Availability definition in ITUT Presentation for MEF

1
Availability definition in ITU-TPresentation for
MEF

• Dávid Jocha (ETH/RLN)

2
Overview

• This slideset is an overview of ITU-T's Ethernet
  Availability definition for the MEF Aligning
  Availability ad hoc group
• Contents
• Related ITU-T recommendations
• ATM, IP, MPLS,
• Sliding window comparison
• Y.1563 (Ethernet)
• Definition
• Answers to questions from previous meeting
• Universal definition

3
Related ITU-T Recommendations

• Earlier definitions, e.g.
• X.137 (1988, 1997) Availability performance
  values for public data networks when providing
  international packet-switched services
• ATM 
• I.357 (1996, 2000) B-ISDN semi-permanent
  connection availability   
• PDH/OTN
• G.827 (1996, 2003) Availability performance
  parameters and objectives for end-to-end
  international constant bit-rate digital paths
• IP
• Y.1540 (1999, 2007) Internet protocol data
  communication service - IP packet transfer and
  availability performance parameters
• Y.1544 (2008) Multicast IP performance
  parameters
• Common
• G.7710 / Y.1701 (2001, 2007) Common equipment
  management function requirements
• MPLS
• Y.1561 (2004) Performance and availability
  parameters for MPLS networks

4
ITU-T I.357 (ATM)

• A given second is considered to be an SESATM if
• User information cells are presented during this
  period of time to the connection portion and
  either the Cell Loss Ratio (CLR) 1/1024 or the
  Severely Errored Cell Block Ratio (SECBR) 1/32,
  where CLR and SECBR are computed over the
  considered period of time
• NOTE   The above CLR threshold is intended to
  support QoS classes in which the CLR objective is
  ? 105. Appropriate CLR thresholds for other QoS
  classes are for further study.
• User information cells are not presented during
  this period of time to the connection portion,
  but the ATM connection is considered to be unable
  to provide acceptable cell transfer performance,
  because an interruption has occurred within the
  connection portion. This interruption prevents
  cells from being transmitted on the connection
  portion during the considered one-second period
  of time, should the user attempt to transmit
  cells. An interruption corresponds to a failure
  occurring within the connection portion, either
  of the physical layer or of the ATM layer.
• Availability
• The onset of unavailability begins with the
  occurrence of ten consecutive SESATM. These ten
  seconds are part of unavailable time.
• A period of unavailability ends with the
  occurrence of ten consecutive seconds, none of
  which are SESATM. These ten seconds are part of
  available time.
• The ten-second criteria are supported using a
  sliding window with one-second granularity.
• Availability Ratio
• The AR is defined as the proportion of scheduled
  service time that the connection portion is in
  the available state. The AR is calculated by
  dividing the total service available time by the
  duration of the scheduled service time.
• During the scheduled service time the user may or
  may not transmit cells.

5
ITU-T G.827 (CBR Paths)

• Availability
• A period of unavailable time begins at the onset
  of ten consecutive Severely Errored Second (SES)
  events. These ten seconds are considered to be
  part of unavailable time. A new period of
  available time begins at the onset of ten
  consecutive non-SES events (a non-SES event is a
  second that is an errored second, but not an SES,
  or is error free). These ten seconds are
  considered to be part of available time.
• A bidirectional path or connection is in the
  unavailable state if either one or both
  directions are in the unavailable state
• SES
• The specification of SESs is dependent upon the
  system under consideration (please refer to the
  appropriate Recommendation).
• Available time
• Available time is obtained by adding all the
  periods of available time from the observation
  period in a generic observation period.
• Unavailable time is derived by adding all the
  periods of unavailable time during a generic
  observation period.
• Availability is defined as the percentage of
  available time (to total time) in a generic
  observation period. This is also known as the
  Availability Ratio (AR).
• Unavailability is defined as the percentage of
  unavailable time (to total time) in a generic
  observation period. This is also known as the
  Unavailability Ratio (UR).

6
ITU-T Y.1561 (MPLS)

• SLB
• A severe loss block (SLB) outcome occurs for a
  block of packets observed during time interval
  Tlb (1 second ) at ingress MP0 when the ratio
  of lost packets at egress MPi to total packets in
  the block exceeds s1(0.15).
• Evaluation of successive blocks (time intervals)
  should be non-overlapping.
• The minimum number of packets that should be used
  in evaluating the severe loss block outcome is
  Mlb.(FFS)
• Availability MPLS service availability function
• The onset of unavailability begins with the
  occurrence of ten consecutive SLBs. These ten
  seconds are part of unavailable time.
• A period of unavailability ends with the
  occurrence of ten consecutive seconds, none of
  which are SLB. These ten seconds are part of
  available time.
• The ten-second criteria are supported using a
  sliding window with one-second granularity.
• Percentages
• Percent MPLS service unavailability (PIU) The
  percentage of total scheduled service time that
  is categorized as unavailable using the MPLS
  service availability function.
• Percent MPLS service availability (PIA) The
  percentage of total scheduled service time that
  is categorized as available using the MPLS
  service availability function.

7
ITU-T Y.1540 (IP)

• Availability
• IP service availability function The IP service
  is available on an end-to-end basis if the packet
  loss ratio for that end-to-end case is smaller
  than a given threshold.
• If IPLR 0.75 ? unavailable state
• If IPLR
• The minimum number of packets that should be used
  in evaluating the IP service availability
  function is Mav 1000
• The minimum duration of an interval of time
  during which the IP service availability function
  is to be evaluated is Tav 5 minutes
• Percentages
• Percent IP service unavailability (PIU) The
  percentage of total scheduled IP service time
  (the percentage of Tav intervals) that is (are)
  categorized as unavailable using the IP service
  availability function.
• Percent IP service availability (PIA) The
  percentage of total scheduled IP service time
  (the percentage of Tav intervals) that is (are)
  categorized as available using the IP service
  availability function.
• Directionality
• This unidirectional definition of availability is
  motivated by the fact that IP packets often
  traverse very different routes from SRC to DST
  than they traverse from DST to SRC. If, from an
  IP network user perspective, a bidirectional
  availability definition is needed, a
  bidirectional definition can be easily derived
  from this unidirectional definition.

8
ITU-T Y.1544 (multipoint IP)

• Point-to-Multipoint IP service availability
  parameters
• The point-to-point unidirectional availability
  service function defined in Y.1540 should be used
  to evaluate the availability of the multicast
  path between a Source and any individual
  Destination.
• Group IP Service Availability Given D, a set of
  N Destinations (or group) intending to receive
  packets from a Source, the Point-to-Multipoint IP
  Service Availability parameter is defined as the
  ratio of Destinations in the (point-to-point)
  available state, Nav, (during a specific
  evaluation interval Tav), and the total
  destinations N (where point to-point availability
  is as specified in Y.1540).The fraction of
  destinations in the available state during Tav
  can be expressed as
• Mean Group IP Service Availability The mean
  fraction of available destinations over a
  result-recording interval, , iswhere I is an
  integer.

9
G.7710 (EMF)

• Parameters (definitons from G.828)
• Errored Block (EB) A block in which one or more
  bits are in error
• Errored Second (ES) A one second period with one
  or more errored blocks or at least one defect
• Severely Errored Second (SES) A one-second
  period which contains ?30 errored blocks or at
  least one defect. SES is a subset of ES.
• Background Block Error (BBE) An errored block
  not occurring as part of an SES
• Severely Errored Period (SEP) A sequence of
  between 3 to 9 consecutive SES. The sequence is
  terminated by a second which is not a SES
• Background Block Count (BBC)
• UnAvailable Second (UAS)
• Definition
• A period of unavailable time begins at the onset
  of x consecutive SES events. These x seconds are
  considered to be part of unavailable time. A new
  period of available time begins at the onset of x
  consecutive non-SES events. These x seconds are
  considered to be part of available time. SEP
  indicates a severe error condition, which does
  not result in unavailability.

10
Sliding window
11
The (non-)sliding window effect
Non-sliding window
available
available
available
available
available
Indicator
Sliding window
available
unavailable
available

• In this simplified example n3, which is the
  number of consecutive small time intervals over
  which we assess availability. Problematic
  sequences for any "n can be found, except n1,
  but in this case availability turns into SES
  (severely errored seconds) or SLB (severe loss
  block).
• The threshold(s), Cu and Ca are neglected in this
  example (note ITU-T uses only one threshold).
  The "indicator" shows if the small time interval
  is considered as
• 0 "acceptable" non_SES non_SLB, or
• 1 "unacceptable" SES SLB.
• "n" is usually 10 in the ITU-T definitions, 4 in
  the MEF example.

12
Simulation

• Simulations were made to compare the sliding and
  non-sliding window approach
• Non-real data, non-real loss
• The input signal was simplified to be over or
  under the threshold
• As seen on the figure in previous slide
• Was created by a two-state Markow-chain with 0.05
  and 0.25 change probabilities, the run length was
  1000.
• The window size was 4 (except for the last figure)

13
Results (1)

• The sliding-window provides finer granularity
• Details and zoomed parts of the figure on next
  slides

14
Results (2)

• The sliding window will declare more (shorter)
  intervals to be unavailable.
• For example consider the unavailable pink
  interval. Its length is more than 4 seconds, but
  only less than 4 seconds fall to a given
  non-sliding window period, therefore it is not
  recognized as unavailable by the non-sliding
  window method.

