`6/28/1999
`
`Robert Hanmer
`Lucent Technologies
`hanmer@lucent.com
`2000 North Naperville Road
`PO Box 3033
`Naperville, IL 60566-7033
`voice: 630 979-4786
`
`Michael Wu
`Lucent Technologies
`mwu@lucent.com
`2000 North Naperville Road
`PO Box 3033
`Naperville, IL 60566-7033
`voice: 630 979-5615
`
`Copyright (c) 1999, Lucent Technologies. Permission is granted to
`copy for the PLoP 1999 conference. All other rights reserved.
`
`Abstract
`An examination of the successful techniques that have allowed earlier networks and network elements to be
`successful must be done to understand how to evolve networks. The patterns presented here describe
`techniques implemented in many Lucent Technologies switching systems to deal with traffic congestion
`issues. Handling congestion is one of the key roles of a network management system.
`
`Congestion is the term used by telecommunications system designers to describe the situation when the call
`load exceeds the available resources of the system.
`
`These patterns discuss ways that these systems deal with congestion. The approaches can be generally
`classified as protective or expansive controls. Protective controls reduce the amount of work by restricting
`access to resources. This protects the system from too much work. Expansive controls allow the system to
`use additional resources not available to call processing under normal conditions. This expands the
`possible actions that the system may take.
`
`Introduction
`“Congestion occurs when the call load exceeds the capacity of available routes or trunks, or of
`common control equipment (for example, call registers that temporarily store call data while the
`call is being processed.) Congestion can also occur when the call load cannot be handled in
`available real time.” [MG]
`
`Telecommunications system designers have studied the problem of system resource congestion for years.
`These same problems are evident in the distributed and client server computing systems of today. This
`article describes solutions that have worked in telecommunication systems over time.
`
`Network management functions within telephone switching systems are designed to maintain a high level
`of service during periods of unusual traffic on systems that are engineered for less than full capacity.
`During these periods, machine performance naturally degrades due to resource contention and network
`delays. Even when normal traffic levels are restored the system continues to struggle to achieve normal
`performance levels. [GHHJ]
`
`Internal congestion arises from two different categories: peripheral capacity and processor capacity.
`Common control or peripheral equipment capacity is related to how a system is engineered and how
`requests for service arrive. In the telecommunications world these requests for service are “traffic” and
`consist of telephone calls being handed off from one system to another as they progress throughout a
`network. Processor capacity (i.e. processor real time) is taken as the key driver behind a system’s
`
`RSH/MW
`
`1
`
`07/17/99 BLTJ
`
`hanmer629.doc
`
`000001
`
`YMAX EXHIBIT 1028
`YMAX CORP. V. FOCAL IP
`IPR2016-01258
`
`
`
`“overload” capabilities. Peripheral capacity issues will be discussed here. For solutions to these problems
`refer to “A Pattern Language for Improving the Capacity of Reactive Systems” by Gerard Meszaros
`[Mesz].
`
`We begin by discussing “patterns”, the form in which the solutions we present are described. Then the
`context of telecommunications traffic and how it appears to switching systems is discussed. This is
`followed by the actual traffic congestion patterns.
`
`Background
`
`Patterns
`This section lays the foundation that is required to understand the patterns presented in the remainder of
`this paper. A pattern is a description of a good, working solution to a problem that occurs again and again.
`A pattern contains a statement of the problem, the context where the problem exists, and a consideration of
`the various trade-offs involved in the solution, as well as the actual solution. Patterns only capture
`solutions that have stood the test of time and can be used whenever the problem occurs again. Generally
`patterns must be customized to a system’s unique characteristics to be used.
`
`What is contained in this article are solutions to problems related to peripheral control equipment
`congestion. These patterns describe functionality needed by a switching system to concentrate upon
`profitable work by avoiding unnecessary work and to decide in real time how calls are routed in a network.
`They each solve smaller parts of the larger problem of dealing with congestion, however when taken
`together they form the nucleus of an effective network management system.
