Re: The TransRelational Model: Performance Concerns
Date: Sun, 05 Dec 2004 01:42:37 GMT
Message-ID: <hItsd.41440$VL6.6102_at_clgrps13>
i think the comments below are 'jumping the gun'. i'd like to see a real implementation before i said similar.
permit me though, to offer a little provocation about something that i've wondered for some years, ever since even little machines could support hundreds of megabytes of fast memory at fairly low cost: i know for a fact that a medium-small airline (say 200 planes) used a real-time reservation database (i use the term 'database' advisedly because it was based on PARS / ACP) that occupied 300 MB on disk. of course there are lots of bigger databases than this, but there are even more that are smaller.
lots of databases use 'cache' as intelligently as they can but with lots of complications and arbitrary heuristics and so forth. wouldn't it be smart to have a database whose native query and update methods were designed to operate solely in memory? except perhaps for an optionally-symchronized log of updates, there would be no 'checkpointing' or similar mechanisms. in fact, such an engine might be able to support dozens or maybe hundreds of concurrent users without all the multi-tasking and storage coordination plumbing that is so common otherwise. of course, one wouldn't choose the same algorithms that are used for disk-based db's, rather ones that would facilitate memory re-use in an elegant way. comparatively speaking there would be almost no seek delay, almost zero latency and so forth. (i'm also intrigued with the potential storage economy of TRM as i've seen others with similar potential.)
i wonder would we be seeing a different world now had the old core memory been as cheap and plentiful as today's.
magoo
hungrylion2004 wrote:
>>Fabian Pascal comments on this very thread: >> >>http://www.dbdebunk.com/page/page/1548800.htm >> >>Unfortunately he doesn't actually have anything to say >>except "trust me, it won't be a problem," although he >>does promise a reply from "Steve Tarin, the inventor >>of TRM ... soon." >>
>
>
> And of course he advised his flock to buy the Intro...8th Edition and
> the upcoming "...Go Faster" which I suspect will be in the same manner
> as "What not how" -- that is, they'll use every trick to come up with
> a 100 page "book" when its really about 50 pages.
>
> An interesting question would be "Why are you involved with
> RequiredTech". Of course, the standard answer will be they are db
> experts who are doing consulting work. A real startup wouldn't give
> Pascal/Date the time of day. They would target the big online
> commerce players and naturally the Truly Relational malarkey would be
> of no interest and neither would publicizing their work.
Received on Sun Dec 05 2004 - 02:42:37 CET