If I told you that I could do something to make your most mission critical app:
- Perform better.
- Cost less.
- Have higher overall availaiblity.
…you would either: A) assume I’m a vendor trying to sell you something and write it off as such; b) assume I’ve made an error – and ignore my advice; c) investigate, and if you agree, you would consider that deployment model for the app.
Or, you’d be a person who likes “just keep it the way it is” SO much that you are willing to pay more, for less performance, and suffer more downtime. Ergo, there’s something wrong with you.
Well – here you have it: If you virtualize your Oracle environment on vSphere on EMC, it can perform better, cost less, and have higher overall availability. Oh – I’ll throw in better agility through the test/dev/UAT/QA/Prod process while I’m at it. This is what today’s news is about.
Caveat: Now, admittedly, I do work for a vendor, which by definition I work for someone who is in the business of trying to sell you something. Our ongoing growth and success at EMC and at VMware (see Q2 results for EMC “EMC Reports Record Second-Quarter Revenue and Profit” and VMware “VMware Reports Second Quarter 2011 Results”) means that we’re doing it and providing value to our customers (thank you customers!).
But, on the topic of mission critical apps being moved to commodity hardware and virtualized, I feel particularly passionate (and have posted on it here, here, and many other places) – and think the reasons people don’t move forward aren’t technical. So – forget me as a vendor. Go ahead and ignore what I say. Here’s the voice of the customer:
“In a physical server model, cloning our production Oracle database configuration for test and development was time consuming, complex and expensive, leading to an inappropriate environment that was unacceptable. Virtualizing our Oracle database servers with VMware vSphere® on EMC VMAX and EMC VNX storage system allows us to quickly create as many test and development environments as we need using the same virtual infrastructure in just minutes. In addition, virtualizing our Oracle servers has allowed us to leverage more Oracle instances per physical core with the possibility to run different OS and Oracle software versions on the same host, improving the return on investment in our Oracle software. After the virtualization of Oracle with VMware, our staff was quite surprised to find that Oracle performed even faster as a virtualized application than it did before. I believe that Oracle runs 10 percent faster within a virtual environment. Our ability to virtualize Oracle has saved the bank hundreds of thousands of euro.” - Andrei Maier, System Architect of Swedbank.
Here’s another:
“Through virtualization of our mission-critical application environments, including Oracle, we have been able to consolidate from fifty physical servers to four VMware vSphere® servers, a nearly fifteen to one consolidation. I can now deploy an Oracle test database in under an hour and have it ready to test the latest release of our banking application. Through the use of EMC DataDomain, we have achieved a de-dupe ratio of 20:1 during backups allowing us to scale our virtual environment cost effectively,”…“With our EMC and VMware cloud, IT has become an enabler of growth and services. Accommodating this growth used to be a time consuming and expensive process. Now, we’re able to have server and storage resources ready in a few hours and with no additional cost, when in the past it took weeks. Our agility and responsiveness have improved dramatically.”…
On top of that, they also use VMware Site Recovery Manager and EMC RecoverPoint to support their virtualized Oracle environment – resulting in a much improved recovery time objective (RTO) protects the bank from downtime, taking minutes, instead of hours or longer, to switch over to the recovery location in the event of a disaster. Remember – Oracle Dataguard is fantastic – but doesn’t provide end-to-end DR for the full app – and you need to protect more than just the database.
…“RecoverPoint with VMware SRM is a lifesaver,” Smithey states. “If I have a recovery situation, we just press a couple buttons and RecoverPoint and SRM do the job. And our IT people are the heroes.” Mark A. Smithey, Vice President, technology services at The Washington Trust Company.
I’ll give you one more:
“Before virtualizing, utilization rates were typically in the range of 15-20% for our Oracle database servers. Since deploying VMware vSphere® on EMC Symmetrix VMAX storage platform we have almost doubled our server utilization rates through leveraging more Oracle instances on the same infrastructure, increasing our return on investment. In addition, migrating our Oracle database servers to the VMware platform and Linux has led to increased database performance and we can now leverage the functionality of VMware vSphere for faster server failover and high availability,” - Haim Inger, Chief Technology Officer, CLAL.
So there you have it. Note – each of the customers – in their own words – after virtualizing Oracle on vSphere (even before vSphere 5!) on EMC, they experienced:
- Higher performance
- Lowered costs
- Improved availability
All these customer stories, plus new reference architectures validating these results and de-risking implementations, all pulled together with standardized services to help you in making a custom case for your environment is the news of the day.
