News from HPE’s Nonstop Division

Public clouds are just as good as the underlying technology they use

The “Best of Nonstop” end of the year review is a time to think about what we accomplished this year and maybe a time to ask ourselves the most fundamental questions about the directions of our industry and how Nonstop is evolving within that landscape. I believe Nonstop leadership in availability remains mostly undisputed. For those of us who know the underlying technology that is required to achieve 5 nines or more, it is easy to spot if a new technology is effectively addressing availability requirements. We know that any solution claiming to be highly available cannot afford to NOT have linear scalability because if it does not, then the availability is not there for your latency SLAs.  Such solution cannot afford to NOT have Single System Image, because complexity is the worst enemy of availability. It cannot offer data consistency as an optional attribute, at least for applications that rely on one single version of the truth, as data integrity is paramount for the applications. Unfortunately, it is easy to see that other vendors are still cutting corners when it comes to providing 5 nines.

So, when another outage happens with a major cloud provider, should we be surprised? First, we should not be surprised simply because public clouds are advertising 3 to 4 nines availability to start with (depending on the services you use). “3 nines” means almost 9 hours of downtime so we should expect visible outages every year. And yes, you can get to 4 nines, but that typically requires you to purchase multi-zone solutions and even if you don’t mind the increase in price, your latency SLAs may start to vary to access a different region and only to still face one hour downtime per year.  To be fair to Cloud Solution Providers (CSP) they can only provide what is available to them as technology and while there is certainly a marketplace type choice offering, this does not mean that there is a full software stack, that streamlines 5 nines availability for the application. The complexity of building a cluster, with no single point of failures, does not go away when implementing it in the public cloud (Remember, the public cloud is just running your infrastructure in somebody else’s data center).

In fact, I think the Open-Source’s led industry, that is leveraged by CSPs, is going the opposite direction. By now it should become visible that the Open-Source universal pattern for distributed systems is to separate the working nodes, from a distinctly implemented server hosting the “control plane”. In such design, the working nodes are indeed distributed and provide high availability, however the control plane itself is a centralized service. Is that central service a single point of failure? Well, it depends, if that control is plane implemented using… yet another cluster! This time the cluster is not to provide any useful work for your application, it is to address the lack of built-in high availability that comes with that control plane. And then the worse news comes out, you have a full software stack to manage, each layer with its own control plane, its own options for clustering, consensus algorithms and so on. If public clouds follow this universal pattern, then I’m not surprised they cannot achieve better than 4 nines when they must manage this:

open sourceOne thing to admit is that when you see that complexity, no wonder you would want another vendor to deal with it! Imagine the set of skills needed! I rather be trained at Nonstop! But the key point is, as you may know, on Nonstop systems, the control plane is not a centralized service that requires its own clustering. Sometimes it is simply a process pair with its millisecond, invisible takeover. Sometimes it is truly decentralized software like SQL/MX. If CSPs want to provide more than 4 nines, they should investigate deploying Nonstop technology to solve what may look to be impossible with other technologies. Nonstop is working on making this easier in the future and that is such an exciting prospect for Nonstop. Public clouds are just as good as the underlying technology they use. Nonstop could change that!

So yes, what a year 2025 we just completed! With the release of the NS9/NS5, KLT and TDE, but we may have an even more exciting year coming up!

Wishing everyone a wonderful holiday season!

 

Yours truly,

Roland Lemoine

HPE Nonstop Product Manager

See you in Orlando

Author

  • Roland Lemoine has been working on NonStop for 23 years and is currently the product manager for database and blockchain languages and development products. Previous experience includes customer support for middleware products, Open Source advocacy and a strong UNIX background.

    View all posts

Be the first to comment

Leave a Reply

Your email address will not be published.


*


This site uses Akismet to reduce spam. Learn how your comment data is processed.