Understanding the Cost Premium of Public Cloud for Active Data

November 1, 2016 - 27 Comments

If you’ve read any of my past blogs, you’ll know I focus a lot on the economic impact of Cisco UCS. Today I’m comparing the TCO of hosting data intensive workloads on-prem vs. in the public cloud.

What started out as a 2RU rackmount consolidation discussion (more on that next week) turned into a public cloud vs. on-prem discussion. I was asked the question of how a S3260 managed by a customer in their own data center or colo would compare to Amazon S3 storage. I was completely blown away with what I found.

The short story is the on-prem solution is 56% less expensive over three years with a 13 month breakeven point. If you don’t like CapEx, using a Cisco Capital lease, your monthly payments can be less than your payments to Amazon. [I wonder if this is why companies like Dropbox, Moz, and HubSpot are moving away from public cloud?]

Let me break this down for you.

Amazon publishes all of their pricing here. That makes it very easy to price out a configuration. I priced out 420TB of S3 storage in the US East N. Virginia region. I included an always on 10Gbps connection and business class support. Note there are no data transfer charges. The breakdown is below.

Amazon S3 Pricing Table

Next, I configured a S3260 with 60 10TB drives for a total of 600TB of raw storage. Why am I comparing that to 420TB of S3 storage? Overhead – RAID, file system, spare drive, and the fact no one would run at 100% utilization. I used our MSRP for our solution which is effectively an apples-to-apples compare to Amazon’s published web price. Your mileage may vary based on discounts from Cisco Partners or from Amazon.

S3260 Pricing Table

There is enough savings that I could buy a second and third S3260 (1,800TB!) and still have some change left over.

If you need more compute power, you can replace four hard drives in a S3260 chassis with a second server node. What’s 40TB among friends when you are doubling your CPU and memory capacity?

Best part of the S3260 is that it is managed with UCS Manager right alongside B-Series blade servers and C-Series rack servers. One tool, one set of processes and procedures to manage your environment regardless of form factor.

While we are all tempted by instant gratification, it’s pretty clear that storing data in the cloud comes at a premium over the long haul. There can be very good reasons for data to live in the cloud. Just as there are very good reason why data should be on-prem. It will depend upon the cost, application, speed of access, privacy, security concerns, etc. The list is a long one and you will need to decide where the best place for your data is.

Hopefully I’ve piqued your curiosity enough reach out to your Cisco Account Team or Partner to learn more about the S3260 Storage Server.


In an effort to keep conversations fresh, Cisco Blogs closes comments after 60 days. Please visit the Cisco Blogs hub page for the latest content.


  1. Another Server Guy,

    Thanks for sharing your experience. If you are new to UCS, there are several special programs offered to give additional discounts. I would ask your partner about those. But you bring up the most important feature of S-Series to me – UCS Management. You can use the same tools, processes, and procedures to manage the S-Series just like the B-Series and C-Series.

    Did you see my blog on consolidating multiple 2RU rack servers on to S-Series? http://blogs.cisco.com/datacenter/do-you-need-more-storage-but-fewer-servers. One of the comments was about reduced software licensing. If you have an enterprise or site license, this probably isn’t a factor. But if not, it should also be added into the equation.

  2. This TCO really isn’t any surprise for those of us who design on and off prem solutions. But I wouldn’t want to put my `active` critical data on a single box like this. It’s a stressful business hosting critical data and I’d rather give that to someone else who can offer those 9’s – even if it does cost loads more.

    But I love this box for Veeam and Commvault so you should create a TCO for on-prem Veeam repositories (2016, ReFS Storage Spaces) against Veeam Cloud Connect providers offerings, as that will show you MASSIVE savings as IaaS cloud based backup repo’s are very expensive.

    So the solution are comparing in this blog is all wrong, you should target it at Backup, CCTV, Archive solutions – Not general purpose `data intensive workloads`

    • Another Server Guy,

      Thanks for adding to the conversation. I would never recommend a single box for any solution. I myself practice the 3-2-1 rule of backups – three copies, two media, one off-site. I’m also glad you like the product. There are many different ways to add to the analysis I’ve done. Both Veeam and Commvault (as well as others) offer solutions with us. My goal with this compare was to start a conversation not to try and cover all the possible permutations. Your comments as well as all the others have done just that. Cheers.

      • After reading, I ended up getting a quote from our reseller – it was for 58 x 6TB HDDs

        It costs a little more p/TB then a similar spec Dell R730xd – But I guess its more dense , uses less power and cooling than multiple smaller boxes and also plugs into existing UCS management software.

        Not enough savings for us to consider changing, but worth considering for those already with UCS.

        • I am Bill’s colleague and must re-emphasize his comment, people buy Cisco UCS for the value it brings operationally. Most storage optimized servers are just sheet metal with a bunch of drives where vendors compete on lowest price. None offer this level of infrastructure management to make adds, moves and changes zero-touch using our embedded policy based automation framework with UCS Management.

          Also keep in mind this is all done inband over data links “less cabling” via the top of rack switch. This TOR switch gives you a single point of management to any number of servers physically connected via a centralized GUI or programmatic API. When plugged in UCS servers are automatically recognized by the management system as additional resources whereas traditional servers are bricks requiring much more effort to stand them up and get them online and available for use.

          This is the value of a unifying compute, storage, networking and management into a single stack. Check out this oldy but a goodie that covers this point much better than I can.


  3. Interesting blog and discussion.

  4. So Bill,

    The MSRP game/comment is exactly why amazon is appealing to so many. Are you a huge company able to negotiate volume discounts or packaging together other crap? Hey, MSRP and Cisco have you covered.

    But what if you aren’t? Hello channel and margin hogs.

    • otherRob,

      Thanks for joining the conversation. I’m confused how your comment applies here. If a customer did not negotiate in any way and wasn’t getting any discounts but was still paying less for the Cisco solution than the Amazon solution, is that wrong?

      In Amazon’s latest quarter, AWS as a whole is reported to have 101% higher operating profits. Sounds like they have some margin as well.

      Let me know if I missed something.

  5. Think that you need to update “crack”under the DC overhead section of the BoM to read CRAC or computer room air conditioner. That appears to be what server-dude is referencing.

    AWS S3 storage automatically provides two current facilities worth of data resiliency. Please update your BoM to include those two secondary sites, the software to do the synchronous replication prior to acknowledgement to the end user that the object has been written, and the network links between the sites as all of these charges are baked in to S3 standard.

    Also, where have you figured in the cost for exposing these services through a portal/api/and cli? I guess UCS director could do this but you would need to work code the integrations with the tools i that you mention earlier – Veeam/ Commvault/Object storage partner etc. and whatever piece of software (could you show that cost please) you plan to leverage to maintain your synchronous replication between multiple sites. Please add the PS costs estimates of those integrations.

    Finally, I find it interesting that you’re even mentioning backup partners in the comments at this point as to compete with S3 in an on prem model. With a 99.99% uptime model you’d only be allowed to be down for 52.5(ish) minutes in a year. To recover from catastrophe leveraging a backup partner in that timeframe (and this wouldn’t be my first choice – you’d need at least a replicated site in a real world scenario), you’d surely need backup disk on prem (and replicated) and people who manage that. Can you show the decision that you’d make there to meet that SLA and those associated costs?

    • Ron,

      I really appreciate your time leaving such detailed & constructive feedback.

      I’ll put together and go through it in more detail and will hopefully be able to respond late next week when I’m back in the office.

    • Ron – also fixed the CRAC. Thanks for that!

    • Mr Bill is very conservative with his numbers. No one pays MSRP, do you ? Real 3Y TCO for customer for a fully populated server is about $40,000 USD, plus power and cooling.

      • PerfGuy,

        You are exactly correct. Customers would pay much less. I can’t comment about the price you quote as it would be up to the customer’s account team or partner. My main point is we purposely stayed away from an ASP and used MSRP which would be similar to Amazon’s web price.


  6. Thanks for some interesting comparisons.

  7. Steve.

    I moved your last comment to the trash as it was making a personal attack and I didn’t deem it professional. If you’d like to restate it, I’m happy to continue the conversation.

    You appear to be posting from an Amazon IP address. I’m very open about who I am and who I work for. I would appreciate the courtesy if you would do the same.

