RE: vmware & Oracle

From: Matthew Zito <mzito_at_gridapp.com>
Date: Thu, 19 Jul 2007 14:01:04 -0400
Message-ID: <C0A5E31718FC064A91E9FD7BE2F081B1B97934@exchange.gridapp.com>

 

It is a fair point that it adds an extra layer, but they've invested a huge amount of effort in mitigating that. Also, it matters *what* VMWare you're running. Both workstation and server run their virtual machines as processes - so there is a huge userspace jump, and IO is definitely impacted.  

In VMWare ESX server, the OS of the physical machine is a customized version of Linux with a custom kernel, and the virtual machines have a much more direct IO path, as well as typically the VMWare filesystem on the physical machine, as opposed to a standard ext3 filesystem for vmware workstation, etc.  

So, if you look at IO impact statistics from someone, make sure they're comparing the version of VMWare you'd want to run.  

Matt

---

        From: oracle-l-bounce_at_freelists.org
[mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Allen, Brandon

	Sent: Thursday, July 19, 2007 12:52 PM
	To: Sean.oneill_at_organon.ie; List, Oracle-l Freelists
	Subject: RE: vmware & Oracle
	
	
	I have no experience with it, but this comment from the list

archives indicates that vmware slows down I/O by adding an extra layer between Oracle and the disks:
http://www.freelists.org/archives/oracle-l/01-2007/msg00094.html          

        I'd be curious to see specific numbers from someone running a fixed set of Oracle I/Os (e.g. a large table scan & a large index range scan) against the same hardware with and without VMWare. Anyone on the list done anything like that to get precise timings?          

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

Received on Thu Jul 19 2007 - 13:01:04 CDT