EXPERT RESPONSE
Unfortunately, the rule of thumb here depends, but here are some quick things to consider:
- Make sure you take into account all packets and cell overhead when you figure out your utilization figure. A "155Mbps" OC3 ATM link will be 100% saturated at about 130 Mbps (depending on packet size distribution).
- If the average utilization is 70 or 80% you can be pretty sure that the peak utilization is 100%.
- When your load peaks to 100% capacity you will experience loss and delay for some fraction of the traffic. The impact of this loss and delay is going to be application dependent.
- If there is no QOS policy implemented in the network, then this loss and delay can hit any and all applications and users. So if you run a network with no QOS, you will need to be more conservative with the utilization guidelines. Some people would recommend targets as low as 25% average utilization.
- If you have a QOS policy, then high-priority traffic can be protected from the impact of traffic peaks. If you have low-priority traffic which is elastic (i.e. it can expand to fill whatever bandwidth is available -- bulk data backup over TCP would be an example), then you may be able to engineer the network for much higher average utilizations (along the lines of the 70-80% you suggest).
The best way to determine whether you have enough bandwidth is to track the loss and delay for each application and/or QOS class. This can be done on a budget with any modern ping implementation that allows you to set DSCP markings. If the application is being impacted by high loss and delay, then it's time to upgrade the bandwidth. You may also want to consider WAN optimization. By using data reduction techniques you may be able to carry several times more traffic over the same link. At the same time most WAN optimization devices also make it easy to implement QoS policies and track application and network path loss and delay statistics.
|
