Telecommunications Operations Resiliency:
Labor Shortages and the V oice Network
Andjelka Kelic!
E. P. Mathews’
Walt Beyeler’
'Critical Infrastructure Modeling & Simulation, Sandia National Laboratories,
PO Box 5800, MS 1137, Albuquerque, NM 87185-1137
2 Bell Laboratories, Alcatel-Lucent, 600 Mountain Ave, Murray Hill, NJ 07974
Tel: (505) 844-5266 / Fax: (505) 284-3850
akelic@ sandia.gov, epmatthews@ alcatel-lucent.com, webeyel@ sandia.gov
Abstract
Models of the voice telecommunications infrastructure have focused on the availability of
the network during a disruption without accounting for the workforce necessary to
provide repair and recovery functions for that network. This paper describes a system
dynamics model of the maintenance operations of the voice telecommunications
infrastructure and explores the effects of large and prolonged worker absence on the
ability to keep the infrastructure operating. Analysis shows that the voice
telecommunications infrastructure is highly resilient to the loss of a large portion of its
workforce.
Keywords: telecommunications, labor, operations, PSTN
Introduction
The National Infrastructure Simulation and Analysis Center (NISAC)! is intended to
provide DHS with an effective understanding of the performance of fourteen critical
infrastructures, especially under extraordinary circumstances. NISAC draws on industry
experience and knowledge regarding individual infrastructures, developing new models
to represent the processes and cross-infrastructure interactions that may precipitate or
control disruptions.
Detailed models of the voice telecommunications network have been constructed to
explore the effects of the loss of particular pieces of the infrastructure on congestion and
the ability of users to make voice calls (O’ Reilly, et al[2006], Jrad, et al [2005]). These
models have not included the repair and maintenance functions that keep the network
operating and respond in the event of a large scale failure. The rate of restoring the
' The National Infrastructure Simulation and Analysis Center is a joint program at Sandia National
Laboratories and Los Alamos National Laboratory, funded and managed by the Department of Homeland
Security’s (DHS) Preparedness Directorate.
system is controlled by the rate of moving, installing, and testing equipment. In order to
estimate outage duration those processes must be represented.
In September 2005 Sandia National Labs and Bell Laboratories, working in partnership,
began developing a model of telephony operations that included human resources,
warehouse replenishment, and the driving forces of exceptional network damage and
routine maintenance and repair. By the end of the year, the Telephony Operations model
had been designed and the initial version built. Simulations were run, including a
baseline and several disaster scenarios, to validate the model and to understand its
weaknesses. This paper documents the model, baseline data, model structure,
assumptions and the effect that absenteeism has on repair and recovery.
Overview of Model Structure
The Telecom Operations model consists of three interconnected systems: the network
infrastructure state, workers, and the warehouse models. Figure 1 and Figure 2 show the
governing loop diagrams represented in the model. One of the key elements of interest in
the model is the amount of functioning equipment in the network; this is shown in Figure
1. This figure shows how equipment failure affects the perceived damage in the network,
and how material is used in the repair process. Figure 2 shows the causal loops that
govern the response of the workforce to damage. The model includes the effects of
fatigue on the workforce which results in an increase in repair time and thus an increase
in the amount of damage in the network at any given time.
he Equipment = Perceived 5
Failure/Damage Rate
iS
Material Required
AtSie rol - Functioning
— Equipment eg %
~ Tea
Materials Dispatched for Repairs
Shipped
from Warehouses
Repairs Drain Warehouses Requats 8 oj —
Repairs Use Materials
‘Material On Site
Equipment: limited
Repair Rate
Figure 1: Network Failure Rates and Material Supply Overview
Modem telecommunications systems comprise a great variety of specialized equipment
and skills. We balanced the competing demands for parsimony and accuracy by defining
four kinds of critical equipment: switches, frames, transport elements, and local copper
loops; and two kinds of repair workers: network operations center (NOC) workers and
field technicians. The similar structure of the material and worker flows allows us to use
arrays to manage these distinctions.
Workers do not provide repairs for damage in the network, they provide repairs for
damage that is recognized in the network. While this distinction may appear trivial, it is
precisely the gap between occurrence of damage and recognition of damage that is
addressed by the majority of network operations’ center (NOC) support software and
processes. Parameters that control the behavior of this gap provide a model of the tools
and software platform in use by the repair organization.
