Tuesday, September 27, 2005

Circuit Congestion Is Rarely The Problem With Business IT Performance

My friend Ken Hilving of Hilving & Associates tackles the potentially complex question of broadband performance and congestion with simple story like explanation. Very inventive and certainly informative. Every business should understand this explanation before tackling any perceived issues they may face from what they think is congetsion of their broadband circuits.

....Circuit Congestion Is Rarely The Problem With Business IT Performance....

Not everyone is technical, so let me discuss performance and congestion with a story.

Company A has a product to output. There are 3 workers required to create the product, workers 1, 2, and 3 (W1, W2, and W3). The product is produced in a multistory building. W1 works on the first floor. W2 works on the fourth floor. W3 works on the fifth floor. The process flow is as follows.

W1 initiates production. After a set number of steps, he must travel to W2 to request more stuff to work with. This means he leaves his work station, walks to the stairs, takes the stairs up to the fourth floor, and then over to W2's workstation where he makes his request. W2 takes the request and converts it to a stuff list. He then leaves his work station, walks over to the stairs, takes the stairs to the fifth floor and then to W3's workstation. W3 pulls the stuff from his shelves, and gives them to W2. W2 carries the stuff to the stairs and then down to his work station on the fourth floor. There, he hands the stuff to W1, who carries his stuff down the stairs to the first floor. This process is repeated several times during the production process to output one product.

Business is good, and W1 is unable to keep up with demand. Output is stuck at 2 products per day. A meeting is held with key managers - department heads over the 3 workers. After much discussion, it is determined that all that walking on the stairs is the problem. The building owner is contacted, and a pair of elevators is installed, production doubles to 4 products, and all managers are well rewarded for their keen ability.

Business grows, and additional staff is added to do the same job as W1. Eventually the staff grows 10 fold, but production does not seem to rise above 40 products per day. Another meeting is held, this time with the building owner in attendance. One of the managers comments about how he sees the W1 types sometimes waiting for the elevator. Its another "A-HA" moment, and everyone decides more elevator capacity is the solution. The building manager lays out choices. Company A can add additional elevators at the same cost as the original. However, Company A can choose to add 28 elevators for the same cost as adding 8 elevators. The cost is included as an increase in the monthly rent. The managers choose to go with the 28 elevator solution, and anticipate the huge production output that will result.

Following installation, production jumps to - 4 products per worker per day. No change. Worse, the rent just went up to cover the cost of the 28 elevators.

W1, W2, and W3 are nodes in a network. The building represents the WAN. The production is a distributed application, be it simple file sharing or a database query setup. The stairs represent the original circuits, perhaps dialup, and the initial elevator a T1. The building owner is the telecom provider. The 28 elevator solution is a DS3 upgrade. Moving on and off the elevator is the router or switch function. Leaving and returning to the workstation is the local node communications processing, and the actual work is the application processing.

Congestion, which is represented by elevator wait time, was never the key delay. The work flow and the location of resources is where the delay occurred. If companies were to examine their processing, they would uncover similar situations with their applications. Measured against the total time to complete tasks, the communications piece is typically under 20% of the total and often under 10%. Any improvement in communications performance only improves that small piece of the total time.


This might be clearer with a simple time formula:

Production Time = W1+ W2 + W3 + travel time

As long as travel time exceeds work time, reducing it may add enough time to increase the units produced each day. However, once the total travel time plus travel time for one unit is less than the work time, further reduction in travel will not increase units produced.

In our story, we hit this at 4 units per W1 type per day. I know congestion was not the problem because adding workers did not cause a drop in production for any existing worker. They each continued doing 4 units per day. If an elevator queue was the issue, one or more would have dropped to 3 units per day as workers were added.

For Company A, the biggest time element is now somewhere within W1, W2, and/or W3 work elements. This is where any performance gains will have to come from.


Post a Comment

Links to this post:

Create a Link

<< Home