Disk and shadowing

[Piet Timmers]

Two disks:

Total blocks 71114623

Disk1: Sectors per track 96
Disk2: Sectors per track 254

It is possible to shadow on OpenVMS or should these values be the same (also cylinders and tracks differ).

Greetings,

Piet

[Ian Miller.]

I think as long as total blocks is the same on both disks then you are fine.

[Ken Robinson]

You don't say which version ov VMS you're running. Starting with V7.3-2, you can shadow dis-similar devices.

Ken

[Mohamed K Ahmed]

Shadowing has to be done with disks with the EXACT configuration (sectors, cylinders....etc)
Starting with new versions, you can shadow and disregard the sectors and cylinders, and focus on total size

Mohamed

[Martin P.J. Zinser]

Documentation about this can be found in the 7.3-2 new features manual (e.g. at
http://h71000.www7.hp.com/doc/732FINAL/aa-rv8xa-te/00/00/28-con.html

Greetings, Martin

[Willem Grooters]

Piet,

I _think_ geometry doesn't matter.
So: yes, you can shadow.
However:
I understood that even disks that appear equal in size MAY in the end be incompatible due to SIZE. As I can deduct from other information, it may happen if different controllers (controller-types or -brands) see them different; when the disks are virtual - defined in a SAN or as a RAID beyond a HSx, it can happen as well (Told so on the last TUD).
By that, shadowing will fail if the paths are different - when NOT VMS 7.3-2. Under 7.3-2 however, I _think_ the biggest of the two would be NOT the primary disk, unless properly initialized (taht is, to a size, less than max, equal, or smaller then the full size of the second one) - see another thread.

[Mohamed K Ahmed]

I only have one OpenVMS system 7.3-1, the disks are in a MA8000 configurations, all the disks are connected to the system via HSG80 controller and SAN Fiber switch. I have built the system with 7.2-1, and at that time the disks have to be exactly the same to be able to be shadowed with their full size, Otherwise, the system won't shadow them and would error out.

Mohamed

[Lokesh_2]

Hi,

I think we can go ahead and experiment. Just check if both the disks show same size with $show device /full . If yes, then initialize both the disk and mount them in shadow set. If it works , fine. Else, we may get some error which will suggest something.

Hi Piet, can you post here the output of $sho dev/full for both the disks ?

Best regards,
Lokesh

[Uwe Zessin]

Early versions of volume shadowing required even the device names to be the same. That has been lifted in a version I don't know. Between that version and V6.2 the exact same geometry was a requirement due to the geo-dependent placement of homeblocks.

Todays disk drives report completely invalid 'geometries' to the outside that don't reflect their real arrangement. The reason is that modern disk drives put more 'sectors' (=more data) on the outer cylinders. This is possible, because the circumfence (hm, a web dictionary I have consulted says 'circumference' - what is correct?) is greater and the write heads can keep the bit density the same. This is sometimes called 'zoned bit recording' or 'banding'.

Since OpenVMS V6.2 a new, geometry-independent homeblock placement algorithm is used as a default. See HELP INITIALIZE /HOMEBLOCKS

Starting with OpenVMS V7.3-2 it is even possible to shadow disk drives with a different number of logical blocks.

[Martin P.J. Zinser]

Hi Uwe,

to learn about the difference between the circumference and a circumfence check
Terry Pratchetts "The color of magic".

Greetings, Martin

P.S. I do know that this is OT ;-)

[Mobeen_1]

Piet,
You should have no issues mirroring your disks (with different sectors) depending on what version of OpenVMS your servers are running.

As many of our friends here have mentioned, if you are on OpenVMS 7.3-2, you should be able to mirror these 2 disks and if you are in prior versions you would not be able to do so.

rgds
Mobeen

[Piet Timmers]

Hi,

Everybody says: on OpenVMS V7.3-2 it is possible, before not. We are using V7.2-2, so this will probably give problems.

The question is, how to prevent?
I buy two disks from HP, for our SAN, both 36 GB/10K, so I expect them to be the same, so they can be used in a shadowset, but this is not true, do I have to specify the hardware details when I buy a disk.

Piet

ps. we have had some problems shadowing these disks, only no answers from HP.

[Mobeen_1]

Piet,
I think the best way to deal with while making any purchases from HP is go by their part numbers. That will rule out any miscommunication.

It looks like you have 2 different types of 36GB HDD. I would suggest that you talk to HP and tell them that you need the same type and get an exchange.

rgds
Mobeen

[Uwe Zessin]

Piet,
most disks of the same family report the same geometry to the outside. With SBB based disks, which I have tracked a time in the past I have only seen 2 exceptions.

Honestly, I don't recall if I have ever mirrored disks with the same number of logical blocks, but a different geometry. I am pretty sure it will work since OpenVMS V6.2 and up.

[Uwe Zessin]

Mobeen,
that does not work. It is totally normal to receive a batch of disks with the same number, but when you look at them you will see different HDAs.

[Mobeen_1]

Piet,
Since we have mixed responses. If your server is not in production or if you are doing this on your testbed, would it be possible that we experiment on this.

May be lets init the disks and see if the blocks are same and then try to mount as a shadowset and see what error we get?

rgds
Mobeen

[Mobeen_1]

Uwe,
Correct me if i am wrong. I was all along thinking that

"If there is any change in the disks, then they will have different part types"

Its news to me that we could have different disk types under the same part number :-)

rgds
Mobeen

[Uwe Zessin]

I have just asked one of my colleagues to make sure I don't halluscinate and he agrees with me.

Now, I don't claim that the different disks report different geometries - I have given up taking a snapshot of every modell number I encounter. E.g. BD07264546, BD0726459C, BD0726536C all were sold as 232432-B22.

I agree with you that Piet should just try it out.

[Jan van den Ende]

Piet,

independant on geometry, if the _TOTAL_ number of blocks is EXACTLY equal (after initialisation), THEN (and before &.3-2, ONLY then) can you shadow them in a set.
btw, is your SAN EMA or EVA? If you get to EVA, there IS not even "a" drive (with a geometry), it is the SAN manager who presents you a "logical" drive, that physically may be spread out over many drives. As long as you get your SAN manager to present equal-sized chunks (VMS will see drives), then you can shadow. Needs at least VMS 7.2-2Hx. Starting VMS7.3-2 you can add bigger drives to an existing set, but there are still some issues that make it unattractive in production environments. "To be addressed in a future release" (probably some patches coming up).

All in all: on 7.2-2 you SHOULD be able to shadow ANY drives wiyh the same "total blocks after INIT", irrespective of geometry.

PS. ik heb nooit reactie gehad op mijn (prive) mail tav je vorige probleem. Is dat niet aangekomen, heb je geantwoord en is dat niet aangekomen, of is er wat anders?

jpe

[Jan van den Ende]

PS,

are you really sure of the disk specs you gave?

I notice even numbers of sectors per track, and an odd number of total blocks?


jpe

[Willem Grooters]

If I remember well, this issue came up in the discussion on shadowing dissimilar devices in 7.3-2, during the TUD in the Netherlands last year. I even do recall some collegues (same company) did recognize the situation that two disks in a SAN were initialized equally, but the size finally DID differ - just one, or a few blocks - causing inability of shadowing. The problem would be solved with this (then new) feature.
But yes - 7.3-2 at least...
(I know the slides should be available somewhere for Interex-NL members, but I cannot remember where...)