VMware is really expensive

costcalcA while ago somebody forwarded me a research paper from an “independent” research firm in which the cost of VMware and Oracle VM were compared. Interesting!

Now you might wonder why, as someone working for EMC, I would care about such comparisons. Why would I be bothered by VMware in the first place?

Well, full disclosure coming up, but here a few things you might or might not be aware of:

  • EMC aquired VMware in 2004 and is the largest share holder of VMware
  • EMC and VMware are separate companies with separate Profit & Loss, etc
  • EMC and VMware have a strong partnership
  • Both companies are free to work with other partners

This means VMware will work with non-EMC storage vendors (and they should, if for nothing else, to keep us sharp) and vice versa, EMC works with other virtualization solutions (like Microsoft Hyper-V and even Oracle VM for Solaris and Intel).

So being an EMC person, I could not care less if a customer would use Oracle VM (OVM) or VMware as long as the primary objective is met: cost savings and service level improvements. Now I leave the discussion which one of the two is better, technically speaking, to the people from both respective companies. But sometimes I get the feeling Oracle positions OVM only when they are forced to (due to pressure from VMware) and not because they really want to drive cost savings forward for their customers (otherwise they would have used OVM within their Exadata machine to achieve much higher database CPU utilization ratios). It is simply not in their best interest.

So why do I care? I explained this to our sales people once that this is a three-stage rocket:

  1. Virtualizing Oracle Databases brings huge cost savings and significant operational benefits for our customers
  2. VMware is the best platform to make this happen
  3. EMC has the best infrastructure and integration to run VMware – and Oracle for that matter  (OK, I admit, biased opinion but this is what I believe)

In the white paper, and frankly, I’ve seen the statement coming from many different sources, it is claimed that VMware is much more expensive than Oracle VM (mainly because Oracle VM is free – sort of).

So although technically speaking this could be true (VMware offers a free product as well called ESXi, but for large scale consolidation you need features like live migration and workload management tools, which means you need something like VMware Enterprise or Enterprise Plus).

The problem with research like that is that they manipulate your perception. In this case, by silently excluding database or application licenses from the comparison (which would make sense only if you don’t plan on virtualizing any applications). So I tried to make a comparison myself: given a typical database stack, how much is the cost of VMware compared to the total license cost?


  • An Intel-X64 environment
  • Dual-socket, 6 cores per socket (common mid-size server)
  • Only looking at DB and Virtualization licenses. I kept OS licenses, storage options etc. out of the equation to keep things simple
  • Oracle Enterprise Edition with the common options: Advanced Compression, Partitioning, Diagnostics & Tuning pack
  • VMware Enterprise Plus (the most high-end, therefore most expensive license, just to make my point)
  • List price for both (but with equal discount it does not make a difference)
  • No maintenance included, just purchase price

I’m not going in details of the whole calculation, but all the sources are on Oracle’s and VMware’s websites to make this calculation yourself.

This was my outcome:

relative cost chart

As you see, VMware takes up less than 2% of the total license cost of this server. As food for thought, try to imagine what happens if you add Oracle RAC to the mix, or Active Data Guard. Or go from 6 cores per socket to 8 cores per socket (more common these days, and remember VMware is licensed per socket, not per core, so the more cores, the lower the relative cost).

By the way, did you know that with VMware, SuSE Enterprise Linux is included?

So, the title of my post was just a teaser. Hope I got your attention :-)


Update: You also might want to check Chuck’s blog post on Pluggable Databases vs. Database Virtualization

7 Responses to VMware is really expensive

  1. Hi Bart, Oracle has reputation for over pricing it’s products so in percentages the hypervisor will always be peanuts. I would like to challenge you with a different question:
    What is the amount of Money spent on VMware in this case?
    And if you replace VMware with another hypervisor what is then the amount of money spent?


    • Bart Sjerps says:

      Hi Maarten, long time no speak!

      For VMware cost I used http://www.vmware.com/products/datacenter-virtualization/vsphere/pricing.html
      There you will see that for Enterprise Plus the current list price is $ 3,495 (per socket). This is the most expensive and feature-rich option. vSphere Standard might do the job as well for $ 995. These are list prices, street prices might be much lower. You can find out what you need by comparing the options here: http://www.vmware.com/products/datacenter-virtualization/vsphere/compare-editions.html

      Now the vSphere licenses are not really for the hypervisor. You can get that for free with ESXi, or XEN, or OracleVM, KVM or others. The value-add is in the way you can manage and move workloads. I don’t know those that well but, as I am looking at virtualizing Oracle DB specifically, a few thoughts:
      – I would not put Oracle on Hyper-V – at least not for mission-critical production. Not that it would not be possible but I don’t think support would be up to speed. But prove me wrong ;)
      – I would not even consider the classic KVM and XEN for now (for prod databases) – until I can be assured the stack is supported by all parties.

      Which leaves only two reasonable options for now, VMware and OVM (which is basically XEN but supported by Oracle)

      – Like you say licensing is peanuts compared to database licenses. So if you can prove that VMware is, say, 2% more efficient than OVM then you made the business case for VMWare over OVM. How do you prove that? That’s a whole separate discussion but here a few starting points.
      a) Oracle has no interest at all in virtualizing databases. It kills their license cash cow. So don’t expect them to help you all the way to get there. They might even create some roadblocks (like they do with VMware). So if you wanna go OVM, do me a favour and at least find another partner to help you
      b) VMware explained on their blog why their product is better: http://blogs.vmware.com/apps/2012/04/oracle-vm-4x-more-marketing-4x-less-substantiated-facts.html

      Of course this is a biased reply to Oracle’s marketing machine so I recommend you check all the facts.

