Unfortunately we are emulating the geometry of the last generation of
ECDK devices (3390's) by creating virtual volumes within an array of
SCSI disks.

In its day the model 9 was a bulk data storage model with poor performance
characteristics.  The performance of virtual 3390 volume however is a
function of the real hardware used to build the SCSI array.  Some systems
programmers still persist in defining only 3390-3 volumes concerned over
the performance issues associated with real 3390-9's.  This results in a
larger number of volumes to manage than necessary.  Given today's
multi-terabyte
DASD farms, the number of logical volumes becomes staggering.

It is however true that I/O is queued at the device level and there may be
situations
where more emulated volumes are better.  However one can do no more
transfers than the
number of upper interfaces available (number of channel connections to the
subsystem).

In developing a capacity management system for a client last year, reporting
for DASD
services times only listed volumes whose services times exceeded 8
milliseconds.  We
have come a long way since the 1st cached controllers in the mid-80's.

3390's were state of the art disk drives 10 YEARS ago.  Even if you were
given the required 3990 controllers and some 3390 drives, it would still be
cheaper to purchase a new Shark, EMC, etc. storage subsystem due to the cost
of maintenance and the cost of power to spin them and to cool them.  That is
to
say that real 3390's are for all practical purposes no longer in use.  I
have not
been in a data center in the last 6 years that had any 3390's install.
Maybe I
have been lucky or oblivious to their presence.

Has anyone heard of any IBM plans to provide non-ECKD DASD support for zOS?
We can't go on emulating 33X0 volumes much longer.

Don Stubbs




-----Original Message-----
From: Linux on 390 Port [mailto:[EMAIL PROTECTED]]On Behalf Of
Peter Webb, Toronto Transit Commission
Sent: Thursday, April 18, 2002 11:07 AM
To: [EMAIL PROTECTED]
Subject: Re: How to pre-allocate a 4GB File?


> I'm sadly out of touch with mainframe hardware; it seems strange to me
> I can get 120 Gbyte disks for my PC and you folk are using 2.3 Gbyte
> drives on your mainframes. I know the theoretical limit is larger than
> that, and was back when S/360 was announced.
>
>
        One advantage to the limited disk size is that since you can only
have one I/O active to a disk volume at one time, spreading data over many
smaller volumes allows more I/Os to be active at the same time. This can
dramatically improve system I/O throughput. IBM even warns that using
3390-9s can degrade performance, think what using a 3390-120Gigabyte would
do! In the PC world, I believe you can see the same sort of effect by using
RAID0 across multiple volumes, versus one large volume.

Reply via email to