Thank you for your clear and comprehensive reply, Alan. I conclude that if I am concerned about possible I/O queuing with emulated FBA I should consider spreading the I/Os for a particular linux guest across mini-disks but it is OK to have a large SCSI LUNs shared with other guests because CP may perform parallel SCSI I/Os. (and I may want to consider the arrangement of the LUNs in relation to the physical disk arrays and controllers of course). Yes, I know that there is some overhead in using emulated FBA rather than linux native SCSI but it is so much more convenient, especially for test systems.
Keith Gooding -------------------------------------------- On Thu, 20/3/14, Alan Altmark <[email protected]> wrote: Subject: Re: [LINUX-390] PAV and FBA dasd emulated on SCSI To: "Keith Gooding" <[email protected]> Date: Thursday, 20 March, 2014, 20:54 On Thursday, 03/20/2014 at 12:34 EDT, Keith Gooding <[email protected]> wrote: > I understand that without PAV an ECKD volume cannot have have multiple I/O > operations in progress and therefore the operating system (or is it the channel > ?) may have to queue I/Os to a volume. I know that this applies to z/OS and I > assume that it also applies to z/VM and linux, and may be a reason to avoid > using very large ECXD volumes. > > Does the same apply to a traditional channel-attached FBA volume (ie not > FCP-over-SCSI) ?. > > For mini-disks on z/VM emulated FBA on SCSI, I assume that at the z/VM can > issue several simultaneous SCSI IOs to the LUN but will each linux guest still > limit itself to one I/O per mini-disk ?. > Does the use of DIAG access to the mini-disk make any difference ? You are correct that PAV is needed to concurrently start more than one I/O to the same volume. The I/O queues are in the OS, not in the channel. In fact, it's the channel subsystem that rejects a second I/O to a device that has an outstanding I/O. This I/O architecture applies to ECKD and FBA, virtual or real. For FBA-on-SCSI, the guest can do only one I/O, but as you suggest, CP can do multiple I/Os to the SCSI LUN if he wishes. Whether the guest uses SSCH or DIAG, the rules are the same. That's half the equation. Not only does the channel-attached FBA I/O architecture have to be updated to support PAV, the host would have to be updated to deal with it. CP will not do PAV or advanced copy services on FBA devices, even if they are modified to support it. CP and other parts of VM assume that real FBA devices are of 1980s vintage. Alan Altmark Senior Managing z/VM and Linux Consultant IBM System Lab Services and Training ibm.com/systems/services/labservices office: 607.429.3323 mobile; 607.321.7556 [email protected] IBM Endicott ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 ---------------------------------------------------------------------- For more information on Linux on System z, visit http://wiki.linuxvm.org/