P Overti
1)
Damage Under. Fatigue Compounds Repair Time
Repair
Perceived Damage — +
+
Workers Respond to Damage — * Workers
Dispatched Fatigue
Dispatch Depletes Worker Pool
Workers Available
Equipment Repair
Rate
* +
Labor- limited Mean Repair Time
Repair Rate
Figure 2: Worker Dispatch and Repair Overview
Tracking the operating state of the network is the driving goal of the model. Failures
occur in the network at varying rates, and are repaired at varying rates. These rates
depend on the type of network element, the status of the worker resources and the status
of the materials to be delivered from a warehouse. Ina disaster scenario, extraordinary
failures occur in addition to ongoing "business as usual" failures. Repairs can be
performed either remotely or not, depending on the nature of the failure, and certain
repairs are performed in two stages ("initial patch" and "repair'). The network
infrastructure state portion of the model represents the state of the network in response to
failure and repair.
The worker portion of the model captures the activities of the human resources involved
in operations. Workers arrive and leave, are dispatched on tasks, become fatigued as
individuals or overloaded as groups. In a disaster scenario, worker presence may apply
different assumptions, overriding "business as usual" process and worker availability may
be severely altered (as during a snowstorm or epidemic). The state of the available
workers can limit the rate at which repairs are initiated and completed, and the number of
failures in the network impacts the dispatch requirements of the workers.
The activity of a human worker making a repair on a network element will require the
delivery of a spare or repair component from the warehouse. The warehouse, as it
delivers spare parts, must replenish its supply from its factory sources without over-
ordering or running short. In the event of extraordinary demand, as during a disaster, the
warehouse delivery processes will throttle the rate at which repairs can be made, and may
cause repairs to be effected in non-optimal order, or cause worker resources to sit idle,
waiting for parts.
Details of the model structure can be found in the following sections.
Network Infrastructure
The network infrastructure portion of the model drives the worker dispatch and repair
functions and is shown in Figure 3. Ina normally operating and maintained network,
critical components fail at a particular rate, governed by the “component damage rate”
variable. Some of the damage can be repaired by operations center workers (see the
variables “front end close time” and “average front end close rate” in Table 1) and other
damage requires a field technician.
After they fail, components become a part of the pool of “damaged infrastructure
equipment” and “infrastructure unreported damage.” These failures can only be repaired
once they are noticed, either by network operations workers through monitoring
equipment, or from customer notification. The rate of damage being noticed is tracked in
the variable “damage reporting rate.” Once damage in the network is noticed, it goes
from “infrastructure unreported damage” to “perceived damage” and can then be
repaired.
The amount of damage in the network determines the workers and material necessary for
the repair. The rate of repair of failures is dependent on the type of component and on
the state of worker and replacement component resources. The default values for
constants in the model can be found in Table 1 and are further discussed in the section on
baseline model runs. The worker segment of the model is described in the next section.
o
Field Closeout,
Rate
Labor Limited Repair Rate
Repair Rate without Repairs In Process
Material Limits. |“?
Undamaged
Infrastructure |= |-—— Rate of Initiating
Hoan _— Workers To
Dispatch for
Percent FE New Damage
Close
omponent Perceived Damage
Re ‘Rate Component
ie Damage Rate
Damage
Field Tech Material Damaged Reporting Rate!
Limited Repair Rate Infrastucture = $I
Equipment
Infrastructure
Unreported Damage
-<Material Limited
Repair Rate>
Figure 3: Network Infrastructure Model Component
Workers
Failures in the network cannot be repaired until they are noticed and the appropriate
resources are dispatched. The worker dispatch portion of the model tracks the human
resources associated with repair and is shown in Figure 4 and Figure 5. Two different
categories of workers are tracked: operations center workers and field technicians.
Operations center workers staff the network operations center and monitor equipment.
These workers can fix problems with equipment that are software related - such as
resetting a piece of equipment. Operations center workers do not require replacement
components on order to solve a problem.
Field technicians travel to the physical site of the piece of equipment and repair physical
problems such as splicing a cable or replacing a piece of equipment. If repair material is
not available, field technicians will not be dispatched to the site of the problem.
