[comp.arch] disk latency

jcallen@Encore.COM (Jerry Callen) (02/24/90)

In article <787@geovision.UUCP> gd@geovision.UUCP (Gord Deinstadt) writes:

[regarding put data in more than one place in a disk farm]

>You could still do it within a single drive, though; just write the data
>twice.  You would get a certain amount of fault tolerance, too, though
>not as much as a real disk mirror.  Does anybody do this?

IBM's VSAM can optionally do this with indexes. VSAM (at least, one flavor
of it) is a keyed access method, and the "replicate" option causes index
pointers to be written as many times as will fit around a track to
reduce latency. A pre-requisite is the "imbed" option, which puts the
lowest level of index pointers on the same cylinder as the data it points
to, to reduce seeks. The highest levels of the index are kept in memory.

>Ahhh, probably not... they'd only gain on the rotational latency
>anyways, and seek time dominates.

As main memory has gotten cheaper (and, therefore, bigger), the replicate and
imbed options are dying out; it's faster to just cache the whole bloody
index in memory.

>Gord Deinstadt   gdeinstadt@geovision.UUCP

-- Jerry Callen
   jcallen@encore.com