhead with a stateless SEM, we conclude the experiments with Table 5 showing the bundle processing overhead. Only 1024- and 2048-bit SEM key size was considered. (512-bit keys are certainly inappropriate for a SEM.) The CA key size was constant at 1024 bits.

Table 5: Bundle overhead in mRSA with a SEM in a stateless mode (in milliseconds).

SEM key size	Bundle overhead
1024	8.1
2048	50.3

<384>>