Both categories of workers arrive on shift and then are dispatched to perform tasks. The
rate and length of dispatch is dependent on the amount of damage in the network and in
the case of field technicians, available replacement components. Workers dispatched for
extended periods of time become fatigued and their productivity decreases. This portion
of the model allows for temporary removal of personnel from the workforce due to
absence - either in small volume due to ordinary illness or large volume due to epidemic
or emergency conditions. The parts of the model governing the removal will be
discussed in the section on worker absence.
The portion of the model shown in Figure 4 calculates the number of workers that need to
be dispatched to complete repairs in the desired amount of time. The necessary workers
are determined by the “perceived damage”, “workers required per unit of damage”, and
the current “repairs in process”. The total number of necessary workers is then compared
to the current “workers dispatched” to determine how many additional workers are
necessary. The “repair time tolerance” represents how long a piece of the network can
wait before being repaired (in addition to the time it takes to repair that piece of
equipment). Field technicians require materials to conduct repairs and will not be
dispatched if the material is not available. Operations center workers do not require
material to perform their repairs, so will always be dispatched, even in the event of
materials shortages. Operations center workers do not travel to other locations when they
are dispatched, they are simply assigned to a problem.
time to dispatch
Repair Time Desired Worker Masia Wotkers@——*
Tolerance: Disa Ra foray atch Rate rn
Workers Available
Fraction of Desired Rate
Desired Worker Pemnitted by Available
Dispatch Rate by Type Workers
‘New Damage and In
Process Repairs
Maximum Usefil Worker Dispatch
; ‘Addtonal Feld
Rate
Fraction of Workers ~<Workers Dispatched > Tects
Dispatched for New ‘ ry
Net Worker Additions
Workers Retum to
‘Available Rate
Field Teths to Release
for In Process Repairs for New Tasks
; Ze Field Tech Workers
Workers To Disgatch ;
; <__ Workers Required per Needed to Use All
meNew Darerp, Unit of Damage
Necessary Workers for
In Process Repairs
Perceived Sones
Process>
Damage>
Figure 4: Worker Dispatch Calculations
Materials
Y
Workers Dispatched
Workers move through four states in the model as shown in Figure 5. They begin off-
shift, go on-shift (“workers available”), are dispatched (“workers dispatched”) and then
either go off-shift again or return to available status depending on the length of the repair.
Movement among states is governed by work schedules, workload, and fatigue. The
extraordinary event portions of the model are designed to remove workers from the pool
of off-shift workers due to absenteeism. The model does not currently represent hiring
additional workers.
Available Workers
Departure Rate
Workers Available Workers Off Shift
Worker Recovery
Time Workers Time extraordinary Extraordinary
Available works Worker Event
Recovery Rate
Worker Dispatch, <calculated
Rate shift length>
Workers Retum to Rate of Releasing
Available Rate <— ‘Workers From
Completed Tasks Workers A fected by
ro Be Extraordinary Event
omph
Workers Dispatched
fp» going off shit
Figure 5: Worker Dispatch Model Component
Overtime and Fatigue
The model assigns workers to overtime when the amount of damage in the network
increases beyond what can be repaired within the repair time tolerance by the total pool
of workers as shown in Figure 6. Even when there is outstanding damage, workers will
continue at the nominal rate of repair when the total level of damage in the network is
considered to be at normal levels of routine damage. As the amount of damage begins to
exceed routine, workers will begin to extend their normal shift with overtime. Under
severe conditions, workers may double their regular shift up to a sixteen hour shift to
maintain network health; longer shifts are prohibited due to typical labor regulations in
the United States.
As the worker shift length increases, and the duration of extended shifts increases, the
workers become fatigued. For example, a worker can work a double shift occasionally
and productivity will not suffer. However, if workers are continually working overtime,
even for a few extra hours every day, their productivity will begin to suffer and repairs
will take longer. Fatigue is a delayed degradation of worker effectiveness, and creates a
positive feedback to the demand for additional dispatched workers. This formulation is
similar to [Hines, 2005], however the specific function for calculating worker fatigue
based on overtime needs to be further researched and replaced with one from industry
data or what is consider to be standard for the type of repair work performed by the field
and operations center workers.
