Thread Links |
Date Links |
||||
---|---|---|---|---|---|

Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |

*To*: P1619-2@xxxxxxxxxxxxxxxxx*Subject*: Re: [P1619-2] another P1619.2 question: the EME2 mix function*From*: Hal Finney <hal.finney@xxxxxxxxx>*Date*: Wed, 24 Mar 2010 10:40:50 -0700*Delivered-to*: mhonarc@xxxxxxxxxxxxxxxx*In-reply-to*: <3896c5d61003231007i25e499f0sd2e71b7997ce9e1b@xxxxxxxxxxxxxx>*List-help*: <http://listserv.ieee.org/cgi-bin/wa?LIST=P1619-2>, <mailto:LISTSERV@LISTSERV.IEEE.ORG?body=INFO%20P1619-2>*List-owner*: <mailto:P1619-2-request@LISTSERV.IEEE.ORG>*List-subscribe*: <mailto:P1619-2-subscribe-request@LISTSERV.IEEE.ORG>*List-unsubscribe*: <mailto:P1619-2-unsubscribe-request@LISTSERV.IEEE.ORG>*References*: <3896c5d61003221617k67c4bf9bg19f6f960bcb0e63f@xxxxxxxxxxxxxx> <ed843b101003221942y370ea9e1i23da20fa145d1342@xxxxxxxxxxxxxx> <3896c5d61003231007i25e499f0sd2e71b7997ce9e1b@xxxxxxxxxxxxxx>*Reply-to*: Hal Finney <hal.finney@xxxxxxxxx>

On Tue, Mar 23, 2010 at 10:07 AM, Laszlo Hars <laszlo.hars@xxxxxxxxxxx> wrote: > 5. EME2 is too slow for sequential implementations in modern disk drives, > where data is streamed at 12 Gb/s, or more. However, table B.2.1 of the > Standard is not applicable for parallel implementations; in Galois fields > multiplication with a power of x is not a shift-XOR operation, if products > with smaller powers are not available. The user would need some guidance how > complex is a 128x128 bit Galois multiplication in this special setting. Keep > in mind that typical applications work on 4KB sectors, so one has to > multiply with up to x^255, which is not of any simple structure when taken > modulo of the fixed 128 degree polynomial. In HW implementations large > lookup tables are expensive. One thing to look into is that you could still use 512 bytes as the EME2 block size even though the hardware sectors are 4096 bytes. Then you could have more parallelism and perhaps could cycle a LFSR fewer times to generate the multiplications by x powers. Another possibility along these lines would be to do a full Galois multiplication to multiply by x^128, then to do the LFSR 128 times, so it takes half as long. Or extend this idea to multiply by x^64, x^128 and x^192, then shift 64 times, and so on. No doubt as Matt says, Doug Whiting would be a good person to ask. Hal

**Follow-Ups**:**Re: [P1619-2] another P1619.2 question: the EME2 mix function***From:*Laszlo Hars

**References**:**[P1619-2] another P1619.2 question: the EME2 mix function***From:*Laszlo Hars

**Re: [P1619-2] another P1619.2 question: the EME2 mix function***From:*Matt Ball

**Re: [P1619-2] another P1619.2 question: the EME2 mix function***From:*Laszlo Hars

- Prev by Date:
**Re: [P1619-2] another P1619.2 question: the EME2 mix function** - Next by Date:
**Re: [P1619-2] another P1619.2 question: the EME2 mix function** - Previous by thread:
**Re: [P1619-2] another P1619.2 question: the EME2 mix function** - Next by thread:
**Re: [P1619-2] another P1619.2 question: the EME2 mix function** - Index(es):