`
`The names of patterns appear as underlined italics.
`
`Telephone Switching Systems and Traffic Congestion
`A telephone switching system is a system that is intended to connect customer telephone circuits when the
`customers desire to communicate. These connections are created as a call is being established and are
`released when the customers hang up. Route selection and the selection of idle bandwidth are routinely
`done by these systems without human intervention.
`
`Telephone switching systems are real-time, fault tolerant systems that attempt to maintain themselves
`without requiring constant human supervision (refer to Minimize Human Intervention [ACGH+]). But the
`systems generally consider that sometimes People do know best [ACGH+], so provisions for human
`oversight and altering normal actions are built into the system.
`The patterns are primarily drawn from experiences with the 4ESS(cid:212)
` Switch, a high-capacity toll and
`tandem switch (refer to the Hierarchical Routing pattern below for an explanation of these terms).
`Traditionally the connections between switching offices are referred to as “trunks”. The term “line” refers
`to the connections between switching offices and individual telephone sets. This pattern language only
`contains patterns related to congestion on trunks. Generally at the trunk level there are not restrictions that a
`particular call use a particular route, unlike when a telephone being called resides at the end of only one
`specific line.
`
`Typically a telephone network requires a large number of switching systems, since individual systems have
`not been built that have the capacity required to concentrate all of the telephone traffic. Each of these
`systems is a node in the telephone network. Between these switching systems there are a large number of
`possible routes.
`
`A large network in which every node is connected to every other node would be infeasible. Every route
`requires ongoing maintenance of the physical medium as well as maintenance within each switching
`system node that it connects to. But with more routes in the network, any specific telephone call has fewer
`switches to traverse; resulting in higher quality and less time required to complete the call.
`
`A balance between number of routes with their associated expenses and the quality of service that is
`achievable and desirable in the network must be struck. The impact on traffic congestion should be
`RSH/MW
`2
`07/17/99 BLTJ
`
`hanmer629.doc
`
`000002
`
`
`
`obvious. If every switch is connected to every other with trunks in sufficient quantity then congestion will
`be non-existent. With fewer links, the possibility of routes concentrating on individual switches increases
`as well as the probability that that a switch will be subject to control equipment congestion.
`
`Typically these networks have evolved with time. New switching systems replace older, smaller ones.
`Switching systems are quite expensive, so they are used as long as possible before this replacement occurs.
`Over time networks become very heterogeneous. This complicates any strategy for dealing with
`congestion since the switches may not all have the same abilities to control congestion.
`
`The evolution of the network also requires that the systems be designed to be flexible in their routing. The
`various trunks that are installed in a new switching office might not be the same trunks as when the office
`is retired or replaced at the end of its useful life.
`
`When a telephone network is being engineered usual telephone usage patterns are considered. Since
`switching equipment is expensive the systems are engineered to take advantage of the fact that each
`individual telephone line has much idle time. Most customers do not make continuous telephone calls.
`The systems are designed to allow expensive peripheral equipment to be shared among a number of lines or
`trunks. This common control equipment can be allocated as needed, and since most lines or trunks are not
`in use continually this sharing is feasible without diminishing the quality of service. But, shared resources
`can be needed by too many calls at the same time, resulting in peripheral capacity limitations.
`
`Natural disasters are another complicating factor within the network of switching systems. A common
`reaction to a disaster is to call your relatives to determine their status. When everyone does this a focused
`traffic load aimed at switches in the disaster area occurs. This is a kind of peak traffic that the network
`engineers cannot foresee, and hence frequently causes overload and control equipment congestion.
`
`Many of these concepts: a variety of nodes, infeasibility of directly connecting every switch, and shared
`resources can be found within the surface transportation network as well as the telephone network. Each of
`the patterns in this language presents one or more examples from this other network to illustrate the
`principles discussed by the pattern. These examples also help show the power of these patterns in that they
`are not limited solely to an electronic communications network.
