Re: Admiral Tuttle (Should be silent ad for Oracle)

[magnesium.club.cc.cmu.edu!news.sei.cmu.edu!firth@uunet.uu.net (Robert Fi]

In article <9307090525.aa28379@dsc.blm.gov> cjames@DSC.BLM.GOV (Colin James 062
1) writes:

>The email path of the message is apparently from "oracorp.com", and
>Oracle has a lot invested in parallel processing to speed up the
>kludge performance of an interpreted query language.

That sounds very odd to me.  In my experience, which is not all that
limited, the critical performance bottleneck in any information system
is the speed with which objects on mass storage can be accessed, which
is almost entirely an I/O peoblem and to a small extent an indexing
problem.  And if the issue is the performance of queries, the one
optimisation that dominates all the others is the reorganisation of
the relational primitives to minimise the size of the intermediate
retrieved sets of tuples (eg apply the better filters first).

Neither of these has much to do with parallel processing - indeed,
without serious investment in heavily multiported mass storage
devices, I'd expect parallel processors to degrade performance, because
of collisions in the device controllers and a reduced ability to schedule
I/O requests to minimise latency.

Does anyone have some hard facts about this issue?