What ESDI Controller ships with the new AT&T 6386s?

Piercarlo Grandi pcg at rupert.cs.aber.ac.uk
Sun Oct 22 22:29:01 AEST 1989

In article <2542 at infmx.UUCP> aland at infmx.UUCP (Dr. Scump) writes:

    I have two basic questions:  1) do the 25 and 33 MHz ESDI models include
    DPT controllers?,  and 2) if not, has anybody tried to use the DPT
    caching ESDI controller in one of these machines?  Does anybody know a
    reason why they *wouldn't* work?
    Also, are there caching SCSI controllers that would work with the
    Model S, and how would the SCSI combination perform compared to ESDI
    with the DPT caching controller for ESDI?

Of course since you also want reliable operation, you *MUST*
*MUST* have an UPS if you use a caching controller with a
database system. If not, and the system goes off before the cache
has not been flushed, all the careful algorithms used by your
DMBS to ensure consistency, rewcoverability, and other
trivialities will have been made pointless, and your customers
will be *very* angry.
    We want to use several of these machines in an environment where 
    disk performance will be the biggest performance bottleneck.

A caching controller does not improve disk performance *AT ALL*. It simply
does the same job as the OS cache, or the DBMS cache, in a different way;
it reduces the average perceived disk latency, but not the maximum one,
at the price of keeping disc blocks in *VOLATILE* storage.

A caching disc controller is only useful if you have no control over either
(e.g. you have reached the limit of expandability of your system), and you
can guarantee that its cache is not volatile.

The OS or DBMS caches need not be guaranteed non volatile because
the OS or DBMS can bypass them for safe writes, while the
volatile storage in a caching disc controller is used even for
writes to what the OS or DBMS think are raw devices.

To have good disk performance, you had better rely on tuning the OS, the DBMS,
their cache sizes, and get fast disc controllers and fast discs (e.g. SCSI,
multiple discs with multiple controllersfor overlapped io, and discs with
tranfer rates of at least 1-2 Meg/sec. and access time of less than 18 ms.).
Piercarlo "Peter" Grandi           | ARPA: pcg%cs.aber.ac.uk at nsfnet-relay.ac.uk
Dept of CS, UCW Aberystwyth        | UUCP: ...!mcvax!ukc!aber-cs!pcg
Penglais, Aberystwyth SY23 3BZ, UK | INET: pcg at cs.aber.ac.uk

More information about the Comp.unix.i386 mailing list