How AppNeta Battles Network Bottlenecks
by May 30, 2017

Filed under: Networking Technology

The age-old bottleneck has plagued IT since the first wired connections were made. Network bottlenecks happen when the communication between network endpoints doesn’t have enough capacity to transfer data without overflowing network constraints. The problem could lie on the internal network, internet WAN or in the destination network, but are no longer commonly in the office. Most offices have a 1GB LAN, and the internet backbone has a high capacity, so the ISP handoff is often where congestion occurs. These aggregation points between disparate networks need to share resources to move longer distances.

One network myth we like to debunk is that bandwidth is the only metric related to finding and fixing bottlenecks. It’s actually the end-to-end capacity of connections that you should measure to find the issue faster. That capacity number tells you the highest achievable network bandwidth you can get on the most congested hop of a connection between client and application. Knowing this information can help provide realistic expectations. Adding bandwidth to a troubled network when the root cause is not congestion will not solve the problem.

Monitoring Boston bottlenecks through Azure

Finding bottlenecks in Azure’s cloud using AppNeta Performance Manager.

Finding the Elusive Bottleneck

It’s harder now to find bottlenecks than it used to be, mostly because of network complexity and ownership. There are so many network paths, and hops along those paths, in a given application transaction. You’ve now got a lot of avenues of investigation to find the network culprit. Before cloud really took off, the standard approach was using leased lines or MPLS, which were always single-provider. It was much easier to know who was at fault. Now, with critical applications running via SaaS or cloud and accessed over the open internet, any carrier could be at fault. You have to consider your entire network infrastructure when trying to eliminate bottlenecks.

Today’s enterprise network includes the cloud provider and SaaS app provider networks you use, and the networks connecting any corporate remote locations or users. Depending on the size of your business, this could be a sprawling, global infrastructure, or a smaller operation. Either way, it’s more complicated than it used to be.

Bottlenecks are often caused by insufficient bandwidth at connection points where large file downloads and transfers occur, as well as by continual data streams overloading infrastructure. If it is your own infrastructure, adding more bandwidth is one fix, but takes time and money and doesn’t really get to the root of what’s causing the bottleneck. If the root cause is on your network, device stats can show you a bottleneck—otherwise you’ll hear from users about it.

You can detect bottlenecks when the root cause is on your own network through device stats, anywhere else you get it through user complaints

Another potential bottleneck-causing problem—which isn’t so common for the average business user—is that network cabling isn’t keeping up with user connectivity demand today. There just isn’t always enough bandwidth available within internet service providers’ networks to get it to the users clamoring for it. Some solutions in the works are continuing to replace old copper cables with fiber optic cabling and optimizing servers and mobile network features.

Preventing and Getting Rid of Bottlenecks

Knowing the location of the bottleneck is the first step toward eliminating the problem, and ideally it’ll help you avoid a recurrence if the root cause is owned infrastructure. Of course, there are some tried-and-true ways to prevent network bottlenecks from happening at all. These include setting up QoS levels and monitoring to see if they’re met, increasing link throughput and setting up port channeling. Now, however, there are other methods in the mix. Load balancing and software-defined routing has become an art unto itself, with dedicated appliances available just for this task.

There’s also WAN optimization and SD-WAN, both of which help make the wide-area network (WAN) run more efficiently without increasing throughput over those connections. And adding redundancy techniques and designing modern network architectures, like leaf-spine, can maximize bandwidth and make bottlenecks less likely.

However, getting rid of bottlenecks does not just involve increased bandwidth. You also need to decrease latency on your network paths. One way to do this is choose your cloud provider carefully. Matching up your remote offices with the cloud provider’s data center that’s closest can take a lot of pressure off of long-distance networks.

We monitor network capacity end-to-end, which gives our users a lot more valuable data than just monitoring network bandwidth in the office. It’s also a lot quicker to find the bottleneck when you’ve got a full picture of your networks in-house along with the public internet and cloud provider networks.

Monitoring your network this way means that you don’t have to overprovision bandwidth for the bottlenecks that come up. It will add efficiency in managing the network to know exactly where the bottleneck is occurring, especially when you’re letting the ISP know how to fix a problem on their end. Another essential part of preventing bottlenecks is knowing exactly which applications are running on your network and when, as well as who’s running them.

A lot of your network infrastructure is out of your control these days, but you don’t have to be at the mercy of bottlenecks.

Network bottlenecks on this dia

Finding bottlenecks in Google’s cloud network.