Ah, but while that was true of a ³real² 3390, is that also now true of
emulated 3390¹s which are split across varying numbers of essentially SCSI
disks? A single 3390 mod 27 might be split up over several 9 gig physical
disks in order to implement the emulation. Is the controller smart enough to
be able to start an I/O to each, even though the I/O¹s were sent to the same
3390 address?

The waters get muddier every day....

-- 
Robert P. Nix          Mayo Foundation        .~.
RO-OE-5-55             200 First Street SW    /V\
507-284-0844           Rochester, MN 55905   /( )\
-----                                        ^^-^^
"In theory, theory and practice are the same, but
 in practice, theory and practice are different."



On 3/25/10 1:43 PM, "Kris Buelens" <[email protected]> wrote:

> There is maybe some misunderstanding: (leaving out PAV a while) a device can
> handle only 1 IO at a time, guests know that, CP too.  So indeed a linux will
> not send a new IO if the previous one to that disk hasn't ended yet, the guets
> will queue it.  With several guests with minidisks on the same disk, CP will
> queue the requests.  Multipathing doesn't change that.
> Multipathing helps in cases where the device is not busy, but the channel or
> control unit is.
> 
> PAV simplified: with PAV, you make appear a single disk as if it were many
> disks, but for each PAV address the old rule is still valid: one IO at a time.
> If you guest is PAV aware, it can also avoid queuing, if not, putting many
> guests on the same volume and let CP exploit PAV can improve IO rates.
> 
> 2010/3/25 RPN01 <[email protected]>
>> Another point I¹ve not seen mentioned, and I¹m not sure if it¹s true or
>> not...
>> 
>> Given a dedicated volume to a Linux guest, won¹t the guest start only one I/O
>> to the device at a time, and wait for it to complete? If you break up a
>> larger volume into several minidisks (like a mod 27 into mod 9¹s) aren¹t you
>> allowing the z/VM multipathing to do its job more efficiently, even if you
>> give all the smaller volumes to the same Linux guest?
>> 
>> Like I said, I may be totally off base, but this is the impression I¹ve
>> had...

Reply via email to