To continue reading for free, register below or login
To read more you must become a member of SearchNetworking.com
It depends. When answering this question, it's usually balancing the law of averages while also managing risk. It's a lot like playing a game of table shuffleboard where the object is to get into the scoring zones without having your puck end up in the gutter. Except in this case if you miss, end-user connections may start to experience performance problems due to discards or drops. So the question becomes how precariously close to the threshold can I get without dropping packets, causing performance problems, and affecting the end-user response times?
For reference I've included the method for utilization:
| Data Volume / (duration * Speed) (Speed = 1.544Mbps for a T1 circuit) |
As a word of warning, be sure to watch your units. Data Volume tends to be stored in Octets. Be sure to either convert them to bits or convert the speed to MBps.
Also since I'm doing in this a scientific proof method (my science teacher would be so proud), my assumptions include that currently this T1 circuit has no QoS policies or class maps defined that would allow you to manage the traffic that gets dropped more effectively. I'm also assuming that you don't have a lot of real-time/VoIP/multiplayer gaming traffic requirements. These applications are more sensitive to packet drops and congestion issues.
The first thing that you want to evaluate is over which period of time are you looking for average utilization? A T1 circuit at any given instant is either in use or it is not. If you were to poll the data volume transferred every second, your utilization numbers would be more precise than polling every 5 or 15 minutes. However polling a device (or using NetFlow to estimate bytes transferred) at second-level granularity would probably put too much strain on the device so most customers tend to compromise on 5 or 15 minute intervals. For the given time interval, you want to evaluate the utilization. If it's 80% utilized over 15 minutes, it's a fairly heavily used circuit. You might consider looking at the NetFlow data to identify what traffic is being transferred over that circuit and identify if QoS or other mechanisms might help alleviate that traffic concern.
You might think that 80% utilized over a 15 minute interval isn't all that heavily utilized, especially in these economic times where we want to squeeze every penny out of these circuits. One thing of note that you want to be concerned about is microbursts. The serialization delay of an interface determines how quickly the packet gets placed on the wire. If there are too many packets either in the egress buffer waiting for access to the wire or a server/app sends a fast rat-it-tat-tat of packets in a small amount of time, the egress buffer on the T1 circuit can discard the packets. So by trying to keep under 80% utilized in 15 minute intervals, you're precariously close to keeping the T1 circuit busy while allowing for some leeway for microbursts. This is just a rule of thumb but hopefully now you're empowered to identify how deep into that egress buffer you want to get before the end users start to complain. May the force be with you!
|