`
`Language Map
`The following diagram shows the relationships between the patterns in this
`language. The diagram shows patterns that enhance the solutions of other patterns,
`resolve previously unresolved forces in a pattern, or take advantage of an earlier
`pattern to provide some new system capability. In the example to the right, pattern
`B refines pattern A, helping to solve unresolved forces or new problems that A
`introduced.
`
` A
`
`B
`
`RSH/MW
`
`3
`
`07/17/99 BLTJ
`
`hanmer629.doc
`
`000003
`
`
`
`Hierarchical
`Routing
`
`TSG
`
`Route To
`A TSG
`
`Automatic
`Controls
`
`Protective
`Controls
`
`Expansive
`Controls
`
`SDOC
`
`Selective
`Trunk
`Reservation
`
`Automatic
`Out-Of-Chain-
`Routing
`
`Final
`Handling
`
` RSH 6/28/99
`
`The Patterns:
`
`Trunk SubGroup (TSG)
`Intent: Group trunks by common properties so that appropriate ones are easy to find.
`
`Problem: From the myriad of different trunks in an office, how do I select the one for this particular call?
`
`Context: Because of the heterogeneous nature of the switching systems in the telephone network, each is
`capable of performing call setup signaling with only a subset of the available protocols. Each of these
`protocols has different advantages and disadvantages for each switch that uses it. Each trunk has attributes
`related to the protocol used upon it. In addition to this each trunk will have a variety of fixed attributes
`such as endpoints and targeted traffic.
`
`Forces: A switching network element will have many, many trunks, sometimes up to 100’s of thousands.
`Each of these has several different attributes that can be used to describe them. These attributes are of
`interest to the routing process so that appropriate classes of service can be allocated to each call traversing
`the network. Some of the example attributes are:
`
`Destination, Translation domain, switching domain, traffic use, far end, INWATS, incoming
`screening class, incoming traffic separation class, directionality, transmission delay, signaling
`characteristics. [MG]
`
`RSH/MW
`
`4
`
`07/17/99 BLTJ
`
`hanmer629.doc
`
`000004
`
`
`
`The system cannot just pick any trunk from the pool of available trunks; it must have the right attributes.
`This requires that some record be kept of the attributes on a per trunk basis. These records can be used by
`the routing selection functions to locate the correct one to use.
`
`Solution: When initially engineering the office group trunks into “trunk subgroups” (TSG) based upon
`common properties. These properties will allow trunk selection (or in the case of Network Management
`trunk category selection) to be easier since it will be easier to identify similar trunks at route selection time.
`
`NN
`
`HH
`
`FF
`
`CC
`
`SS
`AA
`
`KK
`
`RR
`
`BB
`
`II
`MM
`
`PP
`
`EE
`
`GG
`
`JJ
`
`LL
`
`TT
`
`Rationale: This helps two aspects of switch application software: Network Management and Call
`Processing. Network Management is helped because finer divisions allow finer degrees of control. Call
`processing is allowed to select services by allowing calls to be routed to a particular type of facility.
`
`Resulting Context: Trunk resources with specific attributes can be identified quickly because they are
`within the same TSG. This simplifies selection and allows the system to Route to a TSG at execution time.
`
`Highway Example: Associate properties with each highway, such as maximum speed, how direct they are,
`tolls, etc. These can be used during route planning to provide the desired type of travel experience. Many
`modern trip assistant software packages actually allow the user to select which categories of roads that they
`will travel on. This allows the user to find routes with the properties that they desire.
`
`Hierarchical Routing
`Intent: Define a fixed way to route calls to neighboring switches
`
`Problem: How do you design your network so that even unsophisticated switching systems can route calls
`through it automatically.
`
`Context: You can't connect every office to every other. The method has to be easy enough for pre-
`electronic switching equipment to implement in hardware or with simple wiring changes. You want to
`design the network to make optimal use of its links and of its common resources.
`
`Forces: If you directly connect every node in a network you have n(n-1)/2 links. From experience we
`know that there is much expense associated with managing large numbers of links.
