Is Your Data Center Network Aware?

July 20th, 2015 by

What do you do when you get a port buffer utilization alert? Or maybe a better question is do you even get port buffer utilization alerts? In my experience most IT departments separate the network team from the infrastructure team resulting in one epic game of finger pointing. But, there is no way to truly control your network utilization.
Until now. And, we’ll get to that part in just one second. First, let’s look at the way we try to troubleshoot network issues today.

The Problem: Network Guessing Games

Network utilization issues are identified via one of two methods. By far the most frequent method is an alert. An end user calls the support desk to complain that his/her machine is running “slow.” The team then goes and digs through all the necessary steps…

  1. Check the VM
  2. Check the Host
  3. Check the Storage

And assuming that all those pieces check out, the final step…

  1. Point the finger at the Network

Then it’s up to the network team to try and resolve the issue or prove innocence.

Method number two is an alert on the network side. Our friends in the networking industry publish numerous articles on how to properly configure thresholds in order to know about network issues. But, this is just a guessing game.

What happens if the threshold is set too high? Your users are experiencing degraded performance before you get an alert. What happens if the threshold is set too low? You receive endless alerts and can’t decipher signal from noise rendering the system unusable.

Now let’s assume the thresholds get set in the proper goldilocks zone. Now you get an alert. What do you do about it? How long does it take you? At the end of the day even if a threshold is set perfectly to alert you when there is network congestion making the decision on what to do about it is a complicated task made even more complicated by the fact that you have very little time to react.

What Can You Learn from a Network Switch?

This is the exact problem that VMTurbo’s network switch integration platform is meant to prevent. By understanding network port buffer utilization at the switch level VMTurbo is able to determine the actions necessary to prevent network congestion.

For example take the case where an application server communicates frequently with a database server on a different host. The port on the application server host is becoming congested due to traffic generated by other VMs using the same port.VM-Move-Network-Switch

VMTurbo identifies this congestion and makes the decision to move a VM in order to prevent port congestion before it becomes an issue.

Sounds like a simple decision right? But, it’s not. VMTurbo uses a market based approach to determine which VM to move, taking into account all of the relevant information about the demand for compute resources as well as the underlying host capacity.

Consider this, which VM do you move? Is it the VM using the most network resources, or maybe a few VMs using less resources? And, where should those VMs go, all to a host with the least utilized port, or some other combination to get the best results?

Then think about the fact that when you move a VM it’s not just network utilization that’s changing. You’re also changing compute utilization on the host. Identifying the best resource allocation decisions to make across all of the components of network and compute quickly becomes an immensely complex task even in smaller datacenters.

Integration with Arista switches in conjunction with your choice of hypervisor or cloud platform allows VMTurbo to understand all of the relationships within an environment, the demand for resources from applications, and the underlying supply of resources. With these understandings, VMTurbo is able to make complex decisions in real time as demand changes in the datacenter.

Extend Control Across Virtual and Hybrid Environments

Using network switches is just another point of control for VMTurbo. This intelligence can be combined with information from netflow collectors, storage arrays, converged fabric, and an infinite number of other control points in order to build a comprehensive datacenter control system.

Leave a Reply

Your email address will not be published. Required fields are marked *