- Fluke Networks
Download Paper or Go
To Comments/Discussion
(Webtorials registration required. Click here if you forgot your username/password.)
Today, nearly every network of any size has some sort of system for monitoring its performance. These Network Management systems have been around for many years and are critical to the network support team's efforts to deliver reliable services. A wide variety of choices are available, ranging from freeware to systems that can cost millions of dollars. Some systems are provided by equipment vendors, while others are vendor-independent.
In general, all of these systems monitor the status and performance of key devices and links on the network. Examples of the types of measurements tracked include:
But users don't complain that "the interface has too much utilization" or that "the router's memory is low". They say, "the network is slow", because response time is what they care about. The problem is that this sort of network management doesn't measure time at all. All these measurements don't really provide you an understanding of what end users are seeing any more than studying each individual tree can tell you about the size and shape of the forest.
Download Paper
(Webtorials registration required. Click here if you forgot your username/password.)
(Webtorials registration required. Click here if you forgot your username/password.)
Today, nearly every network of any size has some sort of system for monitoring its performance. These Network Management systems have been around for many years and are critical to the network support team's efforts to deliver reliable services. A wide variety of choices are available, ranging from freeware to systems that can cost millions of dollars. Some systems are provided by equipment vendors, while others are vendor-independent.
In general, all of these systems monitor the status and performance of key devices and links on the network. Examples of the types of measurements tracked include:
- Key interface utilization
- Device CPU and memory utilization
- Protocol mix on key links
- Device settings
- Device availability
- Status of processes on servers
But users don't complain that "the interface has too much utilization" or that "the router's memory is low". They say, "the network is slow", because response time is what they care about. The problem is that this sort of network management doesn't measure time at all. All these measurements don't really provide you an understanding of what end users are seeing any more than studying each individual tree can tell you about the size and shape of the forest.
Download Paper
(Webtorials registration required. Click here if you forgot your username/password.)
This is a very instructive paper on managing the end-user view and expectations. As noted in the paper, "The sole reason for an enterprise network is to deliver applications to end users. Traditional network management focuses on devices and links, not the end user experience. Some network management platforms attempt to address this through a central polling approach, which can be ineffective in today’s modern networks."
This also brings up an interesting question on the end-user experience. In my 25-plus years of working with network performance, I've noted that "better than normal" service doesn't exist. Rather, there is "normal" service and "degraded" service.
Which leads me to ask the question, is there a rationale for intentionally delivering consistent response rather than always delivering the best possible service? That is, do you ever intentionally not deliver as fast as possible response so the service will still be "good" under normal circumstances.
I'd love to hear your thoughts on this - and on how you might accomplish it. (For example, intentionally delaying acknowledgment of received messages in some cases.)