2006 Application Delivery Discussion

Adding bandwidth has been the industry traditional solution to response time problem. The reason is that it worked. The problem now is that it doesn’t work as well as it did in the past. Application protocols are the reason. With Client/server, the server sent to entire screen down at once. With HTTP and object, the objects are sent in a sequence with additional delays if the object has to be authenticated. The sequencing means that is a lot of dead time, time the adding bandwidth can’t squeeze out. Bad applications protocols that introduce dead time are also a problem with Microsoft’s application protocols CIFS and MAPI. The answer is no bandwidth but a more intelligent network. Add to this that application are outsourced over the internet and more workers are mobile – you can’t add bandwidth in these situation but need smarter networks.

Robin hit the nail on the head. The other interesting trend is that many of our customers are beginning to leverage 'direct to the net' models. In a recent poll within our customer base, we found that there was an equal percentage of customers using business critical applicaitons hosted on the internet and within their own data centers. This model is also something Gartner Group sees as signifcant as they have forcasted that direct internet connections from branch locations will reach 50% by 2009. Obviously if you can vector that Internet traffic off of a corporate WAN and still maintain security .. all the better.

Any analysis of bandwidth vs 'boxes' is important. However, if you can cache content on a LAN next to a user - there is no amount of affordable WAN bandwidth that can match the performance. With that said - there are some pretty clear minimums for everything, such as VoIP, etc. Therefore it's important to optimize what you have and also manage it.

If you have no clue what’s happen on top of your network, bandwidth is your option. I usually start all my appl. perf. / appl. hosting / CoS discussion with two words, visibility and knowledge. Without this two, bandwidth is the only option. One of the good 'bonus benefits' with appl. perf boxes are the reporting tools which gives visibility to appl. traffic on top of the network. Most companies I talk to does not have enough knowledge what’s their network is used for and how it is used.

When you have visibility and knowledge how application works you are ready to do more then 'just' bandwidth. This discussion happens on the application side too, more CPU/disc or rewrite/redo the application structure. There is only one exception where the visibility and knowledge rule can be omitted in some cases; Add bandwidth when the link latency is low, really low and when bandwidth is cheap.

Best Regards
- Per Håkansson, CCIE 2446.

Netli's presenter Tim Knudsen and the CCIE's comment are on point. Global SME's only have partial control, until the CE of their network. The PE through the WAN is an area of uncertainty that must be managed for web based mission critical applications to have LAN response. The Netli solution seams to solve the challenge by, at least, clearly defining the causes.