Labor Required for
Known Damage With
Material Constraints i
Workers? Oe, Pa
Workers Active rn
Repair Time sal rs
wa Field Total Workers Required - ;
to “N —<=—___<Time to Adjust
Dispatched
ae vata fe —
ss Time FE Close
\ ShiftLenghg Total
Workers>
Repair Time FE
Close ~t—Degraded Repair
Time FE Close HousPerDay
Duration of _‘Start time of
are tatign Degradation calculated shift_ ____maximum overtime
Degradation
length,
Effect of fatigue on <ONE SHIFT>
productivityf Time to
Effect of fatigue pital
on productivity nominal shift length
worker overload
Figure 6: Fatigue, Overtime, and Productivity
Warehouse
The warehouse portion of the model tracks the availability of repair components from the
supplier as shown in Figure 7. Both the supplier and the warehouse have desired levels
of inventory, and produce or order supplies to maintain that desired level while being
able to cover incoming requests. The formulation follows Sterman[2000].
~ vannty on ~ amg ‘and Dispatch
i | Onier for ao Retonnce
Reorier Rate fom Waretouse Delver
Time to Place arvhnuse feorder Quantity by q—_
Omer Warhowe Desired Consumption
Desied Tine to Fl Bo gee ee
Ghee hoe Reorder Coverage Minimum Inveni’ry
Warehouses Curent Demand by at Warehouse qiuaieay
Wereoises
Target Supplier
Desired “Inventory |
Prodan [Minimum Supplier Stipment Rate fom Minimum Inventory
¥ _Outpat Coverage Swoker Coverage at
Inventory | Warehouse
Production Gonseliny Matimum Shipment
Time Rate fom Supper
Warhouse
Inventory at Wickes | hooey Shipments foi
‘Supplier ‘Suppliers Replenishment from Warehouse
Production Smee
iA ,
Max Produclon Maximum Consumptio
Rate Rate from Warehouse>
Figure 7: Warehouse Ordering and Inventory
Figure 8 shows the piece of the model that ensures that materials are on site for repair. If
the materials are not available, workers will not be dispatched to perform the repair.
Thus a disruption in the supply chain or an inventory shortage could cause degradation in
the state of the network.
Material Limited
Time to Deploy Repair Rate me
Resource to
Repair A Required per
Materala Materials on Site
OS —— Dispatched al
ial Di ee Material
Material Dispatched Amival Rate
Rate Sr Dann Rate
x :
Re
Figure 8: Warehouse and Material Dispatch Model Component
Simulations: Scenarios and Results
Overview of Approach for Baseline Scenario
We have exercised the integrated model to test its behavior under ordinary and disrupted
conditions. Below we describe the results for a configuration that approximates a generic
mid-sized metro area. Ordinary operations form a baseline on which two kinds of
workforce disruption are imposed: absenteeism and illness. The results demonstrate the
wide range in behavior that the model dynamics can produce, and the way it can be used
to identify constraints on restoration time and effectiveness of mitigating measures under
diverse disruptions. Several of the model parameters are notional, and the model has not
yet undergone extensive testing. The results give a concrete illustration of the model
scope and capabilities, but should not be interpreted as an analysis of a real system.
Baseline Scenario Parameters
The steady state expressed in this model is intended to capture a mid-sized metro area
under normal, non-disaster, conditions. This is the scenario used as the baseline against
which to isolate and analyze perturbations created by other scenario conditions. We
analyzed and included historical data to the greatest degree possible, as allowed by the
level of detail of the model. Under these conditions, with accurate initial values for all
key variables, the system quickly achieves equilibrium in all areas of interest. The
following key data values shown in Table 1 were used as the baseline view:
PARAMETER VALUE
Size of metropolitan area 6,000,000 subscriber lines, including:
Business, Residence, Redundancy, Overbuild
Size of Central Office equipment 6000 one thousand port cards
Average repair time = Switch: 0.5 hour per line unit
= Frame: 0.5 hour per line unit
= Transport: 4 hours per cable break
= Loop: 0.75 hour per residential repair
Average travel time between field = 2 hours: transport sites
sites = _.75 hour: loop (residential) sites
Average Front end close time 20 minutes for switch and frame components
15 minutes for transport components
9 minutes for loop components
Average Front end close rate 25% for switch and frame components
10 % for transport components
40% for loop components
Number of repair workers per damage | 1 per switch and loop
report 2 per frame
4 per transport report (cable break)
Table 1: Default Model Parameters
Baseline Scenario Results
When the model was run using the above parameters, which were chosen to reflect the
information available from actual telecom service providers, the resulting failure rates
corresponded closely with actual failure rates observed in the network. As shown in
Figure 9, the network is approximately 0.25% damaged at any given time in the course of
normal functioning. This means the total number of network elements in the model that
are out at any one time will be close to 15,000; actual telecom operations information
shows that this is reasonable for a city with six million lines. Given the lag time between
damage occurring in the network and the actual reporting of damage so that it can be
repaired, the perceived damage in the network is slightly smaller than the actual damage
levels. Figure 10 shows that the repair rate is steady and that loop repair makes up the
largest portion of the repair rate, and of the damage.
