On Monday, 12/20/2010 at 12:03 EST, George Henke/NYLIC <george_he...@newyorklife.com> wrote: > Why would you NOT want PAV for CMS mds? > > The IO Supervisor has not kept up with the hardware. > > It still thinks of a disk device as a "spinning platter" when in fact it is a > rank of RAID devices striped over numerous HDs and cached in a disk controller > from where it is actually being read thereby permitting multiple IOs to the > same device number.. > > The problem is that the IO Supervisor checks the IOB Busy Bit before issuing a > SIO and, if it is on, unnecessarilly suspends the SIO until the device is idle. > > Instead of changing the IO Supervisor, IBM has opted to fake it out by defining > alias devices for the same device number in PAV. > > Since most workloads these days are still IO bound, why would you still want to > unnecessarilly "single thread" IO, why would you NOT want PAV on CMS mds, SFS, > or whatever?
Not sure what you're talking about, George. The I/O subsystem is architected to permit exactly ONE active I/O per subchannel. The I/O supervisors MUST obey. So, PAV does not fake out the I/O supervisor. It obviates the need to do some Unnatural Acts with IOCP to get a single subchannel with more than one associated IODEVICE so that you can do more than one SSCH (not SIO any more!) to the same device on different subchannels. To get advantage for CMS apps, including DB2 and SFS, you must have more than one minidisk on the physical volume and I/O must be directed to more than one minidisk at the same time. If you don't have enough parallelism in the system, the I/Os will complete fast enough such that all I/Os go to the base anyway. So unless your performance reports are showing device queueing inside CP, PAVs will not help. Alan Altmark z/VM and Linux on System z Consultant IBM System Lab Services and Training ibm.com/systems/services/labservices office: 607.429.3323 alan_altm...@us.ibm.com IBM Endicott
