I am not sure if we are talking about the same PAV... But the one I know you would have to use static PAV and define a base and several alias devices in iocds and let the guest access tem. These would be present in Linux as multiple devices (might also require to use extended node names) that will have to be grouped as a multipath volume (device-mapper manually or EVMS). So you would always group the devices in Linux. Using only one big device will be slower since only one request is processed for one device before the next is handled. So, IMHO the difference is only in the PAV case the decission where to send any request is based on the dm-round-robin scheduler and on which base device on the group while for the may dasds its only based on which device of the volume the data is written...
Stefan Bader SW Linux on zSeries Development & Services [EMAIL PROTECTED] ---------------------------------------------------------------------------------- When all other means of communication fail, try words. Tom Duerbusch <[EMAIL PROTECTED]> Sent by: Linux on 390 Port <LINUX-390@VM.MARIST.EDU> 06.07.2005 17:00 Please respond to Linux on 390 Port To LINUX-390@VM.MARIST.EDU cc Subject PAV vs LVM vs native Sorry about the length of this... I will be running Oracle 10g (64 bit), under SLES9, under z/VM 5.1 on a z/890 with a ficon attached Shark. I have some large databases that I will need to configure. One database will be used for a single application. The other database will be used for multiple (15-20) applications. For now, the data portion of the database will be either: 10 3390 mod 27s or 30 3390 mod 9s or 90 3390 mod 3s That is suppose to all come out to basically the same size. So, I'm trying to make an informed decision on what will be best (or best for us). Assuming that device addresses are not a concern (i.e. that the 255 device limit per Shark LSS isn't going to be a problem) what are the pros and cons of each of the following? PAV: Is hardware supported. It seems that I can't define a full pack minidisk on it, but will have to attach the PAV to the Linux machine...no MDC for these volumes. I may still do LVM of the 10 volumes to present a single image to Oracle, or not. Is there CPU savings in using hardware supported PAV, vs Linux supported LVM? Any other savings or limitations? LVM: Linux supported. May take more CPU cycles then the hardware option. May be used in conjunction with PAV to give a single volume image. May need to use the expanded device names (/dev/dasda1 to /dasdz1 just isn't going to cut it). Oracle supported: You can give any number (within reason) of disks to Oracle which can use them as a single pool of space. Since all of this space is going to Oracle anyway, management of 90 volumes vs 10 volumes shouldn't be a big concern. There may be some advantage for Oracle to be able to manage multiple disks vs letting the hardware or OS handle it for you. Right now, I'm torn between hardware supported PAV (hardware using less CPU cycles to manage..right?) and Oracle supported. With a single LVM image, will Oracle try to do only 1 I/O at a time? I expect the data to be cache unfriendly. We can't afford the money for either an increase in Shark cache or to increase the Oracle SGA to make sure we have a high cache hit ratio. But I expect our total I/O rate for these two Oracle databases to be in the 1,000 to 2,000 I/O per second in the worst case. The Shark has 8 8 pack pairs of 72 GB disks at 15K RPM and 8 GB of cache. Is there a difference in the considerations between an Oracle database that supports one application (i.e. large tables) vs an Oracle database that supports many applications (smaller tables but 20 times as many)? Tom Duerbusch THD Consulting ---------------------------------------------------------------------- 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 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
