Re: Natural keys vs Aritficial Keys

From: paul c <>
Date: Mon, 18 May 2009 18:03:31 GMT
Message-ID: <TphQl.28548$PH1.17785_at_edtnps82>

Walter Mitty wrote:
> My seven most used files are each one byte long, but you should see how
> long the other 1000 files are!
> I can't help but wishing the Claude Shannon's work was not more well
> understood among software engineers.

Once, just to prove a point, I made a packaging utility to replace the static portions of a bunch of system tables in an oddball product with code. From the point of view of end-developers who used this system to produce apps, that stuff was mostly irrelevant and pure overhead, which was measurable for some apps. With every release, the utility would generate a perfect hash, at the expense of using more storage. This was practical because there were only several hundred keys and at that time, memory latency wasn't the big factor it is today, where a 2GHz cpu might be crippled by 500Mhz memory. A typical app could indirect or "introspect" the "catalog" many times for each "key" and it was silly to invoke caching for that purpose but behaviour needed to be identical to "native" or un-optimized behaviour. Such a method of "compiling" tables today might need to cater to latency more than I needed to then. Received on Mon May 18 2009 - 20:03:31 CEST

Original text of this message