[tahoe-dev] Fwd: [zfec-dev] effect of packet size on erasure coding speed

[zooko]

From: "Adam Langley" <agl at imperialviolet.org>
Date: September 14, 2008 11:36:04 AM MDT
To: zooko <zooko at zooko.com>
Subject: Re: [zfec-dev] effect of packet size on erasure coding speed

On Sun, Sep 14, 2008 at 10:20 AM, zooko <zooko at zooko.com> wrote:
> Personally, I'm going to be working on pycryptopp [1, 2], and  
> immutable file
> checking and repair for the next little while.

I'm actually working on a better public key signing system for DNSSEC
today, but I'm guessing that signature length is a big concern for
you, so ECDSA is probably the right solution.

However, I note in [1] that your ECDSA speeds are terrible! :

# signer.sign(len<=1000): 2.4-2.9ms, len=1M: 15ms, len=10MB: 140ms
# verifier.verify(len<=1000): 5.7-6.7ms, len=1M: 18-33ms, len=10MB:  
137ms

Taking the len<=1000 case to be dominated by the ECDSA itself, you
should be seeing verify speeds 5x faster:

http://bench.cr.yp.to/results-sign.html

An ECDSA verify on NIST P192 should be taking about 750us on a 2.33GHz
Core 2. The source code for that it just using the standard OpenSSL
calls (I can get it if you would like)


Cheers

AGL

-- 
Adam Langley agl at imperialviolet.org http://www.imperialviolet.org