SUMMARY:"ASE shared" (PCI based ) Presto device

From: Lucien HERCAUD <"Lucien_HERCAUD"_at_paribas.com>
Date: Tue, 20 May 1997 10:59:23 +0100

     All,
     
     I received two answers both from Digital Eqipment Corporation from :
     
        Alan Davis - Digital Unix Engineering "System Group"
        Chris Jankowski - Open Systems Consultant "Digital Eq.C. Australia"
     
     Both confirmed to me that -inside Digital- such a Presto device does
     not exist nor has anyone thought of creating one.
     For more detail, please read my comments below and the full content of
     the 2 replies.
     
     Regards,
                        Lucien Hercaud
     
     ---------------------------------------------------------------------
     
     Alan, Chris,
     
     Thanks for your answers. I think I know what PrestoServe is and I
     think I understand how it works. The way Digital designed and
     implemented PrestoServe lead to a presto-device which was to be
     attached to the system memory rather than being used on a Peripheral
     Bus. It seemed this gived the best performance (on simulations ?).
     
     I agree with Chris that it would be wonderful if the HSZ50 RAM Cache
     could be used by the SYSTEM as a PrestoServe device rather than only
     as a WB Cache by the raid controller itself. This because I know the
     Software driven PrestoServe device is a much more performant <<thing>>
     than just a Controller driven WB Cache.
     
     So, this is the real meaning of my question: with ASE we have the way
     of "moving" shared disks from one machine to another. Now, if the
     PrestoServe device could be "moved" and "shared" in the same manner,
     ASE and Presto could cohexist. Seems noone inside DEC thought or
     worried about this. For this to happen, the PrestoServe device must've
     laid on an IO attachement (e.g. PCI) not on the memory board. It may
     imply paying the price of loosing some raw performance in Single
     Machine mode in echange of availability+less_performance in cluster
     mode: a very old already-known tradeoff, right ? In addition, why DEC
     couldn't propose TWO PrestoServe versions: Memory integrated for
     Single Machine and SHARED IO based one for Clustered Machines (I think
     to recall that Sun implemented Presto on an IO bus device).
     
     Best reagards.


______________________________ Reply Separator _________________________________
Subject: Re.:"ASE shared" (PCI based ) Presto device
Author: chris_at_lagoon.meo.dec.com (Chris Jankowski) at internet
Date: 19/05/97 16:18


Lucien,
     
Presto is local to a machine - there is no physucal way its content can
be failed over when the machine is shot.
     
But why cannot you use the battery backed up cache in the HSZ50 controllers.
It does fail over and you can have 128MB of it per controller ie. with
8 of them you have a nice 1GB PrestoServe. (:-)).
     
Cheers,
     
Chris
     
 +-+-+-+-+-+-+-+ Chris Jankowski - Open Systems Cons.- chris_at_lagoon.meo.dec.com
 |d|i|g|i|t|a|l| Digital Equipment Corporation (Australia) tel.+61 3 92753622
 +-+-+-+-+-+-+-+ 564 St. Kilda Rd, Melbourne 3004, AUSTRALIA fax +61 3 92753453
     
    "UNIX was not designed to stop you from doing stupid things, because
     that would also stop you from doing clever things." -- Doug Gwyn
     
______________________________ Reply Separator _________________________________

   Lucien,

Presto shared between nodes is not possible. Since
the hw is tied directly to the computers bus, if the system fails there is no
possibility of moving that cache to the other presto board.

Alan Davis/dusg
Received on Tue May 20 1997 - 20:07:00 NZST

This archive was generated by hypermail 2.4.0 : Wed Nov 08 2023 - 11:53:36 NZDT