0.4
0.3
0.1
0 4 #8 12 #16 20 24 28 32 36 40 44 48 52
Actual Damage 3-3-4444 4.—~ 3 Percent
Perceived Damage Percent
Figure 9: Results of Baseline Scenario Model - Actual and Perceived Damage
80 7 F 7 ¥ ¥ ¥ ¥ ¥ ¥ 7
60
40
20
0 se - —————— se =
0 4 8 12 16 20 24 28 32 36 40 44 48 52
weeks
Component Repair Rate{Field Tech, Switch] : baseline cy ai a Ed cs NetworkUnit/Hour
Component Repair Rate[Field T ech, Frame] : baseline 2 2 2 2 2 NetworkUnit/Hour
Component Repair RateField Tech, Transport] : baseline 3 3 3 3 3— NetworkUnit/Hour
Component Repair RatelField Tech,Loop] : baseline ——4 4. 4 4 4 4 NetworkUnit/Hour
Figure 10: Results of Baseline Scenario Model - Component Rate of Repair
200
He
he
He
he
He
he
He
he
be
nn
He
he
be
be
He
150
100
0 4 #8 12 #16 #20 24 #28 32 36 40 44 48 52
weeks
Worker Types Dispatched[Field Tech] : baseline —t +
Worker Types Dispatched[Ops Ctr] : baseline z 2 2 Person
Figure 11: Results of Baseline Scenario Model - Workers Dispatched
nn
a
a
A
A
cao)
2
a
°
5
The results from the baseline simulation correspond to telephony operations in the real
world: The rate of repair reaches equilibrium very quickly and remains stable, with a
small amount of ongoing network damage, as time is taken to notice, react to, and effect
repairs to each damaged component.
In Figure 11, it can be clearly seen that the number of field workers well exceeds the
number of NOC workers required, as it does in reality. The number of workers
dispatched to perform repairs reaches equilibrium quickly, with gradual movement from
initial zero, rather than sudden and drastic movement of the workforce.
Absenteeism Modeling Approach
Two general structures for modeling absenteeism are included in the model. These
structures move workers into and out of a stock of unavailable workers. Flow rates
between the stock and the pool of available workers can be stipulated, or can be derived
through goal-seeking on an exogenous fraction of affected workers. Although, in reality,
workers can leave work at any time, in the model, workers become absent only from the
pool of off-shift workers and return only to that pool.
The variable “epi switch” controls whether or not the model generates its own worker
loss numbers or if the numbers are externally provided.
Extraordinary
worker loss time Duration of oe
Extraordinary Worker Rarnadliay
Start Time of | Worker Loss Value ‘<Initial Total
Extraordinary Worker Workers Available> <N.EnEI Fraction
Event a ¥ <Time> Ne Incapacitated Including
extemal event gi ee
Joss rate . worker
Joss time eplswitch a eray
4 | desired worker
max worker loss be
ea
extraordinary 4
mip Srorkerloss
rate
: | Workers Affected by
Workers Off Shift ‘tian Event
a
Extraordinary Worker
Event Recovery Rate aay aie
epi switch> __— recovery rate
lest
workerloss> Extraordinary Worker
Event Recovery Time
Figure 12: Overview of Worker Illness Models
The model shown in Figure 13 represents the effect of absenteeism on the work force
when the variable “epi switch” is set to “0” (off). The variables governing the duration
and rate of absenteeism are described in Table 2. In this configuration the model does
not account for deaths due to illness, and all workers are at some point retumed to the
workforce.
