batch binary edwards d j bernstein university of illinois
play

Batch binary Edwards D. J. Bernstein University of Illinois at - PDF document

Batch binary Edwards D. J. Bernstein University of Illinois at Chicago NSF ITR0716498 p index calculus Classic F needs to check smoothness < p . of many positive integers Smooth integer: integer > y . with no prime divisors y


  1. Batch binary Edwards D. J. Bernstein University of Illinois at Chicago NSF ITR–0716498

  2. � p index calculus Classic F needs to check smoothness < p . of many positive integers Smooth integer: integer > y . with no prime divisors y ) 2 2 Typical: (log (1 = 2 + o (1)) log p log log p . y 2+ o (1) , Many: typically y 1+ o (1) are smooth. of which (Modern index calculus, NFS: y .) smaller integers; smaller How to check smoothness?

  3. Old answers: Trial division, y 1+ o (1) ; rho, time y 1 = 2+ o (1) , time assuming standard conjectures. Better answer: ECM etc. o (1) ; specifically y p Time o (1)) log y log log y , exp (2 + assuming standard conjectures. Much better answer (using RAM): Known batch algorithms test smoothness of many integers simultaneously. O (1) y ) Time per input: (log O (log log y ). = exp

  4. General pattern: Algorithm designer optimizes algorithm for one input. But algorithm is then applied to many inputs! Oops. Often much better speed from batch algorithms optimized for many inputs. p # speedup. e.g. Batch ECDL: Batch NFS: smaller exponent. Can find many more examples.

  5. Surprising recent example: Batching can save time in multiplication ! Largest speedups: F 2 [ x ]. Consequence: New speed record for public-key cryptography. � 30000 scalar mults/second on a 2.4GHz Core 2 Quad for a secure elliptic curve/ F 2 251 . Software release this month.

  6. Surprising recent example: Batching can save time in multiplication ! Largest speedups: F 2 [ x ]. Consequence: New speed record for public-key cryptography. � 30000 scalar mults/second on a 2.4GHz Core 2 Quad for a secure elliptic curve/ F 2 251 . Software release this month. Note: No subfields were exploited in the creation of this record.

  7. Batched conditional branches are slow and painful. Solution: complete curve operations. 2008 Bernstein–Lange–Rezaeian n � 3, every Farashahi: for n ordinary elliptic curve over F 2 can be written as a “complete binary Edwards curve.” Extremely fast formulas for complete differential addition. With good curve selection: 5 M + 4 S per bit.

  8. Note 1: Need complete curve . 1 Need singularities at blowing up irrationally. Symmetric, Edwards-like: x 2 ( y 2 + y + d ) x ( y 2 + � � � ) + ( dy 2 + � � � ), + y 2 + y + d irreducible. with Note 2: Need complete formulas . Warning: for odd characteristic, x 1 ; y 1 ) + ( x 2 ; y 2 ) = � � ( x 1 y 1 + x 2 y 2 x 1 y 1 � x 2 y 2 ; x 1 x 2 + y 1 y 2 x 1 y 2 � x 2 y 1 is an incomplete addition law on a complete Edwards curve!

Recommend


More recommend