Very Computer

Hi Folks,

Anyone out there have any data on using Veritas VxVM on luns
created with rm6?  More specifically, large RAID0 luns are
created on separate A1000/A3500 arrays, those luns that are
now on the host are then put under Veritas Volume Manager
control, where they are sliced and mirrored.

I feel like such a setup has a considerable overhead due to
each disk/lun access having to transit through the different
volume mangement protocols.  That is to say, there are two
different volume managers that each disk/lun access has to go
through to complete.  One would expect that scenario to have
some latencies and I have noted that in my test cases.  I just
would like to know if there are any other such setups out there
and how are they working, performance-wise?

I realize that there are other ways to configure the rm6 luns
other than RAID0, but that's what we have to deal with for now,
so please let me know if you have any ideas/experience on how
such a setup worked for you.

Thanks in advance, best regards,

mike

sure. It's fairly common to do things like this for better redundancy.

Quote:>I feel like such a setup has a considerable overhead due to
>each disk/lun access having to transit through the different
>volume mangement protocols.  That is to say, there are two
>different volume managers that each disk/lun access has to go
>through to complete.  One would expect that scenario to have
>some latencies and I have noted that in my test cases.  I just
>would like to know if there are any other such setups out there
>and how are they working, performance-wise?

The rm6 isn't really providing significant overhead to speak of.
It's more than offset by the hardware caches (assuming you have
them enabled!) in the A1000/3500 arrays. That's the big win.
The mirroring will provide overhead, no doubt, because you have
to issue 2 write requests, one to each mirrored half. However,
you also get a speadup on reads because it can choose either one
half of the mirror or the other to satisfy the read request in
a round robin fashion.
 There are tradeoffs to any configuration. You just have to
decide what's acceptable for your circumstances. If you did it
all in hardware, you'd either have reduced reliability (because
it will all have to be internal to one A3500 or A1000 yielding
some kind of single point of failure), or increased cost for
a system that has 'redundancy out the wazoo' (tm).

Quote:>I realize that there are other ways to configure the rm6 luns
>other than RAID0, but that's what we have to deal with for now,
>so please let me know if you have any ideas/experience on how
>such a setup worked for you.

>Thanks in advance, best regards,

I wouldn't necessarily worry about adding overhead in the way
you mention. It's extremely minimal, and you can't do much better
without spending a lot more money for extra redundancy.

        Doug