Duration of
f Extraordinary Worker Extraordinary
Extraordinary
worker loss time Event Worker Loss Value
Bie oder extemal event 4 ____<Time>
Event Joss rate epi switch
\ extraordinary
worker loss
rate
al
Workers Off Shift ve: pect by
~<a y
Extraordinary Worker
Event Recovery Rate ge
_— iN
<epi switch>
Extraordinary Worker
Event Recovery Time
Figure 13: Internally Driven Absenteeism Model
PARAMETER DESCRIPTION
Start Time of Extraordinary Worker
The two parameters control the timeframe
Event :
: : of the absenteeism: at what hour of
— of Extraordinary Worker simulation it starts and how long it lasts.
Extraordinary Worker Loss Time The two parameters work in conjunction to
drive the severity of the absenteeism: the
Extraordinary Worker Loss Value workers become absent at the following
rate: Loss Value/Loss Time
How many days before an absent worker
can return to work
Extraordinary Worker Recovery Time
Table 2: Absenteeism Model Parameters
The model variables shown in Figure 14 allow for absenteeism to be input into the model
from external sources when the variable “epi switch” is set to “1” (on). This allows a set
number of workers to be removed from the active workforce at any given time. It can
account for deaths due to illness by never returning workers to the work force (e.g. by
having ten workers in the “workers affected by extraordinary event” stock at all time
steps,).
<N.EnEL Fraction
<lnitial Total Incapacitated Including
oplsiich Workers Available> Dead>
worker
| time desired worker-¢———_ discrepancy
max worker loss —— a
Tate extraordinary
Da i
rate
>|
Workers Off Shift Wont sey
I. reinary
Extraordinary Worker
Event Recovery Rate
es
Spies max worker
assis recovery rate
worker loss>
Extraordinary Worker
Event Recovery Time
Figure 14: Externally Driven Absenteeism Model
Absenteeism Simulation Results
This section documents the baseline absence scenarios results. We find that even a small
perturbation results in noticeable impact on overall levels of network failure. The two
scenarios run were:
e Scenario A: Using internally generated absenteeism - 15% off shift-worker loss
rate per day, with a 30 day absence period starting at day 10.
e Scenario B: Using externally generated goal-seeking absenteeism.
Scenario A.
The worker loss rate in Scenario A is shown in Figure 15. This worker absenteeism
removes 15% of off-shift workers per day for 30 days, placing them in the “workers
affected by extraordinary event” pool for starting at day 30. Workers have on average a
10 day time period to be retumed to the workforce. The resulting worker dispatch profile
is shown in Figure 16.
2
15
0.5
0 4 8 12 16 20 24 28 32 36 40 44 #48 52
weeks
extemal event loss rate[Field Tech] : absence ——+——?——2——+——_ Person/Hour
extemal event loss rate[O ps Ctr] : absence —2—-2——2—2—2—__ Person/Hour
Figure 15: Absenteeism By Worker Type
200
150 $9 >» 4} —_}- + 7-4)
100
50
0 — 2. 2. >. >. 2.
0 4 #8 12 16 2 24 28 32 36 40 44 48 52
weeks
Worker Types Dispatched[Field Tech] : absence —+ + + + Person
Worker Types Dispatched[Ops Ctr] : absence 2 Person
he
nn
he
4
4
Figure 16: Dispatched Workers
As shown in Figure 17, as workers are removed from the pool of available workforce, the
percent of the network that is damaged goes up. There is a slight increase to the percent
network damaged after all of the workers are returned to the available pool due to an
increase in perceived damage. When there are fewer workers in the workforce, damage
is not noticed as rapidly. Once workers are retumed to the workforce, unreported damage
is again noticed at a normal rate and the amount of perceived damage and thus the
percent of the network that is damaged goes up.
0.4
0.3
0.2
0.1
Actual Damage +——t—-2—_4+—_4-—__4-—_ 4. 4+ +> +>-—~_ +4 Percent
Perceived Damage 2—2 -— 3 Percent
Figure 17: Perceived and Actual Damage
Figure 18 shows the necessary overtime for the workforce during the absenteeism period.