      Bottom line: the virtual platform that drives the highest consolidation (i.e. the most databases/vm’s per CPU) while achieving rock solid SLA’s (i.e. availability, performance guarantees, other useful management features) wins. License cost is almost completely non-important in this context.

  2. Pingback: Als je Oracle virtualiseert, kijk dan vooral naar de Oracle licenties « EarlyBert

  3. Hi Bart,

    Great post. My questions are regarding performance and DB cost. I run a benchmark to check how an Oracle Database would perform in a virtual environment versus a physical and also compared the database performance (including data load, transaction per minute,response time at db level) using different hypervisors, and as a result it clearly shows that using Oracle VM the database performs better than using other hypervisors and in some cases even better than bare metal.

    The other point is Oracle Database Licensing, due that Oracle only recognizes hard partitioning if using Oracle VM, it means (If using your scenario of a server with 2 sockets with 6 cores each – and Multithread is on) but creating as per example the VM for the Oracle Database using 4 vCPUS it will need for database license purpose 6 processors license if using VMware and only 1 processor license if using Oracle VM.

    As a conclusion it makes the deployment of Oracle Database on Vmware more expensive than using Oracle VM, or not?

    I have nothing against VMware, (by the way, I love the product) but we are talking about who is more expensive to run an Oracle database here.

    • Bart Sjerps says:

      Hi Francisco,

      Sorry I didn’t catch your comment before. But you seem to be very aware of what you want to achieve, you’re definitely on the right track.

      A few comments and thoughts. First of all, at EMC we have a preference for VMware for technical and non-technical reasons. But if you can achieve the same cost savings with other virtualization approaches, then why not.

      If Oracle VM performs better then that’s certainly interesting. Do you have any clue on explaining the difference? And it’s certainly strange that virtualized performs better than bare metal. Our own experiences with VMware indicate around 4% overhead (but getting better with each new VMware version). If virtualized performs better than bare-metal and you cannot explain why, there’s probably something that undermines consistent results. Bare metal should always perform at least similar to virtualized – all other things equal.

      However, a few thoughts. First of all, it’s important how consolidation works out in the end. What I mean is, say you can prove that OVM performs a bit better than VMware for a single database. Does that hold when consolidating 20 VMs per physical server? If so, then performance wise OVM would certainly be a good alternative from a performance perspective.

      Second, manageability and availability. How do other platforms behave in terms of live migration (i.e. VMotion)? Being able to live-migrate flawlessly is crucial to success. Having automated tooling to do this based on actual workload even more so (VMware DRS/HA and equivalent features on other platforms). Can you set CPU shares (i.e. give one VM preference over another one i.e. segregate prod workloads from test/dev)? Does it work well? Can one runaway VM choke another or can you do some I/O fencing? Does the platform have good integration with storage, networking? Do you have good chargeback/showback tools? How do you prevent “VM sprawl”?
      How do you manage hundreds to thousands of storage volumes if you scale up the number of VMs? (not claiming this is easy with VMware, it gave me a bunch of headaches for sure, but you might find it a bit more evolved than others currently, and best practices are more widespread).

      Is your organization planning to put everything on OVM (i.e. Microsoft, web services, etc) or are you now going for a split architecture (OVM for Oracle, VMware or something else for the rest of the landscape)? You might find yourself spending much more because of this fragmentation than you are attempting to save by using OVM for a portion of the landscape.

      I’m not up-to-date with latest OVM features (I should kick the tyres once more I guess) but earlier versions were nice for R&D but not up to the task (yet?) for mission-critical production.

      Then licensing. That is very often the most significant consideration. Interestingly it is to see that technically VMware allows CPU limiting just like OVM – it is just implemented a bit different (and named different) – and Oracle tries to lock out customers here by not honouring VMware affinity. That said, if you can prove (during Oracle audits) that your database never has been running on non-licensed CPU cores then legally you’re OK (but prepare to have heavy arguments with the Oracle reps).

      I know customers who will not risk such a discussion and asked me for alternatives. Not saying it’s easy but on a few occasions I was successful by proposing a blade system (Cisco UCS/Vblock) with a few “smaller” nodes i.e. with only 4 or 8 cores instead of 16 – plus a few biggies for non-database stuff. Make sure the cores themselves are the fastest in terms of TPC-C (i.e. go for Intel E5 with high MHz ratings).

      Finally, and I’ve mentioned this before, a consideration that is 100% non-technical.

      Do you think a vendor whose cash cow is software licensing, will cannibalize this income flow by helping their customers in any way to reduce that licensing for their customers? I have my doubts. So if you go OVM and it works fine from a technical perspective, great, but be prepared to be on your own in figuring out the best ROI reductions. With VMware/VCE/EMC at least we will do our best to achieve the best consolidation ratios possible.

      Admitted, my response is biased (I’d go work for another company if I didn’t think EMC – and VMware with it – were the best of breed) but I hope it helps in your quest.

      Best regards

  4. Nice post, unfortunately some people will still only see the up front cost and not even consider the operational costs into that equation.

    Unfortunately, SUSE Linux Enterprise Server is no longer included with VMware vSphere, though (as of July 25, 2014). SLES will still be the most cost-efficient enterprise linux (not counting free distributions like CentOS or the free version of Oracle Linux) on virtualized environments especially since Red Hat stopped offering basic subscription), a priority (Red Hat equivalent: Premium) costs less than a standard Red hat subscription.

    • Bart Sjerps says:

      Hi Ronny,

      Agreed on the cost issue. But if people don’t consider all cost factors then it’s their own lost opportunity. We can’t do much more than make them aware.

      Thanks for the info on SLES and indeed too bad it’s no longer included.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: