Blog Post

So … do you really need a PaaS?

Stay on Top of Enterprise Technology Trends

Get updates impacting your industry from our GigaOm Research Community
Join the Community!

Whether using a Platform as a Service (PaaS) is a pre-requisite for building and deploying modern applications or overkill has been the subject of much debate of late. Folks on the Cloud Foundry or Heroku(s crm) side of the fence understandably would (and do) say PaaS is table stakes. Red Hat(s rhat), which launched its OpenShift Version 2 this week would certainly argue that PaaS is ready for enterprise use.

Others —  even some who were previously PaaS boosters — privately say that for most companies, PaaS is just too complicated for most humans, including developers, and that there are easier, more elegant ways to do much of what PaaSes promise. One developer who has been a Cloud Foundry user said  “Micro-PaaS frameworks like and PaaS-inspired technologies like Docker are much more popular and are a growing trend.”

Docker is basically a container that lets devs write an application once and then deploy it across many environments.

Be careful what you ask for

Much of this discussion, admittedly, was self-inflicted. I recently asked on Twitter for companies using PaaSes to come forward. And the usual suspects, including Time Warner(s twx) Music, an ardent user of Cloud Foundry, did so. But my issue is that we keep hearing about the same few companies and not a ton from anyone else. Pivotal made a splash when it announced General Electric(s ge) as a backer of its Cloud Foundry effort, but I’m still not seeing a ton of adoption from big, mainstream companies.

Adron Hall, coder, recon and messenger (he’s not kidding about that title, by the way) at Thrashing Code, took that tweet and ran with it, soliciting input and blogging the results. He notes that PaaS may follow in the footsteps of the Application Service Provider (ASP) wave of the early 2000s (he and I are probably among the few who remember the ASP phenomenon). A quick recap: ASPs offered software applications for rent. The boom went bust but resurfaced later in another guise that we know as SaaS. He suggests that PaaS may follow that same route, that the best of that model will survive and perhaps flourish but not necessarily under the PaaS moniker. (If you want to go back in time on ASPs, check out this story from 2000.)

Is PaaS overkill or tablestakes for next-gen app building?

Via email, Hall also noted something I’ve been seeing: many large enterprises see no reason to “move wholesale to a PaaS” since they already have the infrastructure necessary to interact with their application tier. Big banks and consumer packaged goods companies “have systems in place for deployment of all their apps and such. If they think consolidating is somehow helping them, that would be an interesting discussion, as would the reasons that CIOs & CTOs really think they should be moving to PaaS.”

James Watters, head of product at Pivotal, insists that naysayers are on the wrong side of history.

“I think what you’re seeing is just like when people pushed back on IaaS 5 years ago… Docker is popular because it makes the IaaS approach easier but it’s not a competitor to a PaaS, it just shows that IaaS is now the sleepy incumbent.”

He agreed that most companies have not tested out PaaS yet, but said that’s a function of supply, not demand. “The next two years will change that,” he said, when organizations align themselves to this approach and realize what he said are huge savings PaaS can offer in building and deploying applications.

Jared Wray, the founder of Tier 3 who is now the  cloud CTO at CenturyLink(s ctl), is busily converging AppFog and Iron Foundry, two Cloud Foundry-based PaaSes. In his view, big companies need more than what the lightweight offerings do.  “Docker is good for many things but PaaS is about removing the operations from your application — how to do large-scale load-balancing, scale out, and versioning of services…there are a lot of things that the framework of Cloud Foundry does that are above what Docker does,” Wray told us on the most-recent Structure Show podcast,

So that leads back to the beginning: And so I ask (again): Do big companies need full-fledged PaaSes to build and deploy the best possible applications? Please comment below.

14 Responses to “So … do you really need a PaaS?”

  1. memesweeper

    On-premise PaaS needs to remove more complexity than it introduces to be worthwhile. Of the current generation — at least the two I’ve used hands on — this isn’t true; they are underperforming whilst being behemoths in their own right.

    Hosted PaaS is a different kettle of fish — you can get PaaS benefits without having to install, manage and debug the thing. So, if you can live with hosted PaaS for your apps, take that route and enjoy the benefits and avoid the headaches.

    • memesweeper, which ones did you try? I have not been able to try the 3 i am looking at.

      You do make a good point. It will take effort to implement an OSS PaaS. That being said, so does IaaS and VMs and physical servers etc. If a company is small, it probably is not work the effort. Of course that, again goes for the things mentioned before.

