Operating System - OpenVMS
1748180 Members
4210 Online
108759 Solutions
New Discussion юеВ

Re: Shadowing between data-centers.

 
SOLVED
Go to solution
The Brit
Honored Contributor

Shadowing between data-centers.

In my configuration, I keep my production systems and 2 units of my Production Shadowsets in my Primary Data Center. For Data Protection reasons, I have a third Shadowmember hosted in a different data center ~3 miles away. My storage is all FC with the fabrics spanning both data centers via an OC48 using Ciena FC Gateways.

I was just poking around in the VMS help on "Set Shadow /Site", and I came across the following which I thought was kinda interesting...

....In a Fibre Channel configuration, shadow set members at different sites are directly attached to the system. For the Volume Shadowing and cluster software, there is no distinction between local and remote in multiple-site Fibre Channel configurations....

Does this mean that the /Site qualifier is ineffective in concentrating "Read" requests to the shadow units in the Primary Data Center.

Note:: This question is specifically about the qualifier. I appreciate that simple latency will cause the majority of the reads will be satisfied locally.

Dave.
10 REPLIES 10
Jan van den Ende
Honored Contributor

Re: Shadowing between data-centers.

Dave,

do the math.

>>>
hosted in a different data center ~3 miles away
<<<
Sorry, I have to switch to metrics.
3 miles (FIBRE distancw, I will assume !!) ~ 5 km. ~ 10 km round trip.
Speed of light in glass = 200,000 KM/SEC =>
round trip = 1/20,000 th of a second ; or much less then the loss within any device etc.
This means, that for practical purposes your sites can still be considered "adjacent"

hth

Proost.

Have one on me.

jpe
Don't rust yours pelled jacker to fine doll missed aches.
Jan van den Ende
Honored Contributor

Re: Shadowing between data-centers.

oops.

I hit Sumit too fast.

So, the conclusion is, that in your situation, the SITE mechanism has no noticeable added value.

Proost.

Have one on me.

jpe
Don't rust yours pelled jacker to fine doll missed aches.
The Brit
Honored Contributor

Re: Shadowing between data-centers.

Hi Jan (and any other readers)

Please note my "NOTE". Math is not the issue, the qualifier is.

I am simply asking whether the /site qualifier is meaningful in a SAN fabric context. The fact that my storage subsystems are only 5KM apart is irrelevent, it could be 50, 500, or 5000km,

the question about the /site qualifier remains the same.

Thank you.

Dave.
marsh_1
Honored Contributor

Re: Shadowing between data-centers.

dave,

but that is the reason that you would use this qualifier.

The Brit
Honored Contributor

Re: Shadowing between data-centers.

I agree, that is why I am confused by the statement in the help screen for the "set shadow/site" command.

Dave.
Wim Van den Wyngaert
Honored Contributor

Re: Shadowing between data-centers.

We have 2 sites about 5 km apart. May be the fiber is 10 km. We use shadowing but 1 site is mscp served to the other (not fc). We use /site. No idea if something is gained but our FDDI has limited capacity (100 Mbit) so any traffic avoided is a gain.

For shadowing it makes no difference where the disks are. If you indicate that it must concentrate the reads on 1 disk (via /site), it will do so. Even if it's the slower one (e.g. far away).

Wim
Wim
Volker Halle
Honored Contributor
Solution

Re: Shadowing between data-centers.

Dave,

the intention of this sentence should be to point out, that the volume shadowing or cluster software can NOT distinguish between 'local' and 'remote' disks, so you may want to use the /SITE qualifier to indicate, which disk is at which site.

See a recent 'OpenVMS Shadowing in Action' presentation (page 51 ff.) from Keith Parris:

http://www2.openvms.org/kparris/hptf2005_VolShad.ppt#500,1,HP OpenVMS Volume Shadowing in Action

Volker.
Jan van den Ende
Honored Contributor

Re: Shadowing between data-centers.

Dave,

Traditionally, SHADOWING will issue the READs only to the drives attached to the local system.
And here is the catch: to VMS _ALL_ FC disks are "locally connected".
And that would defeat the read performance advantage of reading only locally __if the time to transfer the request and the answer grow to significant amounts__
As a way to resolve that, the SITE mechanism (and the related COST mechanism) were introduced.
It is thereby possible to tell VMS that a node and a specific drive are co-located, or not, and so re-introduce the read-locally-only principle.
And the value only applies at "significant (in transfer time) distances"

hth

Proost.

Have one on me.

jpe

Don't rust yours pelled jacker to fine doll missed aches.
marsh_1
Honored Contributor

Re: Shadowing between data-centers.

dave ,

i agree it could have been clearer i.e

" ...In a Fibre Channel configuration, shadow set members at different sites are directly attached to the system. For the Volume Shadowing and cluster software, there is no distinction between local and remote in multiple-site Fibre Channel configurations.... so you may wish to use this setting to concentrate read requests on local fc disks"

:-)