When workers are initially removed from the workforce, the overtime required from the
remaining workers goes up. It continues to rise as more workers are removed from the
pool of available workers. Once workers are retumed to the pool, necessary overtime
begins to fall as the backlog of damage is worked off. Some amount of the backlog
remains for the duration of the run due to repairs having a repair time tolerance, such that
they don’t need to be completed immediately.
16.5
0 4 8 12 16 20 24 28 32 36 40 44 48 52
weeks
calculated shift length[Field Tech] : absence +—-»—2—_2—_3—2—3- Hour
calculated shift length[O ps Ctr] : absence 2—2_3-_3 2 22 Hour
Figure 18: Necessary Shift Length by Worker Type
Scenario B
The worker loss rate in Scenario B is shown in Figure 19. The model goal seeks the
number of workers absent to match the desired fraction of total workers absent resulting
in the number of absent workers shown in the figure. The fraction of workers absent
never returns to zero, simulating workers who never retum to the workforce. The
associated model results are shown in Figure 20.
Similar to scenario A, the percent network damaged peaks once the workforce has
retumed and the backlog of unnoticed damage is perceived. As shown in Figure 21, once
the absenteeism begins, workers are required to work overtime. The model does not
allow the workforce to go beyond a double shift, thus the necessary shift length never
goes beyond sixteen hours. Overtime continues through the end of the model run as the
backlog of damage slowly gets resolved.
400
300
200
0 4 #8 12 #16 20 24 28 32 36 40 44 48 52
weeks
Workers Affected by Extraordinary Event{Field Tech] : epidemic. —3——3——33_+
Workers A fected by Extraordinary Event[Ops Ctr] ; epidemic ~ = 2— Person
A
g
g
Figure 19: Absenteeism Scenario B
15
0.5
0 4 #8 12 #16 20 24 28 32 36 40 44 48 52
Actual Damage --——2——-—-—_4.-__49—__4-—__4-—__4.—__ 4.4. Percent
Perceived Damage 2 2— Percent
Figure 20: Absenteeism Scenario B results - Actual and Perceived Damage
20
16.5 a a ce
13
9.5 a
6
0 4 8 12 16 20 24 28 32 36 40 44 48 52
weeks
calculated shift length{Field Tech] : epidemic +—_1—__+—_+—_+—+ Hour
calculated shift length[O ps C tr] : epidemic 2 3 2 Hour
Figure 21: Necessary Shift Length by Worker Type
Conclusions and Future Work
The process of modeling telephony repair activities as well as analysis of the results
shown here and additional scenarios not presented here, lead to several conclusions:
The complexities of telephony repair are quite amenable to the modeling process.
Simulation of historical events with known parameters showed close correlation
between the resulting model repair times and the historically recorded repair times.
The same is true for worker effort and staffing level values.
The complexities of telephony repair are well modeled by SD flows, because
although staff, inventory and failures occur in discrete units, the repair activities occur
continuously over time and in fact, can be considered at any time to be percentage
complete. The need to allocate staff to repair events lead quite cleanly to an
interpretation of repair events as a stock of repairs-needed, calculated in terms of
time, i.e. "staff-hours".
In terms of insight derived from the results of running the model, it has been shown
that repair and recovery functions in the voice telecommunications infrastructure are
resilient to the loss of a significant portion of the work force. However, certain
activities, such as sharing of warehouse inventory across regions and emergency staff
augmentation during a crisis, is an increasingly effective response as the severity of
the emergency increases.
Future work for the model includes testing of workforce management strategies in a crisis
situation such as long-term worker loss or large scale outages.
References
A. Jrad, H. Uzunalioglu, D. Houck, G. O'Reilly, S. Conrad, W. Beyeler. “Wireless and
Wireline Network Interactions in Disaster Scenarios,” MILCOM 2005.
Gerard O’ Reilly, Ahmad Jrad, Theresa Brown, Stephen Conrad, “Critical Infrastructure
Analysis of Telecom for Natural Disasters,” Networks 2006, November 2006.
Hines, Jim “Molecules of Structure: Building Blocks for System Dynamics Models,”
from www.vensim.com, July 2007.
Sterman, John. Business Dynamics. Boston: McGraw-Hill, 2000.
