Scale Out or Scale Up? Backup is Not Primary Storage When It Comes to Architecture Design

Posted by
[vc_row][vc_column][vc_column_text][mk_dropcaps style=”simple-style”]A[/mk_dropcaps]t Quantum, we have been delivering both approaches for decades with our scale-out products, engineered for large and data-intensive workloads, and our scale-up approach for data protection. Why both? Because Data Protection and Production functions have different requirements.

Before we focus on the data protection side of the world, let’s make sure we all understand the differences between “scale-out” and scale-up.”

Scale-out, also known as horizontal scaling, generally refers to adding more backup appliances (CPU, RAM and storage) in order to literally scale infinitely.

Scale-up, also known as vertical scaling, generally refers to adding more capacity to a single backup appliance (add storage).

Why choose the “scale-up” approach for data protection?

“Scale-out” makes sense for Production environments, as they often need to simultaneously increase performance and capacity when dealing with unpredictable workloads.

Imagine, for example, that your marketing team creates a fantastic promo. While this will impact your Production system in terms of performance, it won’t affect at all your backups. Data Protection is not Production. They follow different life cycles, even if they have some common constraints.

Why we didn’t choose the “scale-out” approach for our DXi deduplication appliances, even if we know how to create horizontal scaling solutions?

Here are four important reasons:

Reason #1: Why add CPU and memory when all you need are more TBs of storage?

Deduplication appliances have been created to store more backups in less space. That’s the main purpose of deduplication. Of course, it has to deal with the backup windows and so on (we will come back to that in a minute), but it’s all about doing more with less. With a scale-out approach, every time you run out of capacity, you’ll need to add a new appliance, hence adding CPU and memory when all you need is TBs. But you’ll also add a new device that you’ll need to manage, and connect to your network and host, consuming power, cooling, switching and data center space. Don’t you want to save data center resources for your critical workloads instead of wasting them?

Reason #2: Don’t assume all appliances are created equal in terms of performance

What about ingest performance? Deduplication means nothing if you can’t ingest fast enough to meet the backup window. One of the benefits of the scale-out approach is being able to predict performance. Have a 5 TB/h box but now need 10 TB/h? Just add a box and there you go. Sounds easy. When you select an appliance, you need to do a proper sizing with the vendor. What kind of data are you protecting? How many applications are your using? What’s your retention? Protocols? Available bandwidth? And more. This is an exercise you’ll have to do whatever the approach, scale-out or scale-up.

I have heard countless times that scale-up appliances can’t keep the same performance at all capacity points. Perhaps that was true a decade ago. Nowadays, CPUs are powerful, storage IO connections are constantly improving and Quantum DXi has a secret ingredient: StorNext.

StorNext is Quantum’s proprietary file system and certainly one of the most powerful in the industry.

StorNext is the heart of our scale-out products for production storage, and we implemented it in our DXi Deduplication appliances. This is the secret sauce that allows DXi appliances to keep the same performance whatever the capacity points. Again, why add an appliance when you can deal with your backup window using the one you have? In particular, when this appliance is scalable?

Reason #3: What about hardware failures?

While scale-out offers more fault-tolerance options in the paper, let’s be pragmatic. What is the component that fails more often? A controller or hard drives? A few year ago, the Computer Science Department at the University of Carnegie Mellon, for example, found that the relative frequency of drive failure was 49.1% versus 0.6% for the controller.

Because hard drives are often the weakest link, it’s not a question of if they will fail, but when. Most (read “all”) deduplication appliances are still using traditional RAID technology, even with high-density drives that require weeks to rebuild. A rebuild is a resource-intensive process when you use a traditional RAID configuration, resulting in slow backups and missed backup windows. DXi uses DDP, not RAID, which enables faster rebuilds and lower impact during rebuilds.

Reason #4: Forklift upgrades

And what about the forklift upgrades? Is it true that it’s a nightmare to replace a deduplication appliance? Consider the diagram below.

A three-step, easy-to-apply process; just let the replication work for you in background.

At the end of the day, there is no clear answer, as every environment is unique. At Quantum, we believe that scaling-up with a solution like DXi can make sense: no need to add an appliance to increase performance, no need to add CPU/RAM/space/cooling/power/network when you need to add few TBs to deal with data growth, and last but not least, no need to keep older appliances and pay for more expensive maintenance contracts!

The good news is that DXi is the only deduplication appliance with capacity-on-demand. We ship more TBs than needed, which can be activated at any time with a license key. Still a big fan of scale-out? No problem, most modern backup applications can make a scale-up appliance look like a scale-out one. Just look at Veeam SOBR (Scale-Out Backup Repository).

To learn more about our industry-leading deduplication systems, check out[/vc_column_text][/vc_column][/vc_row]

Leave a Reply

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