Tuesday, March 19, 2024

march, 2024

Is Machine Learning the Answer to Network Latency?

Is Machine Learning the Answer to Network Latency? Not yet.

It used to be that every IT team could define and monitor clear network paths between their enterprise and data centre. They could control and regulate applications that ran on internal systems because they installed and hosted all of data locally without accessing the cloud. That level of control afforded greater visibility into issues like latency, allowing them to troubleshoot and resolve the problems quickly. Fast forward a decade and the proliferation of SaaS applications and cloud services has complicated network performance diagnostics to the point that it requires a rethink.

What is the underlying cause of this trend? The simple answer includes added complexity, distance and a lack of visibility. Whenever an organization outsources its data or applications to an external vendor rather than hosting it locally, it effectively adds a third party into the mix of network variables. Each of those points introduces a potential weak spot that can impact network performance. These services have – for the most part – become quite robust and reliable, but outages in a service can now impact millions of users. The simple truth is that there are many moving parts of the network landscape that enterprise IT teams can no longer control.

One way that companies try to ensure performance is to lease a dedicated MPLS tunnel into their headquarters or data center. But this approach is expensive, and most enterprises don’t do this for their branch locations. The result is that data from applications like Salesforce, Slack, Office 365 and Citrix no longer travels through the enterprise data centre because it isn’t hosted there.

Is Machine Learning the Answer?

To some extent latency can be mitigated using traditional network performance monitoring methods, but by its very nature, latency is unpredictable and difficult to manage. So, what about using artificial intelligence? We’ve all heard examples of technology taking great strides by adopting some form of machine learning. Unfortunately, however, we’re not at a point where machine learning techniques can significantly minimize latency. We cannot accurately predict when a specific switch or router will become overloaded with traffic. The equipment may experience a sudden burst of data, causing just a millisecond or even ten milliseconds of delay. The fact is, once that equipment is overloaded, machine learning isn’t yet able to help with these sudden changes, which typically result in a queue of packets waiting to be processed.

The solution is to fight latency where it impacts users the most – as close to their physical location as possible. In the past, engineers used Netflow and/or a variety of monitoring tools at the data centre, knowing that most of the remote traffic hit their servers and traveled back to their clients. With so much more distribution of data today, only a small percentage of remote data hits the servers, making monitoring from the DC far less advantageous. Rather than relying solely on this kind of centralized network monitoring model, IT teams should supplement their traditional tools by monitoring the data connection at every remote location or branch office. It’s quite a shift in thinking compared with today’s practices, but it makes sense: if the data is distributed, then network monitoring needs to be, too.

Applications like Office 365 and Citrix are good examples because most of us use various productivity and unified communications tools on a regular basis. These applications are likely to connect to Azure, AWS, Google or others, rather than the corporate data centre. So, if the IT team doesn’t actively monitor that branch office, then they completely lose sight of the user experience at that location.

For some organizations latency can lead to a host of cascading problems, especially when large data files or things like medical records are being transferred from one location to another. An obvious casualty of latency is VoIP call quality, characterized by frustrating conversations that don’t flow naturally, but it can also be a major annoyance for large data transactions such as database replication, causing regular processes to take much longer than anticipated.

Focus on end-user experience

There is no doubt that the proliferation of SaaS tools and cloud resources has been a boon for most organizations. The challenge facing IT teams, then, is to rethink their approach to network management in a decentralized network. One important issue certainly is the ability to effectively monitor that service level agreements (SLAs) are being met. More importantly, however, is the ability to ensure quality of service for all end users. For that to happen, IT pros need to see exactly what users are experiencing, in real time. This kind of shift to a more proactive monitoring and troubleshooting style will help IT professionals resolve network or application bottlenecks of all kinds before employees or customers even notice them.

Savvius

Jay Botelho
Jay Botelho
Jay Botelho is the Senior Director of Products at Savvius, Inc., a leader in actionable network visibility for network performance management and security investigations. Jay holds an MSEE, and is an industry veteran with more than 25 years of experience in product management, product marketing, program management and complex analysis. From the first mobile computers developed by GRiD Systems to modern day network infrastructure systems, Jay has been instrumental in setting corporate direction and specifying requirements for industry-leading hardware and software products. He is based at Savvius’ headquarters in Walnut Creek, California.

Related Articles

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Stay Connected

1,595FansLike
0FollowersFollow
24FollowersFollow
2,892FollowersFollow
0SubscribersSubscribe

Latest News