Read on for more performance, best practices, details on licensing, and where to go for help…
Q: What’s the scoop on performance?
A: it’s amazing, and perhaps a bit counter-intuitive, but you CAN see improved performance on virtual, even when compared against a straight physical config. I wouldn’t set that expectation with any customer – it’s a very fair expectation that for even large configurations, no more than 10% performance overhead should be experienced.
BUT – the compare is really “vs what you’re doing” – and in those cases, more often than not, a modern Westmere-EX system will smoke even recent RISC-based systems. Remember that yeah, with vSphere 4.1 there are some workloads that exceed the Virtual Machine version 7 maximums, but they are rare. And those dwindle to nearly none with vSphere 5 and virtual machine version 8.
The other thing is that in many cases, the ability to squeeze more utilization (particularly for the massive number of test/dev, light utilization hosts) means better overall performance too… How? Well, with more smaller VMs, each OS/DB platform has memory sizes which tend to be sweet spots, and often this is well below what you can get for low cost configs these days (with very low cost x86 multicore and DRAM). If you slice up those systems into more VMs, you get better overall performance.
This whitepaper covers a very recent round of testing, and has loads of results (not only on performance, but how to migrate, how to make other things like maintaining some physical RAC nodes). BTW – the testing has been going on for many moons now, and of course, there will be an update as soon as vSphere 5 GAs :-)
Net of some of the testing is in the chart below which speaks volumes:
Q: What’s the scoop on support?
A: The short version? It’s supported. The more important version – no vendor can/should represent a support position of another – so ignore what I just said. More importantly, just listen to the letter of Oracle’s support position. Oracle’s policy toward virtualization with VMware is characterized for subscribers of MyOracleSupport, Document ID #249212.1. Gartner, IDC, and others also have documents for their subscribers that specifically address this policy.
There are two things to know when reading that doc:
- What does “certified” mean? Well – Oracle creates “certified” solutions when they are responsible for all the components – end to end. They don’t “certify” any 3rd party hardware. Yet, you wouldn’t hesitate to deploy Oracle on a Dell server right? Even though it isn’t Oracle certified, right? Remember – certified != support. The only certified solution is Oracle software, on Oracle hardware – end to end.
- What does “..customer can demonstrate that the Oracle solution does not work when running on the native OS, Oracle will resume support” mean? This clause – in effect, that they will do their best to provide support, but reserve the right if their fixes don’t work, push an issue that to VMware to support (since a problem that occurs on VM, but not on physical points to VMware) – is very fair, in my view. It’s also very common across ISVs – clearly they can’t be responsible for supporting VMware itself. But sometimes – this gets “blown out of proportion”. There are no known, open issues (there have been in the past). In my interaction with customers, and in our own EMC IT experience, this never is asked. BUT – just in case it ever does… If you’re using vSphere and EMC together – this document describes how you can literally “repro on a physical” in seconds using EMC Replication Manager. It’s a bit dated, but still current in it’s approach. Another technique is to maintain a physical RAC node (more difficult operationally, but means you always can “repro on physical”)
Q: What’s the scoop on licensing?
A: The short version? Talk to Oracle. But – talk to them understanding the facts, so you understand the details.
First core idea – understand Processor Core Factor (PCF) – in essence a base multiplier. You can use this to your advantage. The x86 processors have a PCF of 0.5, half that of modern RISC based processors. But, in reality, modern CPUs like the Westmere-EX based Intel cores offer great (not just “good”) per core performance.
Second core idea: there is no core “penalty” or “tax” for running virtual – the cost per core is the same. Look at two examples of list Oracle licensing cost:
In those two examples – the fact that one is running on top of VMware has no effect on the Oracle license cost. There is, of course, the material cost of vSphere in the TCO that needs to be factored in (and OVM is free of course). More often than not (by HUGE factors), customers on their own conclude that all the benefits of VMware (not just in higher consolidation ratios, but things like storage VMotion, Storage DRS, SIOC, NIOC, reporting, mangement… and so on and so on) are of huge benefit (not just for Oracle, but as the basic way that they do all x86).
Third core idea – yes, DRS Host affinity is not recognized officially (even though it is auditable, logged, and I know of examples where the customer has pushed Oracle on this point) as a mechanism to “fence” off what physical cores could be used. OVM’s hard partitioning is recognized, but then again, you lose the ability to move workloads live – which is a base capability that is at the essence of the “agility” of virtualization.
