Do multi-tenant environments still create noisy neighbors?
A noisy neighbor isn't just a real-world problem. Learn how a noisy neighbor can affect workload performance and how the public cloud has changed to address this issue.
Everyone has a noisy neighbor story, like the suburbanite who mows the lawn at 6:30 a.m. on a weekend. Unfortunately, this issue isn't reserved just for people who live in close proximity to one another. Cloud users sometimes deal with similar frustrations.
In the public cloud's early years, the notion of sharing resources was new and vendors hadn't yet worked out the kinks to prevent decreases in performance. Today, the noisy neighbor is mostly a historical artifact, though analysts seem to agree that it's still something that could crop up now and then.
The effect of noisy neighbors
A noisy neighbor is defined as one party monopolizing the sharing space in a multi-tenant environment -- an issue that has become commonplace for IT teams.
In the noisy neighbor scenario, one party overprovisions compute, network and storage infrastructure based on expected demand and workload behavior in a multi-tenant environment. Everything performs as planned, until a workload spikes in activity and starts to consume resource capacity beyond its typical behavior. As a result, other workloads that are sharing the same capacity can experience a performance impact.
"This problem has been around since the mainframe and it came along for the ride as businesses raced towards the public cloud," said Mark Bowker, analyst at Enterprise Strategy Group.
Every IT organization has had this problem -- the difference is some plan for it better than others, Bowker said.
Compare single-tenant and multi-tenant environments.
Noisy neighbors and containers
The noisy neighbor problem was resolved for customers of the major cloud vendors, said Deepak Mohan, analyst at IDC. Over the years, they've grown more capable of managing operations, shifting loads and rapidly responding to performance issues. Additionally, with hyperscale providers, users have access to many specialized options to minimize these concerns, including virtual private clouds and dedicated connections. Other powerful resources, such as larger instance types and auto-scaling tools, are also readily available if a workload needs them.
"Maybe other, smaller providers in some areas might have that noisy neighbor issue, but as far as hyperscale providers, I haven't heard of an end user in the last two years expressing that concern," Mohan said.
Sid Nag, research vice president at Gartner, has been thinking about the noisy neighbor problem for a long time, initially in terms of networking and traffic engineering. Nag noted that inquiries on the topic have become less common, but that doesn't mean the problem has disappeared entirely.
As far as hyperscale providers, I haven't heard of an end user in the last two years expressing [a noisy neighbor] concern.
Deepak Mohananalyst at IDC
For instance, while traditional concerns about multi-tenant environments focused on an entity monopolizing bandwidth or CPU cycles, the widespread adoption of containers could shift those concerns.
With containers, the operating system is virtualized -- unlike the VM model. As a result, slices of the OS are dedicated to multiple tenants, which creates its own set of challenges, especially around security, Nag said.
However, the chronic lack of visibility into containers means IT operations and technical teams may not be able to recognize a noisy neighbor issue in multi-tenant environments.
Turn down the volume
First and foremost, it is up to the customers to actively monitor performance of any application that runs in the public cloud. "Cloud providers guarantee availability and have SLAs for that, but if you notice a performance decline it is up to you to raise a red flag," said Dave Bartoletti, a Forrester Research analyst.
If it's an app that's used across a corporate intranet, users are probably not that worried about slight variations in performance. But, if it's an ecommerce site, those variations can be a big concern that would support the argument for using a dedicated or isolated machine.
However, it may not be simply a compute problem. Customers need to see what kinds of shared resources are being used. For instance, you could have a dedicated server -- physical or virtual -- that is sharing Amazon S3 storage. "In that case, a noisy neighbor could still be an issue if somebody is running a big S3 job," Bartoletti said.
If you do find you have a problem, Bartoletti recommends working with your cloud provider to find out what it takes to get access to different kinds dedicated infrastructure -- where you don't have to worry about any potential noisy neighbors in the cloud.
Smaller companies -- even web-hosting companies -- sometimes offer dedicated infrastructure. Furthermore, IBM and Oracle have tried to differentiate themselves based on the additional support they provide for bare metal. AWS has since added its own bare metal offerings, while Microsoft recently rolled out a dedicated host option on Azure.
"These days, pure public shared everything and pure dedicated, hosted bare metal are meeting in the middle," Bartoletti said.