`
`The more systems that must process a call, the longer it takes for the call to be established.
`
`Solution: Identify each switch by a class number or level. In the telephone network five levels are used.
`Class 5 offices are end offices to connect to customer telephone sets. Class 4 offices are called toll offices,
`class 3's are primary centers, class 2's are sectional offices and class 1's are regional centers. Sometimes
`local tandem offices connect class 5 offices without the call requiring a class 4 office.
`
`Route each call as low as possible. If the calling and called party are on the same switch, route the call
`directly. If they are not, go up the hierarchy from the calling party's class 5 office to the lowest switch
`(highest class number) that can connect the calling party's class 5 office with the called party's class 5. For
`example the call may be passed through the local office to the toll office to another toll office and back
`down to the called parties local, class 5 office.
`
` Engineer some links between offices of different classes to provide some shortcuts within the hierarchy.
`
`RSH/MW
`
`5
`
`07/17/99 BLTJ
`
`hanmer629.doc
`
`000005
`
`
`
`The assignment of a class number to a switch is something that is done by the network engineers as they
`are creating the network or allocating work to switching systems. The routes between switches that the
`network engineers design are accessed automatically during call creation to advance the call.
`
`Rationale: A regularity has been added to the telephone network so that even systems that do not have
`computing power as we know it can route calls. The hierarchy provides that the higher the switch is in the
`hierarchy (the lower the class number) the fewer there need to be in a network.
`
`Resulting Context: A large number of ramifications upon system design are introduced by using a
`hierarchical routing method, such as varying feature sets. These are not generally germane to this
`discussion of traffic congestion, except in the differences between lines and trunks that was made earlier.
`Lines connect telephone sets with Class 5 offices. Trunks connect switching systems. These patterns are
`primarily concerned with trunk traffic congestion.
`
`Highway Example: The system in the USA of township roads, county highways, state highways, US
`highways and interstate highways. At each level there are fewer roads than at the next level, and each is
`designed to handle more traffic than the previous level.
`
`Route To A TSG
`Context: Trunks with similar properties have been grouped into Trunk Sub Groups during office
`engineering. Within a toll or tandem switch it is rarely required that a specific trunk be used for a particular
`telephone call. Generally the call must be forwarded from one switch to another by any available route.
`There are usually restrictions upon the types of trunks that are to be used when routing a call. For example
`the number of satellite links might be limited (due to the time delay from the great distances involved).
`
`Problem: How granular should the “routing” function of a switch be?
`
`Forces: Assume there are M routes between two switches and each route has N trunks. To identify a route
`all the way down to an idle trunk will take O(M*N) time.
`
`To identify the appropriate route from among the M potential routes only takes M time. Then to identify
`the idle trunk within that route takes at most N time. Thus to break the route identification into two parts,
`one at the route level and the other at the trunk level takes only O(M+N) time.
`
`RSH/MW
`
`6
`
`07/17/99 BLTJ
`
`hanmer629.doc
`
`000006
`
`
`
`Different algorithms might be more efficient for identifying routes and trunks. The deeper into the routing
`that you have to go to find the answer the more time and memory that you need. Routing to the trunk level
`requires the most levels. Routing to the trunk group requires fewer levels.
`
`There are benefits to using the same algorithm to find both the route and the idle trunk, mostly in the
`maintenance area. Switching system performance is frequently more valued than its ease of maintenance.
`
`There can be efficiencies if the problem being solved can be made smaller and optimal algorithms used for
`each sub-problem.
`
`Solution: Routing a call across a network should occur in two steps. First, identify a route between the
`endpoints. This can be determined by examining the Trunk Subgroups. From there the selection of an
`individual trunk can occur without the overhead of the route selection process.
`
`Selection of an idle trunk is a different problem than selection of a route between switches. Any idle trunk
`within a group will suffice, once the route has been selected.
`
`Route only down to the trunk group.
`
`Then use the most efficient method to get to an individual trunk.