  8. What kind of ‘crack’ are you recommending with these S3260 servers? I’ve never seen that from a Cisco rep.

    Aside from that, your comparison is a simplistic / rather naive comparison.

    • Server-dude,

      Thanks for taking the time to read and comment.

      The comparison was simplistic by design. I don’t think is was naïve however.

      I was trying to distill the comparison down to the absolute minimum. I could have included more things on both sides.

      If you like a more in-depth compare, reach out to your Cisco account rep or partner and lets really dig in.

  9. Are you seriously comparing a single server to S3 which has 99.999999999 durability and automatically replicates data across multiple data centers? And have you considered any of the features you get with S3– encryption at rest, integration with IAM, versioning, lifecycle management, static website hosting, cross region replication, APIs, etc?

    • Steve,

      Thank you as well for taking the time to read and comment.

      Yes, I’m serious.

      I’ve never seen that many 9’s in reference to S3 storage :-).

      As I stated in the blog, there are reasons to go to the cloud and there are reasons to keep things on-prem. We have a number of partners that offer the features you mention along with our own API. I would stack up our hardware and our partner’s (Veeam, Commvault, etc.) software against the cloud. But then, I’m biased.

      Also, as soon as you replicate the data across Amazon’s regions, data charges would be incurred. I didn’t include any data charges.

      I think everyone should take a look at all of the solutions available to them and make the determination what is the best fit. This blog wasn’t a feature comparison but a price comparison. I was trying to keep things simple. I just wanted people to be aware that price shouldn’t be an issue when considering on-prem storage.

      • Veeam, Commvault…these are backup apps, not object stores.

      • You say that this blog was “not a feature comparison but a price comparison”– this is a nonsensical statement. How are you comparing the price of a single server against a solution that has 11 9’s durability, replication across multiple data centers within a region (with no additional cost) and all of the other features I mentioned in my previous comment. Why don’t you do an honest comparison and add multiple UCS servers in multiple data centers to mimic S3 durability, then add all of the partner solutions you mentioned (Veeam, Commvault, etc) and see what the price looks like then.

        • Steve,

          I’d like to point everyone to this blog post: https://blog.cloudsecurityalliance.org/2010/05/24/amazon-aws-11-9s-of-reliability/. It actually address the issue of 9s much better than I can.

          I also pulled this directly from their site:
          Q: How reliable is Amazon S3?
          Amazon S3 gives any developer access to the same highly scalable, reliable, fast, inexpensive data storage infrastructure that Amazon uses to run its own global network of web sites. S3 Standard is designed for 99.99% availability and Standard – IA is designed for 99.9% availability. Both are backed by the Amazon S3 Service Level Agreement. [https://aws.amazon.com/s3/faqs/]

          All I can say is I stand behind my numbers. I can add two more servers and may still have room for software. I would defer the question on pricing since there are so many partners to pull from.

    • Steve, Amazon’s “eleven nines” of “data durability” is among the most creative bits of marketing nonsense ever invented.

      When asked to explain the math they used, Amazon has refused.

      However, I set out to ‘reverse engineer’ the claim. It wasn’t hard.

      If you do the math on the chances that two cheap disk drives (in two different locations) will both fail >>at the same instant<<, and then you further assume that both of these instantaneous failures are catastrophic, such that it is impossible for a data-recovery service to do a forensic recovery from EITHER of the two drives, then this goofy 'eleven nines' number pops out of the equation quite unsurprisingly.


      In terms of availability, Amazon claims "designed for four nines", but their SLA guarantee doesn't cost them anything until they drop down below three nines to two nines.

      So, in reality the comparison Bill is giving us is perfectly valid.

  10. Interesting but certainly not an apples to apples comparison. A single piece of hardware vs an Internet-scale managed service. Akin to comparing owning your own car vs worldwide Uber. Two vastly different scenarios.

    • Rob,

      Thanks for taking the time to read the blog.

      When I said apples-to-apples, I was only referring to pricing.

      I tried to address the scenarios towards the end of the blog. Yes, on-prem vs. cloud storage are two different use cases. We believe there are reasons for both. It just that price shouldn’t be one of them.



    • > Comparing owning your own car vs worldwide Uber
      yes, this is more like owning a car, all the flexibility. Uber is great for infrequent use