RE: EMC Powerpath and ASMLib

  • From: "Matthew Zito" <mzito@xxxxxxxxxxx>
  • To: <jeremy.schneider@xxxxxxxxxxxxxx>, <raj.mareddi@xxxxxxxxx>
  • Date: Thu, 5 Mar 2009 07:49:43 -0500

Assuming you're using EMC storage, and not the generic powerpath support for 
third-party arrays, you will see improved performance and failover with 
PowerPath versus devmapper.  The reason for this is that PowerPath actually 
uses native commands to query the array for the optimal path, while devmapper 
simply round-robins and looks for command timeouts.

If you know you're always going to be using EMC storage, I would recommend PP.  
Plus, devmapper actually had most of the same ASMLIB issues that powerpath did, 
it's just that powerpath's got more "press" because it's much more widely 
deployed.  There was at least one CRS bug in early 10gR2 releases that was 
specifically for DM.

However, both solutions will work fine for the core of load balancing, 
failover, and path recovery.

Thanks,
Matt

-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx on behalf of Jeremy Schneider
Sent: Wed 3/4/2009 8:52 PM
To: raj.mareddi@xxxxxxxxx
Cc: oracle-l@xxxxxxxxxxxxx
Subject: Re: EMC Powerpath and ASMLib
 
We did spend a lot of time on this decision. (We will potentially deploy 
up to a few hundred clusters based on our design.) We decided to avoid 
Powerpath and use Linux built-in multipathing (devmap-based) on RH5. I 
think that ASMLIB is fine with devmap-mpath; we chose not to use it but 
I think that there are some trade-offs both ways... most people probably 
will use it.

There are some good arguments for devmap and I think I remember a fairly 
recent endorsement of it on the ocfs2 list; I also wrote up some 
speculation about reasons against it on a blog post not too long ago 
over at ardentperf.com.

-Jeremy

Raj Mareddi wrote:
> Fellow DBAs,
>
> We gotta make a decision to use raw devices of powerpath or ASMLib 
> over powerpath. I know ASMLib historically got some issues with 
> Powerpath(cause ASM Couldn't understand mutlipathing). But it sounds 
> like these issues are now fixed with Linux kernel 2.6 and 2.0 ASMLib. 
> However, The EMC white paper for best practices doesn't mention ASMLib 
> at all.
>
> Just curious on what you guys using ? or What option would you use and 
> why ?
>
> (I personally prefer raw device names of power path (by using pseudo 
> names), just as explained in 
> http://www.oracle.com/technology/products/database/asm/pdf/asm-on-emc-5_3.pdf
>
> I am sure lot of you are already through this situation. Would you 
> please share your thoughts and experience ?
>
> Thanks,
> Raj Mareddi


-- 
Jeremy Schneider
Chicago, IL
http://www.ardentperf.com

--
//www.freelists.org/webpage/oracle-l



Other related posts: