Note: Descriptions are shown in the official language in which they were submitted.
CA 02205421 1997-OS-15
WO 96!20549 PCTlGB95103025
1
FAULT MONITORING
This invention relates to the monitoring of a telecommunications system
for faults. Telecommunications networks have a large number of components
distributed over a wide area, and it is important to be able to identify
faults when
they occur and to deal with them promptly before they manifest themselves to
the
users of the network as a service interruption. The user of the network, in
turn,
may have a service level agreement with the network operator which specifies
contractual penalties if service interruptions exceed a predetermined limit.
Many
methods are known for identifying faults and alerting the operators to them so
that
remedial action can be taken. Included in the term 'fault' for the purposes of
this
specification are not only equipment or service breakdowns but events such as
overloads which may be due to external causes but require remedial action.
Different users of a system need notification in different circumstances. A
telecommunications network operator, responsible for monitoring equipment,
needs
to know of failures of individual equipment. However, it is possible for an
item of
equipment in a telecommunications network to fail with no immediate effect on
the
level of service provided to a customer, if for example alternative routings
are
available, or if the customer is not using the full capacity of the system.
Conversely a customer may experience a service interruption when no individual
item of equipment has failed, for example if the total demand for use of the
system
by all customers exceeds the capacity of the system. In order to monitor the
service supplied to its customers, the network operator also needs to be
informed
of such service interruptions.
For example, in a telecommunications network, if a channel is running at
full capacity any further call attempts on this channel would fail. If the
rate of
occurrence of such call failures increases, this indicates that the network is
at full
capacity for an increasing proportion of Remedialactionmay
the time. be
' possible, for example by reallocating capacityelsewhere(e.g. second
from a
channel in a time division multiple access In the case service
system!. of
' interruptions to a customer, these can be logged. The nature of the service
level
agreement will determine what information is required, but in most cases a
guaranteed minimum level is agreed. The minimum level may be an agreed
CA 02205421 1999-11-29
W0.96/20~4'9 PCTIGB95/03025
2
proportion of time that the service is available. In other circumstances, for
example where re-start procedures are complex, the number of separate
interruptions may be used as an additional or alternative criterion.
There is a difficulty in fault monitoring, that phenomena which may be of
a transient nature may be indicative of the presence of an actual or imminent
underlying problem, but may instead be no more than statistical fluctuations
of no
significance. A transient fault which appears to clear itself may signify that
the
cause has ceased, or it may signify that there is a underlying problem and the
fault
will recur under certain conditions. A particular special case of this problem
occurs
when an alarm is triggered by a property of the system exceeding a
predetermined
threshold value. If the threshold value is set too high the alarm indicator
will not
be triggered until the situation is already critical, leaving no margin in
which to take
remedial action. However, if the threshold is set too low there will be an
excessive number of false alarm indications. The importance of transient
faults
may differ according to individual customers' requirements. To a customer
whose
terminal equipment includes error-correction facilities intermittent,
transient, faults
may be unimportant. However, to a customer for whom any interruption, however
short, necessitates a re-start operation involving complex security procedures
or a
site visit to re-set equipment, a large number of transient interruptions are
much
worse than a single interruption of longer length.
Transient faults have to be handled differently from non-transient faults.
Transient faults clear themselves, so there is no need to take any remedial
action
to clear them manually. However, because they clear themselves, it is
difficult to
investigate the cause of a transient fault, or to identify patterns which may
indicate an underlying problem.
A prior art fault-monitoring system is known from IBM Technical Disclosure
Bulletin, Vol. 35, No. 7 dated December 1992. Each second, this system records
whether or not a fault is present, by detecting transmission errors. An alert
is sent
if the number X of individual seconds in which a fault was detected in a
period of
1 5 minutes exceeds a first threshold L or if in a period of 24 hours the
number Z of
individual seconds in which a fault was detected exceeds a second threshold
which is, in proportion to the size of the measurement period, much lower than
L.
This arrangement allows statistically significant changes in fault rate to be
CA 02205421 1997-OS-15
WO 96/20549 PCTlGB95/03025
3
detected for both long and short periods, allowing sudden large changes to be
detected promptly, and also detecting smaller long-term changes which, because
of the larger sample size, are statistically significant, without false alerts
caused by
normal short-term statistical fluctuations.
' 5 This prior art system only monitors the number of seconds in which a fault
is present. It takes no account of the duration of the individual faults. For
example, a report of nine "errored seconds" in a 15 minute period could be
caused
by nine individual faults each of less than one second, or a single fault of
nine
seconds. It also makes no provision for separate faults occurring
simultaneously,
or of overlapping duration; only a maximum of one fault per second can be
recorded.
According to a first aspect of the invention, there is provided a method of
monitoring a telecommunications system for faults and generating alarms in
response to the faults, said method comprising the steps of monitoring the
system
for the occurrence of faults, and activating an alarm indicator if the number
of
occurrences of faults counted during a predetermined time interval is equal to
or
greater than a threshold value.
According to a second aspect, there is provided an apparatus for
monitoring a telecommunications system for faults, comprising an alarm
indicator,
detection means for detecting occurrences of faults, counting means for
counting
the number of occurrences of the faults detected by the detection means which
occur within a predetermined analysis period, and activation means for causing
the
alarm indicator to be activated if the number stored in the counting means
equals
or exceeds an activation threshold value.
This arrangement measures the frequency of occurrence of actual faults,
rather than, as in the prior arrangement discussed above, the proportion of
time
that a fault or faults are present. By generating alarm indications in
response to the
frequency of occurrence of faults a better indication is available to the
system
~ operator as to whether the fault requires attention, than if transient
occurrences
~ are alerted directly to the operator. For a user, the system can be used to
identify
the number of times the service is unavailable.
Preferably, the method comprises the steps of establishing an analysis
period;
CA 02205421 1999-11-29
WO 96/20549 PCTIGB95103025
4
monitoring the system continuously for the occurrences of faults;
at the end of a scan interval shorter than the analysis period, counting the
number of occurrences of the fault during the analysis period which ends at
the
end of the scan interval; and
activating an alarm indicator if the number of occurrences of the fault in
the analysis period is equal to or greater than a threshold value.
Preferably, the times of onset and clearing of faults are recorded, and at
the end of each scan interval faults for which a time greater than the
analysis
period has elapsed since the recorded clear time are not counted.
In a preferred arrangement, the alarm indicator is deactivated if the
number of occurrences of the fault in the analysis period is equal to or less
than a
second threshold value, the second threshold value being less than the first
threshold value.
Preferably the alarm indicator is maintained in its current state if the
number of ocurrences of the fault in the analysis period is between the first
and
second threshold values.
In another arrangement a further alarm indicator is also activated if the
number of occurrences of the fault in the analysis period is equal to or
greater than
a threshold value, and the further alarm indicator remains activated until
acknowledged by an operator.
Preferably the duration of each fault is measured, and if the duration of a
fault exceeds a predetermined value an alarm indicator is activated. In this
way a
fault which does not clear spontaneously within a predetermined period, and
must
be acted on, can be readily distinguished from transient fault reports which
can be
stored for subsequent analysis, without the system operator having to be
alerted
to every individual transient fault. This additional feature i~s the subject
of 'our co-
pending Canadian application No. 2,205,603 filed December 22, 1995.
Preferably the times of the onset and clearing of faults are recorded, and
after the onset time of a fault is recorded, a delay period is initiated and,
if the
clearing of the fault is recorded before the expiry of the delay period the
stored
value is incremented, and if the delay period expires before the clearing of
the fault
is recorded an alarm indicator is activated.
CA 02205421 1999-11-29
WO 96/2049
PCTlGB95/03025
Preferably also the system is monitored for occurrences of unavailability of
the system, or a function of the system, to a user.
This alarm indicator may be the same one as that activated if the number of
faults in an analysis period exceeds the threshold value. It may be arranged
to
5 remain activated as long as either the stored value exceeds the second
threshold
or a fault of duration greater than the predetermined value remains uncleared.
The
alarm may have different activation states dependant on whether the stored
value
remains above its threshold, or a long duration fault remains uncleared, or
both.
The counting means may comprise a store, arranged to store the number
of occurrences of the fault condition within each one of a plurality of scan
intervals
whose total duration is that of the analysis period. The durations of the
predetermined analysis period and of the scan interval may be selectable.
Preferably the equipment includes an updating processor arranged, at the
end of each scan interval, to retrieve the data stored in the store and to
supply it
to the counting means, and to instruct the store to delete the data relating
to the
earliest scan interval for which data is stored.
Preferably there is provided discrimination means for distinguishing
occurrences of the fault condition having a first predetermined characteristic
from
those having a second predetermined characteristic, the counting means being
arranged to count those occurrences having each characteristic separately, or
to
count only one of the types. The counting means may also be arranged to be
over-
ridden, thereby suspending its' operation. This allows routine tests and pre-
arranged
breaks in service to be discounted from the analysis, to avoid the generation
of
false alarms.
An embodiment of the invention will now be described, by way of example
only, with reference to the drawings, in which
Figure 1, is a diagrammatic representation of a sequence of faults
occurring in a telecommunications system;
Figure 2 is a functional block diagram showing the various components of
an apparatus according to the invention for monitoring a telecommunications
network for intermittent fault conditions, together with elements of the
network to
be monitored; and
CA 02205421 1997-OS-15
WO 96/20549 PCTIGB95I03025
6
Figure 3 shows the general construction of a computer suitable for
operation of the invention.
Figure 1 shows a time sequence of transient faults of a particular type
occurring in a telecommunications system which is to be monitored by an
apparatus embodying the invention. For example, the system may be a
telecommunications network and the faults may be overloads in a communication
channel, and breaks in service to a specific customer.
In the monitoring apparatus, to be described below with reference to
Figure 2, the time is divided into a number of scan intervals ti, t2, etc each
of
length t. In this embodiment the scan interval is substantially longer than
the
duration of the faults, and it is possible for more than one fault to occur in
the
same scan interval. An analysis period T is defined. In this illustrative
example the
analysis period T is three times the length of the scan interval t. However,
in
practice the analysis period could be very much longer than this.
In this invention, a distinction is made between transient faults and those
of longer duration. When a fault is identified a delay period is initiated.
The length
of the delay period may be configurable for different customers and for
different
services. If the fault does not clear during the delay period an alarm
indicator is
triggered. This alarm indicator may be the same as the one triggered if the
threshold number is exceeded, but preferably the alarms are distinct from each
other, or the alarm has distinct activation states, so that single long
interruptions
and a series of short interruptions can be distinguished, allowing the
operator to
prioritise his actions in response to the alarms. However, if the fault does
clear
during the delay period the fault is classed as a transient alarm and a
service level
transient fault score for the current scan interval increased by one. If the
. predetermined fault occurs and clears more than once in the same scan
interval the
final transient fault score for the scan interval will be the number of times
the fault
occurred in that interval. A separate alarm may be provided which remains
activated after a transient fault is reported, until the alarm is cleared by
being
acknowledged by the operator.
In this example transient faults A, B, C, D, E, occur in scan intervals t2, t4
(twice), tfi, and t~ respectively.
CA 02205421 1997-05-15
WO 96/20549 PCT/GB95/03025
7
At the end of each scan interval the apparatus counts the number of faults
which have occurred during the analysis period T, in this example comprising
the
last three scan intervals, and an alarm indicator is triggered if the number
reaches
or exceeds a threshold value, in this example set at 3. Only when the number
in
the analysis period falls to or below a second threshold value, in this
example 1, is
the alarm .indicator cleared. Different thresholds are selected to avoid the
occurrence of intermittent alarms which might otherwise occur when the rate of
occurrence of transient fault is close to a single threshold value. For
example, at
the end of the scan interval t3 only one fault, A, is counted in analysis
period
ti +t2+t3. At the end of the next scan interval t4 three faults, A, B, C are
counted
in the analysis period t2+t3+t4 and the alarm indicator is triggered. At the
end of
the next scan interval t5 only two faults are counted, because fault A is now
outside the analysis period t3+t4+t5. Although two faults is below the alarm
threshold value of 3, the alarm is not cleared because the number of faults in
the
analysis period has not fallen to the second threshold value.
At the end of scan interval is there are once again three faults; B, C, D,
within the analysis period. At the end of scan interval t7 there are only two
faults,
D and E, because although fault E (in scan interval t~) is added, faults B and
C are
now outside the analysis period t5+ts+t7. However, the alarm indicator again
remains on, because the 'clear' threshold has not been passed. At the end of
period t9, only fault E is within the analysis period t~+t8+t9, and so the
number of
faults has fallen to the threshold and the alarm indicator clears. However,
had
fault 'E' been followed in period t8 or t9 by a group of one or more further
faults,
the second threshold would not have been reached and the alarm indicator would
not clear.
It should be noted that the fault E, although starting in period ts, is
counted as being in period t~, the period in which it clears. This avoids
double
counting of fault E. The clear time is used because at the end of scan
interval tfi
the duration of the fault is indeterminate, and it may be a long duration
fault Csee
fault F in scan intervals t9, tio).
' Referring now to Figure 2, there is shown a functional block diagram of
the functional components of an apparatus 20 for monitoring faults in a
telecommunications network 21, the network including elements 2 to 5. These
CA 02205421 1997-OS-15
WO 96120549 PCT/GB95103025
8
elements may be functional components of the network, or they may be elements
of the service provided by the network to a customer.
The apparatus 20 is realised on a computer of conventional construction
as shown in Figure 3 and which comprises a memory 220, a display screen 222
and keyboard 224, a central processing unit 226 and an interface 228. The
memory 220 may be implemented as a combination of a hard disk, random access
memory (RAM) and Read-only Memory (ROM).The computer has a program stored
in its memory 220 and the program includes a set of program modules
corresponding to the functional components 1, 6, 7, 8, 9, 10, 1 1, 12, 13, 14,
15
and 16 shown in Figure 2.
The apparatus 20 comprises a fault monitor 1 for monitoring the network
21 for faults in various elements 2, 3, 4, 5 of the network 21. The monitor 1
provides an output first to a fault duration discriminator 6 which identifies
whether
the fault is of long or short duration. If the fault has cleared within a
predetermined time the discriminator 6 causes information about the fault to
be
transmitted to store 8. The information stored about each fault includes
information about the network elements 2, 3, 4 or 5 to which it relates, the
time
of onset of the fault, and the time of clearing the fault. The times of onset
and
clearing may be reported by the relevant network element 2, 3, 4, or 5, or by
the
monitor 1. If the fault has not cleared within the predetermined time the
discriminator causes an alarm 7 to be activated, to indicate a non-transient
fault to
the user on the display 222. For some purposes it may also be necessary for a
user to be alerted to transient faults as they occur, for example if equipment
needs
re-setting after an interruption. If such an arrangement is required the alarm
7 is
activated whenever the monitor 1 detects a fault.
An updating processor 9 periodically retrieves the data from store 8, at a
scanning interval t controlled by a timer 10. The updating processor 9
instructs
the store 8 to delete information relating to any faults for which a period T
(the
analysis periodl has elapsed since they cleared. The analysis period T and
scan
interval t are both selectable, subject to the scan interval not exceeding the
analysis period. The analysis period T is typically several times the scan
interval t,
so that at the end of each scan interval the faults in the previous T/t scan
intervals
are retrieved. The processor 9 then passes the data to a counter.11 which
counts
CA 02205421 1997-OS-15
WO 96/20549 PCT/GB95/03025
9
the number of faults relating to each element 2, 3, 4, 5 which cleared within
the
previous analysis period. The values obtained are passed to a comparator 12
which compares the values with activation and deactivation threshold values
stored in threshold store 13, and the result is passed to an alarm controller
14.
The alarm controller 14 also receives an input from an alarm state monitor 15,
and
causes the alarm indicator 7 to be operated (activated or cleared) in
accordance
with its pre-existing state and the results from the comparator 12 according
to the
truth table below.
EVENT COUNT
pFACTIVATION BETWEEN ACTIVATION
THRESHOLD THRESHOLDS THRESHOLD
OR BELOW OR ABOVE
SWITCH OFF LEAVE ON LEAVE ON
ACTIVATED
ALARM NOT LEAVE OFF LEAVE OFF SWITCH ON
ACTIVATED
Thus when the count value is equal to or less than the deactivation
threshold the alarm indicator is off. If it rises to the deactivation
threshold the
alarm indicator 7 remains 'off' until, if it rises to or above the activation
threshold,
the alarm indicator is switched on. If it is between the thresholds the alarm
indicator remains in the 'on' condition. If it falls further, to the
deactivation
threshold or below, the alarm is switched off.
The alarm indicator may be held in its activated state until a predetermined
number of scan intervals have elapsed since its activation, even if the
deactivation
threshold is passed, in order to allow sufficient time for it to come to the
attention
of the operator. Alarms may also be cleared manually. A separate alarm may be
provided which remains activated.after a transient fault is reported, until
the alarm
is cleared by being acknowledged by the operator
A user input 16 associated with the keyboard 224 or another input device
(e.g. a "mouse") allows the fault duration used by the fault duration
discriminator
6, the scan interval used by timer 10, the analysis period used by updating
processor 9, and the threshold values stored in threshold store 13 to be
selected,
CA 02205421 1997-OS-15
WO 96/20549 PCT/GB95/03025
and allows the selection of those of the elements 2, 3, 4, 5 which are to be
monitored by monitor 1.
The alarm indicator 7 provides information to the user concerning which of
the elements 2, 3, 4, 5 has caused the alarm indicator 7 to be activated and
5 whether the activation was because of a long duration fault or an
accumulation of
shorter faults.
The operation of the apparatus will now be described with reference to the
sequence of faults shown in Figure 1. For the sake of illustration, only a
single
element 2 is monitored, the longlshort fault threshold is set to 5 minutes,
the scan
10 interval to 20 minutes, the analysis period to 1 hour, the alarm indicator
activation
threshold to 3 and the alarm indicator deactivation threshold to 1 . These
values
have been selected for illustrative purposes only, and are not necessarily
representative of appropriate values for a practical system. It is assumed
that no
faults are recorded in store 8 at the start of the time period illustrated,
and that the
alarm indicator 7 is not already activated.
Timer 10 triggers the updating processor 9 to operate every 20 minutes.
At the first such operation, at the end of scan interval t~, store 8 contains
no data,
so the updating processor 9 retrieves no data from store 8. Counter 11 passes
a
zero result to comparator 12 which compares this result with the activation
and
de-activation thresholds. Since the zero count is less than both thresholds
the
alarm indicator 7 is not activated. During the second scan interval t2 fault A
is
detected by monitor 1. This fault clears before the 5 minute threshold set by
discriminator 6 and so data concerning fault A is passed to store 8. At the
end of
the scan interval updating processor 9 retrieves the data on fault A, but does
not
delete fault A from store 8, because a time less than the analysis period T
has
elapsed since its occurrence. The data is passed to counter 11 and the value.
1 is
compared in comparator 12 with the values stored in threshold store 13. The
counter is now at the deactivation threshold, but the alarm indicator 7 is
already
not activated and the count is still less than the activation threshold so the
alarm
. indicator 7 is not activated.
Similarly, at the end of the scan interval t3 no faults have been added to
store 8, and no faults have been in the store for longer than the analysis
period T,
CA 02205421 1997-OS-15
WO 96/20549 PCT/GB95/03025
11
so the updating, counting and comparing processes are as for the previous scan
interval t2.
During scan interval t4 two short faults B, C occur. These are added to
store 8 in the same way that fault A was recorded in scan interval tz. At the
end
of scan interval t4 the counter 11 now counts three faults. This value 3 is
compared with the value stored in threshold store 13 and is found to be at the
activation threshold value. The ~ output from the comparator 12 is input to
alarm
indicator control 14 together with an input from the alarm indicator state
monitor
15, which indicates that the alarm indicator is currently not activated. This
causes'
the alarm control 14 to activate the alarm indicator 7.
During scan interval t5 no further faults occur. At the end of the scan
interval t5 the updating processor identifies fault A as having occurred more
than
time T before the present, having occurred in scan interval tz. Fault A is
thus
deleted from the store 8. The counter 1 1 therefore only counts two faults,
(faults
B and C) and this value 2 is compared with the threshold values stored in
store 13.
Although the value has now fallen below the activation threshold it is still
above
the deactivation threshold so that alarm indicator 7 remains in its present
(activated) state.
During scan interval is a further fault D occurs, and another fault E is
taking place as the scan interval ends. As fault E has not yet cleared it
cannot yet
be determined whether it will exceed the long/short discrimination threshold.
At the end of scan interval is the updating processor therefore finds three
faults, B, C, and D stored in store 8, none of which has been stored for
longer than
the full analysis period T. They are thus not deleted from store 8, and are
all
counted by counter 1 1. The value of 3 determined by the counter 11 is
compared
with the threshold values stored in the threshold store 13 and found to be at
the
activation threshold. Since the alarm indicator state monitor 15 identifies
the
alarm indicator 7 as already activated a new alarm indication is not
generated.
In scan interval t~ fault E clears within the long/short discrimination
threshold and so is stored in store 8. At the end of the scan interval t~ the
updating processor updates store 8 by deleting faults B and C, since the
analysis
period T has now elapsed since they occurred. The counter 11 therefore counts
only two faults, D and E, and this value of two is compared by comparator 12
with
CA 02205421 1997-OS-15
WO 96/20549 PCT/GB95/03025
12
the threshold values. Although the value has again fallen below the activation
threshold it is still above the deactivation threshold so the alarm indicator
7
remains in its present (activated) state. Similarly in scan interval tg no new
faults
occur and none are deleted by the updating processor 9 so the number of faults
counted by the counter 11 remains at two and the alarm indicator 7 remains
activated..
Towards the end of the scan interval t9 another fault F starts, but has not
reached the long/short discrimination threshold or cleared, before, the end of
the
scan interval. At the end of the period t9 fault D is deleted from store 8 by
the
updating processor 9 as the analysis period T has now expired for this fault.
This
leaves only fault E in store 8 to be counted by counter 1 1. The value output
from
counter 1 1 to comparator 12 is therefore 1, which is the deactivation
threshold.
The alarm control 14 therefore deactivates alarm indicator 7.
During scan interval too, the long/short discrimination threshold expires
before fault F clears. At the expiry of this discrimination period, the
discriminator
6 identifies the fault as a long fault and activates alarm indicator 7
immediately.
In certain circumstances the system may receive a 'clear' indication when
no 'fault' indication has been received. This may occur for example when the
system is first set up, or if the 'fault' signal is not received because of
another
more general fault making it. The system is arranged to disregard any such
'unpaired' clear indication.
The alarm indicator 7 may give information about the faults such as which
of the elements 2, 3, 4 or 5 is generating the faults, their total duration
and time
of occurrence.
The apparatus 20 described above can be used to monitor the
performance of a network at any level. For example, a network operator
responsible for maintenance of telecommunications equipment may wish _ to
monitor for failures of specific items of equipment. Individual transient
interruptions may be insignificant - perhaps caused by an external cause. For
example if an equipment failure occurs at one point in the system it will
cause an
interruption to the operation of many other items of equipment in
communication
with it. If the network has so-called "self-healing" capabilities, calls can
be re-
routed to avoid the use of the failed equipment, and only a single transient
CA 02205421 1997-OS-15
WO 96/20549 PCT/GB95103025
13
interruption is detected for the other items. However, if a large number of
transient faults are detected at a single item of equipment, this can be
indicative of
a developing situation in which the performance of that item is deteriorating.
This
information can be used to take remedial action before the component fails
completely, either by re-routing call traffic to avoid or minimise its use, or
by
repairing the component.'
A customer of a service provided by the network is usually not concerned
with individual equipment failures provided that the service itself is
uninterrupted.
Such customers often have service-level agreements with the network operator
which specify maximum levels of service interruption (either as a number of
individual interruptions or as a proportion of total timel. The alarm
indicator / may
include means for recording the details of these interruptions to service to
be
monitored and recorded so that the network operator and/or the customer can
have visibility of the performance of the system. The alarm indicator 7 may
monitor the stored details to identify when the proportion of time, or the
total
elapsed time, for which the service has been interrupted exceeds a
predetermined
value, and triggering the alarm when this value is exceeded.