  2. Good article. To me, where we are with PaaS is very reminiscent of where we were with Java Application Servers back in the 90’s. Many companies grappled with whether Servlet engines and POJO were sufficient, and whether complex application servers were overkill. The issue was clouded then, as it is now, by the complexity of the more emerging alternatives, and the fact that the market is slow to adopt a consistent approach. Once the app server market began to standardize around the J2EE spec, vendors were able to focus on optimizing and simplifying, and organizations became more willing to adopt the concept. Ultimately, the App Server advantages of simple session management, automatic clustering and load balancing, etc. won out. I expect the same thing may happen in PaaS, though the Cloud introduces a wild card, and it is hard right now to pinpoint a standard that can be the rallying point for the vastly different approaches we see out there today.

  3. It’s really quite simple in my opinion. Ask yourself this, “Do your teams spend way too much time trying to build and deploy your application(s)?” This includes typical environments during the normal Application Life Cycle(ALM), dev, qa, perf etc… This is especially a critical step that needs to be “fast” if you are trying to reach the Continuous Delivery model.

    If this is a yes, you should use a PaaS.

  4. I’m working on a project that should be deployed on PAAS. Due to its nature (real time stream data analysis, data mining algorithms, high availability and low latency), my concern is that the PAAS offers will not handle such demands.
    Looking at the heroku dynos size (1GB RAM and 2 CPU maximum) it seems to me very low for my application. Perhaps you’ll argue that it can be scale horizontally, but for data mining algorithms sometimes you need both RAM and CPU. What happens if your application consumes more that 1 GB of RAM due to its complexity that cannot be split?
    Another concern is that the fact PAAS suppliers provide external cloud services that cannot fulfill data locality((in order to prevent network glitches, thus latency) near by your applications. How you can deal with this? You might say, find a different service, but what if you cannot find it?
    Another major concern is related with the versions, patches of the used components. How is this handle?
    As a conclusion. PAAS is a promise, but in my opinion, it still not mature for facing real enterprise applications.

    • Spico F, i suggest that your use case (analyitics) is not the intended target for what is commonly known as PaaS. You need something like what netflix has done with their “PaaS”. It is not a product. It is a set of tools.

      For you concern about PaaS and external cloud – that is really determined on how you deploy your PaaS and on what you deploy your PaaS. Some PaaS [product] suppliers just supplied a product – you gotta deploy it yourself.

      For your concern about versions, etc – see how CloudFoundry and Apprenda do it.

      As for your last comment about it not being ready for real enterprise apps, i again point you to Apprenda.

  5. Sacha Labourey

    Hello Barb,

    Asking whether companies need “PaaS” is kind of like asking 15 years ago whether companies needed middleware. Could you build applications without any middleware? You sure could! Now, was that the best path forward to manage your security, pooling, clustering, transactions and messaging yourself? Probably not.

    Raise the abstraction a cloud higher and you get your answer for PaaS. Can you simply instead rely on IaaS, stitch things together your environment with scripts, install your middleware layers, setup your load-balancers, firewall rules, make it an elegant whole and make sure the full thing is integrated with development environments and that it will constantly evolve in a smooth fashion to your users? I am sure “you can”. The question is whether this makes sense or not. You can also build your servers yourself and produce your own electricity. You can.

    What PaaS enables is for companies to invest their money where it truly matters: where they can differentiate against their competition. Do some IT organisations really believe they are being more competitive because they’re faster at installing Linux?

    Now, I think your article is very broad and mixes (probably on purposes) very different phenomenons. The adoption of public PaaSes such as Heroku, Google App Engine or CloudBees happens along a very different story line than then adoption of private PaaSes such as OpenShift and CloudFoundry.

    PaaS and Love,


  6. Thanks for posting this piece Barb. Having worked extensively on IaaS projects, I wonder if the PaaS value proposition has been inadequately presented. A lot of organizations need to worry about the “icky” stuff of patching, upgrades, vulnerability scanning, anti-virus, backups/restores, continuous monitoring etc. These processes tend to be “hand-stitched” in an IaaS environment. In my mind PaaS should offer a cleaner and smoother path for taking care a lot of the activities described above. I know Derek Collison of Apcera gets this and they are working to address this gap.

    • GP, some of what you mentioned is the job of the IaaS (i.e. anti-virus). Some of is both the job of the IaaS and some the PaaS. Obviously they both need continuous monitoring. What the PaaS provides, besides deployment is things like scaling – so it will provision (and deprovision) IaaS VMs as needed.