11.07.2014 Views

Identity-Based Encryption Protocols Using Bilinear Pairing

Identity-Based Encryption Protocols Using Bilinear Pairing

Identity-Based Encryption Protocols Using Bilinear Pairing

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Thus, for parallel version of pairing computation algorithm, one obtains optimal twomultiplier<br />

algorithms for both doubling and addition. For doubling, the four multiplier<br />

algorithm is optimal, while for addition, the four multiplier algorithm is sub-optimal. However,<br />

the Hamming weight of r will be small and hence if we use four multipliers then the<br />

sub-optimal performance will be amortized over the length of the representation of r and<br />

will not be significantly reflected in the final cost analysis.<br />

4.5 Comparison<br />

For the purpose of comparison, we assume that r = (r t = 1, r t−1 , . . . , r 0 ) is represented in<br />

NAF having length t and Hamming weight s.<br />

The irrelevant denominator optimisation was introduced in [8]. Further, [8] uses affine<br />

representation. The total cost including point/line computation and updation is<br />

t(1[I]+8[M]+2[S]) + s(1[I]+6[M]+1[S]),<br />

where [I] is the cost of inversion over IF p and is at least 30[M], see [72].<br />

Izu-Takagi [57] uses projective coordinates for pairing computation in general characteristics<br />

for large embedding degree k. They also consider the BKLS optimisations for<br />

supersingular curves with embedding degree k = 2 for general a. They assume that one IF p k<br />

multiplication takes k 2 [M]. For k = 2, this can be improved to 3[M]. In the following calculation,<br />

we use this fact. Their cost for w-iterated doubling is 6w[M]+4w[S]+13w[M]+(5w+1)[S]<br />

and addition is 6[M]+16[M]+3[S]. Summing over w’s, the total cost comes to<br />

t(19[M]+9[S])+s(22[M]+4[S]).<br />

The work of Scott [82] also proposes the use of projective coordinates in the case a = −3<br />

for certain non-supersingular curves. The paper does not distinguish between multiplication<br />

and squaring. The total cost is 21t[M]+22s[M].<br />

In Table 4.1, we summarize the above costs along with the costs obtained by our algorithms<br />

for the various cases for the curve parameter a. The best case occurs for Algorithm 1<br />

with a = −3. Also the cases for Algorithm 1 for small a and Algorithm 2 are marginally<br />

slower than the best case. However, all three of these cases are much more efficient than<br />

any of the previous algorithms. The algorithms of Izu-Takagi [57] and Scott [82] are more<br />

efficient than the basic BKLS algorithm with affine coordinates.<br />

For Tate pairing applications, r is generally chosen so that the Hamming weight s is<br />

small. On the other hand, for a general r, the Hamming weight s is approximately s = t/3.<br />

In either of these two situations, we summarize the superiority of our method as follows.<br />

• Algorithm 1 with a = −3 is approximately 20% faster compared to the algorithm by<br />

Scott.<br />

• Algorithm 2 is approximately 33% faster compared to the algorithm by Izu and Takagi.<br />

47

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!