Keep the pressure up on Oracle. The fact that DRS Host Affinity rules are not recognized as “isolated” is non-technical. But – it still doesn’t matter. For your high license cost Oracle Enterprise Edition licenses – simply create clusters with less, larger ESX hosts (and you can still optimize around the new vSphere 5 vRAM oriented licensing). Use a larger vSphere cluster for all the lower cost licenses like Oracle SE, SQL Server, or other stuff – and remember – if you’re like most customers, you have 10x-20x the amount of infrastructure supporting test/dev/QA/UAT as you do prod :-)
Q: Where can you go for more help in virtualizing Oracle?
A: A good start is here: http://www.vmware.com/solutions/partners/alliances/oracle-vmware-support.html
I also strongly recommend a great VMware/EMC partner – House of Brick. We’ve partnered together with House of Brick with many customers – and had great results. It’s no coincidence that you’ll see their name as the service partner on VMware’s page.
NET: Virtualize Oracle with confidence on vSphere and EMC. You’ll save money, perform better, get higher availability, and be more flexible.
Are you a customer virtualizing Oracle? Whether it’s on vSphere or not, or EMC or not – I would love to hear your feedback (positive, negative, whatever – let’s just keep it courteous!)
Excellent article, especially so in the distinction made between certified and supported solutions. Licensing and support are the main, and oftentimes misinterpreted, factors for adopting Oracle virtualization. These two components can frequently kill a deployment before any discussions of performance or virtualization benefits even hit the table.
Posted by: Scott Aisenstat | July 29, 2011 at 09:28 AM
Excellent post Chad... as usual.
Agreed on support. No one ever asked Oracle to debug VMware issues. The discriminator is really at which point they hang off the phone. If they do it the moment you report the problem that isn't very good and fair to be honest. If they do as part of the (advanced) troubleshooting in order to rule out the possibility that the hypervisor is the problem .. well I guess that is fair.
Massimo.
Posted by: Massimo Re Ferre' | July 29, 2011 at 10:39 AM
Nice to see more customers doing this. We've been doing this for our customers for years. I'm in the process of migrating some monster Oracle DB's from Sun E25K systems to Linux on vSphere 4.1 and we're aiming at a 2x jump in performance at least (which is the minimum we've seen in the past). Average VM size 3 vCPu and 32GB RAM, With a bunch in the 6 vCPU 94GB RAM range. Will use 1.1TB of RAM for 56VM's on 6 vSphere hosts. Whole cost of the project, including hardware, services and licenses is less than 50% of the annual maintenance of the old systems. How's that for ROI!
Posted by: Michael | July 31, 2011 at 08:22 PM
Chad, great post, and thanks for the mention of House of Brick. For any of you that will be at VMworld, plan on attending the Oracle on VMware superstar panel (BCA1548 Oracle on VMware Panel Discussion) Wednesday at 12:30 PM. Dave Welch, House of Brick's CTO and Chief Evangelist will be on the panel.
Nathan
Posted by: Nathan Biggs | August 05, 2011 at 05:03 PM
We are beginning our migration from sparc to vsphere for oracle. The biggest problem right now has to do with DR. The DBAs like to have log devices recovered at a different point in time with recoverpoint. SRM wont let us do this. We are thinking about taking all our Oracle DB VMs out of SRM but that defeats a lot of goodness we get by virtualizing Oracle. Anybody cracked this nut?
Posted by: Duane | September 07, 2011 at 05:47 PM
@Duane - there is no solution (that I'm aware of) that's super-clean) to have your cake and eat it too on that topic.
Near term choices (well proven):
- use in-guest mechanisms (iSCSI, NFS).
- use SRM for everything else, and use the callout/script mechanism (improved in SRM5) for the Oracle DBA where it's important for more sophisitcated recovery.
Longer term, we're working hard to crack that nut in a better architectural way (see VMworld session VSP3205 - working to get a blog post on the topic.
Let me know if we can help architect the solution...
Posted by: Chad Sakac | September 08, 2011 at 09:58 AM
When we use SRM for DR, do we have to have the DR site fully licensed from Oracle's perspective (all DB and App server licenses)? Oracle folks are saying, the customer should be fully licensed on the DR site even if is passive but if the Oracle DB or App code is installed. As far as SRM, there is nothing that is installed on the DR site and the whole VM images will be migrated as part of the failover. Does it still warrant a full license on the DR site?
Posted by: Srajarat | November 10, 2011 at 02:20 PM