Actually, after eliminating other external factors I can confirm that decrypting a 65MB GPG file into memory and re-encrypting the substrings inside every one of 1,000,000 lines is 19% faster with 1.8.1 than with 1.7. This is now pretty accurate and reproducible. If it matters, this runs on an i7-2600 machine w/o HT (4 real cores).
Thank you Alex -----Original Message----- From: Alex Malmyguine Sent: Friday, May 20, 2016 9:30 AM To: 'peter.dett...@bouncycastle.org'; dev-crypto-csharp@bouncycastle.org Subject: RE: [dev-crypto-csharp] 1.8.1 vs 1.7? Peter, An inconclusive 10% improvement and profiling has skewed towards BigInteger.Remainder() - 46% and BigInteger.SquareMonty() - 25%. I will need to run more tests to gather conclusive stats. I am tempted to try outsourcing the BigInteger methods to OpenCL. Alex -----Original Message----- From: Peter Dettman [mailto:peter.dett...@bouncycastle.org] Sent: Friday, May 20, 2016 1:30 AM To: dev-crypto-csharp@bouncycastle.org Subject: Re: [dev-crypto-csharp] 1.8.1 vs 1.7? Hi Alex, I cannot offer you exact numbers, but there were definitely some significant improvements to the BigInteger.ModPow performance (including faster MultiplyMonty and less calls to it) after 1.7, so I would expect it at least worth a try. Perhaps if you manage to measure an improvement, you could report back to the list on the results. Regards, Pete Dettman On 20/05/2016 2:07 AM, Alex Malmyguine wrote: > Good time of the day! > > > > Would an application which spends 93% of encryption time in > BigInteger.Mod() and BigInteger.MultiplyMonty() benefit from upgrading > from 1.7 to 1.8.1? > > You probably need to know the specifics of the used keys, but I don't > know much about that as it's essentially a black box for me yet. > > > > Thank you! > > Alex >
