- This topic has 5 replies, 1 voice, and was last updated 20 years ago by Rommel.
18th October 2001 at 19:37 #33650Todd TannerGuest
We are having a problem with the extended Erlang B vs. Neil Wilkinson Retrial Tables. I have attached an Excel spreadsheet to show you the conflict. In the attached example, we have six T1 PRI (or 140 circuits) serving a Dept. of Defense Contact Center. We have a required service metric of blocking no more than 20% of offered calls. Neil Wilkinson is telling me we are over (P.40) and Westbay is substantially less.
I have the hourly ATB (all trunk busy), CCS and PEG reports from the switch if the raw data would be helpful. What is my actual blocking probability?
Who’s formula is correct? I hope it is yours.18th October 2001 at 19:38 #33651Westbay Engineers Ltd.Guest
Thanks for your note. Your choice of which traffic model to use depends upon the nature of the traffic. Extended Erlang B assumes random call arrivals. Neil Wilkinson does not; it assumes that the traffic is peaked. It is usually used to dimension trunks groups that carry overflow traffic from other trunk groups. Neil Wilkinson’s blocking figure is higher because it thinks the peak traffic (rather than the busy hour traffic) is higher than it probably is (becuase the traffic is more peaky).
The short answer is that you’re not comparing like for like. Unfortunately, we do not support the Wilkinson algorithm at present, although we hope to in the future.18th October 2001 at 19:39 #33652Todd TannerGuest
Thank you for the response. I almost understand it. I use NW to determine “actual” load in trunk groups where I have an “all trunks busy or ATB” status during the measured hour. An “ATB” condition creates an offered load consisting of the normal random arriving calls as well as “peaked” or non-random retrial calls. Your extended ErlangB allows a selectable retry percentage which I assume is subject to the same random and peaked calling simulated loads?
Obviously, I am having a problem understanding the difference between retrial generated loads and peaked calling loads.
Any help available?18th October 2001 at 19:41 #33653Westbay Engineers Ltd.Guest
I understand your situation, and it’s a classic NW application. But it’s not the same as Erlang B. NW assumes that the the peakedness comes from another trunk group that is in an ATB situation. The big peaks (and non-random traffic) occur when the primary trunk group (not the one under analysis) is ATB.
Extended Erlang B is different. It only considers one trunk group and assumes that if a call encounters ATB, that a percentage of those calls will retry. But, the generation of those calls is random, albeit that the offered trafffic is a little skewed when the blocking is high.18th October 2001 at 19:43 #33654Todd TannerGuest
I do not understand how any “retrial” call would be considered random as a “retrial” call by its very nature is non-random and peaked.
[Webmaster’s note: This is an exchange of emails between a customer and our technical support staff. With Mr. Tanner’s permission, we have reproduced it on this forum, as we believe it to be worthy of wider debate.]23rd October 2001 at 23:02 #33655RommelGuest
To help this along, NW is used for overflow trunkgroups, thus giving a peaked traffic pattern. If we look at voice traffic and the models, we would have to choose on a few factors:
call arrival pattern
blocked call disposition
Each of these factors will help us choose a traffic model.
NW model shows peaked arrival pattern, blocked calls retried, etc…
EEB shows random arrival pattern, blocked calls retried, etc…