More on Oracle and VMware–including other must-see resources.

The blog post on EMC’s own journey on virtualizing Oracle (we saved $5M vs. our old configuration, and got 20x more performance) has gotten a lot of feedback (thank you all!)

There was one piece of feedback I wanted to call out in particular before we get to more resources.   Phil Jaenke does a spirited defense of RISC-based approaches to Oracle here.   His main point (IMO) is that people with aging RISC/large-endian systems should consider refreshing those rather than the alternative I presented (replatforming to x86 and virtualizing).

There will be a time to do a point by point discussion with Phil on the topic.   His main point was around Oracle licensing.   He noted that our low CPU utilization number means that we could have saved more by either having less UCS hosts and/or VMs, or by keeping it on RISC and using RISC-based virtualization/partitioning approaches.

But, while we will do the point-by-point (Phil – shall we do it at VMworld?), I actually AGREE with Phil more than you would think.

As I noted in the article (these are points that Phil points out also):

1. The old system was an old RISC-based rust bucket – so how fair a comparison is it to a modern x86, virtualized approach?  BUT, as I noted – that’s what I see at many, many large customers.   Many are on old kit, and afraid to move it due to it’s “mission critical” nature.
2. At EMC, we have a top-down mandate to virtualize as much as possible, to do it via our technologies as much as possible, and to do with with approaches that mirror what we advocate for our customers (hint, we think the RISC vs. x86 debate, while still raging in some circles, is winding down to a close where x86 has won).  We are on our own “Journey to Cloud”.   We look at infrastructure choices that don’t fit our overall strategic view of IT as problematic in our environment.   We want a big pool of shared CPU, memory, storage, network resources that enforces infrastructure policy – and that pool hosts our applications.  The more we can get that simple model down, the more we can invest on things that DO create competitive advantage for EMC.   BUT, as I noted – that’s what also what I see at many, many large customers.   Many have taken a view of: “We need to become more cloud-like, more homogeneity at the infrastructure layer, stop building ‘islands of infrastructure for apps’ unless absolutely necessary, and aim for automation of almost everything we can – then focus on new applications, assuming that uniform pooled/dynamic infrastructure.”  And, in many cases, that’s a view driven from the top-down.

I’m sure – absolutely sure – that I have my own bias based on my worldview, so I’d encourage people to see Phil’s which is a different world-view (on this topic).

Most of all – I’m not saying that the approach EMC took (moving the entire Oracle apps and DB stack to x86 and virtualizing as much as we possibly could) is the right answer for everyone.

What I AM saying is that as people look at what to do with their mission critical apps – they should look at all the options, including the approach I’m advocating (replatform from Unix to Linux, leverage commodity x86 hardware, virtualize) along with the approach their legacy and inertia would normally them and Phil mentions (refresh their aging RISC hardware with more modern hardware).

So – you’re going to be seeing a LOT on this front  (experiences virtualizing mission critical apps and the creation of more and more new apps) through the year.   We have a set of resources for you to leverage if you’re having a debate about what do to with your mission-critical apps.   If you’re interested – read on….

Ok – first, the March Webcast schedule has an Oracle on VMware doozy in it – “Top 5 reasons why to virtualize Oracle” presented by an A-team of Michael West from VMware, Sam Luicdo from EMC, and Jason Kotsaftis from EMC.

Also – coming to a city near you is a VMware/EMC roadshow coming to a city near you.   The first 3 locations are these – and you can register here.

Boston, MA
Tues. Mar. 15th Boston Marriott Burlington

Detroit, MI
Thurs. Mar. 17th
Hyatt Regency Dearborn

Baltimore, MD
Tues. Mar. 22nd
Hyatt Regency Baltimore

Seattle, WA
Tues. Mar. 29th
Hilton Seattle

San Diego, CA
Thurs. Mar. 31st
San Diego Marriott Mission Valley

Orlando, FL
Tues. Apr. 5th
Hilton Orlando Bonnet Creek

Sacramento, CA
Tues. Apr. 12th
Hilton Sacramento Arden West

Houston, TX
Thurs. Apr. 14th
JW Marriott Houston

Toronto, Canada
Thurs. Apr. 28th

No matter what – if you’re using Oracle, you owe it to yourself and your business to evaluate the alternatives of virtualizing on commodity hardware!

Comments

[Alex Maidak]

I find this misleading in general. The database tier is a 4 node RAC cluster on the UCS hardware, no vmware involved. You did not "virtualize" oracle db, but virtualized the oracle application server tier.

I'm not impressed. A java application server is about 10x easier to virtualize then a large database.

[Chad Sakac]

@Alex thanks for the feedback.

I'd encourage you to read the post and the document clearly. It's clearly stated several times (in both), that the Oracle 11i apps tier is virtualized, the Oracle 10g RAC DB tier is running on x86 hardware. As I noted in the original blog post, virtualizing EMC's DB tier is one of the key scaling goals of VMware's next major release.

I will tell you this - large scale Java workloads are actually very hard on VMware (and require careful BP planning) based on how they handle memory. Many large scale DBs are successfully virtualized (not EMC's - NOT YET).

[Betsey Bolton, EMC Global Marketing Programs]

You can register for the March EMC Solutions for VMware webcasts here: http://info.emc.com/mk/get/DBM10514-17091_raf_lp?reg_src=PA_Vmware

[nate]

I don't know, I guess it depends on the workload, but my personal preference for Oracle would be to stick to physical hardware for real production systems.

Mainly to get around the limitation of 8 vCPUs in VMware. A 4-socket 48-core(AMD) (or even 32-core Intel) Oracle SE system will just scream, and be dirt cheap compared to Oracle EE. Most folks can get by just fine with SE and I love the per-socket licensing (though there is a limit of 4 sockets in a system or 4 sockets in a RAC last I checked - but with so many damn cores it's not a big deal).

VMware certainly gives a lot of flexibility though if you want a lot of independent instances.

[twitter.com/that1guynick]

A different viewpoint of my own, after reading this yesterday.

http://that1guynick.blogspot.com/2011/03/oracle-on-vmwareyes-again.html

Would love feedback and continued discussion!

If nothing else, thanks for keeping this topic active, Chad!

-Nick

[Jay Weinshenker]

I'd be more than happy to talk to any people who are worried about virtualizing their Oracle environments on VMware. Sure, in VMware currently you are limited to 8 vCPUs per VM (hopefully that'll get resolved soon), but I'd argue most of the Oracle databases out there don't need 8 (or more CPUs). For most of those, virtualizing under VMware and then leveraging VMware's features like snapshots, HA, DRS, etc make TONs of sense.

My main client runs the following Oracle on VMware:
Oracle EBS, Oracle Agile, Oracle UCM, Oracle UPK, Oracle Hyperion, Oracle GRC, Oracle OEM, and a number of other products requiring Oracle databases. At last check, and I need to re-calculate, the client saved over $2.2M in Oracle licensing costs by being able to run all their databases on only a few vSphere hosts.

[Mark]

Intel Xeon Nehalem EX is simply the most cost-effective platform for Oracle EE today. The per-core performance of Nehalem EX is on par with the per-core performance of POWER7 on many benchmarks, but Oracle's core factor for Nehalem EX is 0.5, and for POWER7 it is 1.0.

Nehalem EX crushes SPARC64 and Tukwila in per-core performance.

A 4-socket Nehalem EX system can produce around 2 million tpmC, and can support 1TB of physical RAM.

You do not need RISC/EPIC "big iron" servers to support the databases for a Fortune 500 company. You can run a large company on Oracle RAC clusters of Nehalem EX servers. EMC and Cisco are proving this.