`
`Rationale: Using the most efficient method to select the route and using a different efficient algorithm to
`select the trunk within the route optimizes memory and execution efficiency of the trunk selection.
`
`This pattern should be applicable to any context in which you have several "pipes" going from A to B with
`possibly different characteristics.
`
`Resulting Context: A system where the most efficient algorithm is used for each part of the routing/trunk
`selection problem, even though that means that multiple algorithms are used. Pick the pipe that goes to the
`right place and then identify which flow within the pipe.
`
`Let trunk group selection identify the major freeways to be taken, but leave the individual lane selection for
`more efficient (time and memory) methods.
`
`Highway Example: Someone offering guidance about a route to take between cities doesn’t also specify
`which lane to be in for the entire trip. Suggestions related to particular highway features might be offered,
`these are like the guidance offered by the different properties that were used to create the various TSGs.
`
`Automatic Controls
`Intent: When conditions dictate, the switch should automatically institute changes to normal behaviour to
`handle extreme conditions.
`
`Problem: People are slow, congestion problems grow rapidly
`
`RSH/MW
`
`7
`
`07/17/99 BLTJ
`
`hanmer629.doc
`
`000007
`
`
`
`Context: Minimize Human Intervention [ACGH+] is in use, saying that the system should be able to take
`care of itself. The system is capable of computation. This eliminates many electro-mechanical switching
`systems from using this pattern.
`
`Forces: People are slow and reluctant to respond. They look away, they miss things and frequently they
`overreact. Computers are good at deciding things if they’ve been prepared in advance for a given scenario.
`Most congestion situations are similar. They might be from a different source, but are basically more work
`than can be done in a given time.
`
`Solution: The system should apply controls automatically. And be able to remove controls automatically
`as well. The system should allow humans to intervene (People Know Best [ACGH+]), but it should not
`rely upon them to intervene.
`
`Rationale: The human element is eliminated, in keeping with Minimize Human Intervention.
`
`Resulting Context: Having stated that the system should apply controls automatically, the actual controls
`need to be defined. Automatically applying controls is effective in both the processor and peripheral
`congestion areas. Some of the controls that can be applied are more applicable to one area than another
`(e.g. Expansive Automatic Controls ), but overlap does exist as you will see.
`
`Highway Example: There are a number of examples of automatic controls within the USA highway
`network. Frequently the highways have embedded sensors that will identify potential congestion and
`display it automatically to alert motorists that congestion appears ahead. Sometimes an automated radio
`broadcasts on a certain frequency to alert motorists. Sometimes the highways will automatically
`reconfigure themselves through reversible lanes that change to accommodate rush hour traffic flows.
`
`Protective Automatic Controls
`Intent: Protect the system from too much work by restricting what work it will allow into the system.
`
`Context: The system is capable of applying Automatic Controls. The metric being watched is throughput –
`the amount of traffic (calls) that will be successfully processed during a period of time, and congestion is
`occurring for the finite resources within the system. For example, during an event where many customers
`are dialing in simultaneously and there is a shortage of the peripheral equipment that processes touch-tones.
`
`Problem: What actions should the switch take when the resource being congested has a finite limit?
`
`Forces: If the system takes on all the work that is being offered for the finite resource, requests will have
`to be queued for service when resources are freed. This takes work to queue. And if the congestion is of
`more than brief duration queue lengths will increase and yield unacceptable delays. The nature of the
`traffic then also comes into play. Human callers get impatient, hang up and retry their call. This results in
`requests being queued that are no longer valid as well as additional requests for service.
`
`Solution: Impose Automatic Controls that restrict how much work the system accepts. Sometimes there
`are things that the system can do that will safely protect itself from too much traffic. These should be
`invoked when the congestion is in elements that have finite limits and no idle capacity that can be used.
`
`RSH/MW
`
`8
`
`07/17/99 BLTJ
`
`hanmer629.doc
`
`000008
`
`
`
`RSH/MW
`
`9
`
`07/17/99 BLTJ
`
`hanmer629.doc
`
`
`
`Problem: How can certain kinds of traffic be allowed? Is there some work that the system is capable of
`handling even though it cannot do some work of different kinds? Can the system process any traffic? Are
`there only certain types that are congested at the moment?
`
`Forces: Frequently congestion is localized. A certain originating or terminating node is congested (as
`during a natural disaster). Or a certain type of resource is much overloaded, possibly because of a failure
`somewhere else in the network. Even in periods of processor overload, or traffic congestion, there will be
`some idle resources that can be used to allow traffic through.
`
`Solution: Request of systems that traffic is routed to that they limit the amount of traffic that they send,
`but only for the particular properties (as identified by Trunk SubGroup) that are in congestion.
`
`
`
`STOPSTOP
`
` GO GO
`
`Highway Example: High occupancy vehicle lanes frequently allow their types of traffic avoid congestion
`on the main thoroughfare through the use of dedicated exit ramps. Since they are dedicated they are much
`less likely to be congested than the main roadway.
`
`Selective Trunk Reservation
`Intent: Deny incoming traffic on TSGs that have few idle trunks during periods of congestion.
`
`Context: The system is designed to automatically apply Protective Automatic Controls. Actions will be
`taken that will allow it to protect itself, to maximize the throughput through this individual system. The
`system has too much work to do and not enough of some peripheral equipment resources.
`
`Problem: How should the system handle requests to the most busy Trunk SubGroups?
`
`Forces: Calls that must be retried several times to be completed tie up resources longer than calls that
`complete on the first attempt.
`
`Solution: The traffic most likely to succeed on TSGs that have many idle trunks. To protect the system
`during congestion, favor those calls over the ones that arrive on busy TSGs. Mark certain outgoing trunks
`in such a way that calls to them are allowed to be processed. Calls to the unmarked trunks will be blocked.
`
`RSH/MW
`
`10
`
`07/17/99 BLTJ
`
`hanmer629.doc
`
`000010
`
`
`
`
`
`STOPSTOP
`
` GO GO
`
`Resulting Context: A system with Selective Trunk Reservation will spend more time on work that will
`actually succeed rather than work that will only delay other work.
`
`Highway Example: On roadways that have less traffic, provide longer green traffic signals than on the
`congested roadways. This favors keeping the less busy roadway flowing smoothly. The other roadway is
`already congested – it won’t make much difference. This is very similar to “Late trains get later”.
`
`Expansive Automatic Controls
`Intent: Protect the system from too much work/traffic by providing new ways to do the work.
`
`Context: Automatic Controls can be applied. In this case the metric being watched is the number of calls
`that reach their destination. This is a more network-centric metric than that being watched in Protective
`Automatic Controls (which is system throughput). Resources within the network are not totally finite.
`New links can be created to expand it’s the network capacity.
`
`Problem: How can we both protect a system from the effects of overload and at the same time not reduce
`overall throughput?
`
`Forces: If purely Protective Automatic Controls are used when confronted with congestion, each switch
`will shrink back upon itself and network throughput will decrease, which is opposite the objectives of this
`pattern. . If some way to expand the range of possibilities available to a system were possible then overall
`network throughput might not decrease. Are there alternate ways that a system can deal with its workload?
`Ways that are normally not used? Can any of these ways be put in? If used for ordinary traffic they don’t
`expand it during time of network congestion.
`
`Solution: Design relief mechanisms into the network that can be accessed only in case of congestion.
`Provide new ways for the switch to do its work that a) has a higher probability of succeeding or b) use
`fewer switch resources.
`
`Impose automatic controls that provide new ways to do the same work. This allows excess work to be
`taken on and the congestion lessened. It is like providing an escape valve.
`
`RSH/MW
`
`11
`
`07/17/99 BLTJ
`
`hanmer629.doc
`
`000011
`
`
`
`Rationale: There are three basic responses that a system can have with confronted with congestion. The
`first is to shrink within itself, taking whatever action is necessary to protect itself from the onslaught of too
`much traffic. The second is to open itself up to the traffic, taking on as much as it can and shedding work
`that is unnecessary. The third is to not do anything, which will generally lead the system into instability.
`
`Expansive Automatic Controls result in the network throughput being managed
`
`Resulting Context: Expansive Automatic Controls provide new ways for the system (element in a
`network) to do its work. This requires work at the network level rather than the level of individual systems.
`
`In some scheduling regimes it is possible to defer some internal maintenance work during peak traffic
`times. This expands the amount of real time that is available to perform revenue-generating work. It also
`allows this pattern to be applied to processor capacity.
`
`Related Patterns: Protective Automatic Controls deals with finitely bounded resources. This pattern deals
`with resources that can be expanded.
`
`Highway Example: An example of an expansive control is when the police direct traffic over side streets,
`possibly in violation of posted regulations, to avoid an accident scene. They have the authority to take this
`action, but they seldom exercise it. Another example is using all lanes of a highway in one direction to
`allow a higher number of people to leave an area threatened by a natural disaster such as a hurricane.
`
`Automatic Out-of-chain Routing
`Intent: During overloaded periods, allow new routes within a hierarchical network.
`
`Context: The system has the capability to apply Automatic Expansive Control. The network employs
`Hierarchical Routing. The network is being engineered to maximize throughput.
`
`Problem: How can the effects of congestion on trunks to the next office in the Hierarchical Route be
`reduced?
`
`Forces: The routing chains of a hierarchical network are quite rigid. And the network designers can look
`at it in advance during the engineering phase and calculate the total possible throughput. These limits will
`be exceeded sometime.
`
`Traffic can be restricted through a Protective Automatic Control, but that will decrease throughput.
`
`A hierarchical network does not connect every system to every other. There are many possible routings
`between switches that are not part of the standard hierarchy of the network. New routes between systems
`might be nothing more than a jumper cable in another office (i.e. an alternate route from A to C might
`really be an A to B route permanently connected to a B to C route), they might not need to be a direct
`cross-country route.
`
`Solution: During periods of traffic congestion, allow the system to use routes that are not normally used.
`They are “out-of-chain”. Since these routes are not usually used they will not be congested.
`
`RSH/MW
`
`12
`
`07/17/99 BLTJ
`
`hanmer629.doc
`
`000012
`
`
`
`RSH/MW
`
`13
`
`07/17/99 BLTJ
`
`hanmer629.doc
`
`
`
`Resulting Context: Data should be collected when calls are terminated abnormally. This data needs to be
`presented to human network managers so that they can make changes to strategy or network topology.
`
`Highway Example: On some freeways there are emergency pull-off locations where it is safe to stop and
`report an accident or vehicle problem without obstructing the normal flow of traffic. After the problem is
`reported or corrected the normal lanes are used to resume the trip up to a normal exit. These lanes do not
`provide special exit ramps. They provide the ability to perform some special actions and then the normal
`exit ramp is used.
`
`RSH/MW
`
`14
`
`07/17/99 BLTJ
`
`hanmer629.doc
`
`000014
`
`
`
`Acknowledgments:
`The authors acknowledge the many very useful comments of the PLoP shepherd David DeLano.
`
`Dana Garoutte and Andrew Peck of Lucent Technologies originally wrote Route to A TSG. It has been
`revised here to fit within the context of this language.
`
`Appendix: Related Patterns
`Pattern Name
`
`Pattern Intent
`
`Dynamic Overload Control (DOC)
`
`Tell connected systems we’re in trouble!
`
`Minimize Human Intervention
`
`Give machine enough smarts to not require human intervention
`
`People Know Best
`
`Assume humans know more than the machine.
`
`Glossary:
`Call Processing: The switching system application that connects telephone calls from Calling party to
`Called Party.
`
`Congestion: When the system receives more requests for service than it has peripheral resources to handle
`them.
`
`Control (noun): Changes to normal call processing instituted during congestion or overload to allow the
`system to either handle additional traffic, or to protect itself from insanity.
`
`DO