home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!nntp1.radiomail.net!csl.sri.com!boucher
- From: boucher@csl.sri.com (Peter K. Boucher)
- Newsgroups: sci.crypt
- Subject: Re: DES (Was: Re: 800MHz
- Date: 21 Jan 1993 19:48:33 GMT
- Organization: Computer Science Lab, SRI International
- Lines: 63
- Distribution: world
- Message-ID: <1jmumhINNe3d@roche.csl.sri.com>
- References: <1444.204.uupcb@ssr.com> <1993Jan9.163011.23230@csi.uottawa.ca>
- NNTP-Posting-Host: affirmed.csl.sri.com
-
- cbbrowne@csi.uottawa.ca (Christopher Browne) writes:
-
- |> If that $56 figure is faintly accurate, then I expect that the NSA
- |> must have some rather high-speed hardware, probably specialized to
- |> DES, that costs quite a bit less than anyone expected. I would have
- |> expected a somewhat higher bill.
- |>
- |> I can give some basic "blanks" to fill in:
- |>
- |> a) Number of messages processed: _____________
- |>
- |> b) Mean Size of messages processed: _____________ (may be important)
- |>
- |> A * B = "total ciphertext"
- |>
- |> c) Cost of electricity and staffing, per hour, for the NSA "DES
- |> Breaker" computer: ____________
- |>
- |> d) Average time to decrypt 1K of message on the "DES Breaker":
- |> ______________
- |>
- |> Actually, one can probably get that $56 just out of figures b), c) and
- |> d).
- |>
- |> $56 = (b) * (c) / (d)
- |>
- |> I'm ignoring capital costs, in this analysis, but I'd suspect that so
- |> does the NSA. Governments don't tend to consider "capital costs" to
- |> be REAL costs, for a variety of reasons.
- |>
- |> And if anyone wishes to publicise the values b), c) and d), that's
- |> THEIR choice. Those three numbers are probably considered QUITE
- |> secret.
-
- You need single-chip search engines (custom-built IC, not standard
- DES chips) with a gate delay of 3 nanoseconds to get a 1-microsecond
- search time. You make a machine with 100 racks of 10,000 IC's/rack.
-
- in 1976, given economies of scale, each IC would cost ~$10
- ($10M/machine). Allow a factor of two, for other hardware and you
- pay $20M/machine. Amortize costs over 5 years, and you pay $10K/day.
-
- 10**6 keys 10**5 sec 10**6 IC 10**17 keys
- ---------- X --------- X -------- = -----------
- sec/IC day 1 day
-
- 10**17 searches ~= 2**56 searches = 1 exhaustive search.
-
- 1 exhaustive search/day at $10K/day = $10K/exhaustive search.
-
- These are 1976 figures. Even if we are optimistic by an order
- of magnitude, the cost of computation falls in 5 years by a factor
- of 10; in 15 years (by now) by 1,000. This results in a cost of
- ~$10/exhaustive search.
-
- (The preceding is condensed from a talk given by Prof. Martin E.
- Hellman of Stanford University at the RSA Data Security Conference,
- 1/17/93)
-
- --
- Peter K. Boucher
- --
- RIPEM public key available upon request.
-