15
Results (3)

• The sliding-window provides better accuracy.
• The unavailable interval is long enough to be
  recognized by both methods, but because of the
  unlucky interval borders, in this example the
  start of unavailability according to the
  non-sliding window is delayed by 1, the end of
  the interval is delayed by 2 units.

16
Results (4)

• Top figure is the original
• Bottom figure only the probabilities were
  changed (to 0.99 and 0.95)
• Here small differences similar to previous
  figures are still present, but they can be
  neglected

• In conclusion if there are relatively short
  unavailable intervals, then the difference
  between the sliding and the non-sliding window
  results is significant
• Sliding window is recommended
• However, if the unavailable intervals are
  relatively rare, but long then the difference is
  not significant
• Both methods are suitable

17
Y.1563
18
SES

• The Ethernet service availability definition is
  based on a model which uses two states
  corresponding to the ability or inability of the
  network to sustain the service in the available
  state. Transitions between the states of the
  model are governed by the occurrence of patterns
  of Severe Errored Seconds in the Ethernet layer
  (SESETH). This Recommendation views availability
  from the network perspective, where availability
  performance is characterized independently of
  user behaviour.
• A Severe Errored Second (SESETH) outcome occurs
  for a block of frames observed during a one
  second interval at ingress MP0 when the
  corresponding FLR (i.e. the ratio of lost frames
  to total frames in the block) at egress MPi
  exceeds s1.
• A provisional value s1 of 0.5 is proposed, and
  different values may also be chosen depending on
  the Class of Service (CoS).
• Evaluation of successive one second intervals is
  non-overlapping. The provisional minimum number
  of frames that should be used in evaluating the
  SESETH outcome is M equal to 100 packets.
• Note that provisional values are subject to
  change following additional study and real
  network experience.

19
Availability

• A period of unavailable time begins at the onset
  of 10 consecutive SESETH outcomes. The
  corresponding period of time is considered to be
  part of unavailable time. During the unavailable
  time period, the Ethernet network is in
  unavailable state. A new period of available time
  begins at the onset of 10 consecutive non-SESETH
  outcomes. The corresponding period of time is
  considered to be part of available time. During
  the available time period, the Ethernet network
  is in available state. Figure 9 illustrates the
  definition of criteria for transition to/from the
  unavailable state.
• This definition of availability has been chosen
  to allow comparison with other link layer
  techniques.
• Because Ethernet service is bidirectional,
  Ethernet network is in the unavailable state if
  either one, or both directions, are in the
  unavailable state. The unidirectional
  availability can be measured by the criteria
  mentioned above.

20
Availability Performance Parameters

• Percent Ethernet service unavailability (PEU)
• The percentage of total scheduled Ethernet
  service time that is categorized as unavailable
  using the Ethernet service availability function.
• Percent Ethernet service availability (PEA)
• The percentage of total scheduled Ethernet
  service time that is categorized as available
  using the Ethernet service availability function
• PEU 100 PEA
• Typically, the users of performance parameters
  need to make comparisons with objectives. This
  section treats the point-to-multipoint parameters
  as a general case. Results collected for a
  population of interest and a set of registered
  destinations should be compared with an
  objective, O, as follows
• percent meeting objective (PMO) The percentage
  of total destinations with point-to-point
  performance that is categorized as meeting the
  stated objective for a specific population of
  interest. The objectives are evaluated over sets
  of point-to-point parameters

21
Considerations

• Availability definition in Y.ETHperf to be
  compatible with
• Y.1540, because the traditional Ethernet is a
  connectionless unreliable service, a server layer
  for the IP layer. Therefore the availability
  description should be similar to IPs.
• No window
• Y.1561, because recent Ethernet extensions, like
  IEEE 802.1Qay (PBB-TE) realize a
  connection-oriented approach, like MPLS.
  Therefore Ethernets availability description
  should be similar to MPLSs.
• Sliding window
• MEF 10.1, because it defines also availability
  for an Ethernet service, and compatibility is
  recommended
• Non-sliding window
• Consider universal definition
• Liaison
• The MEF liaison also inquired about the
  derivation of the Availability Service Function
  currently defined in Y.ETHperf. A point was made
  that a comparable MEF definition has been
  standardized for 4 years. There were several
  contributions that led to the current definition,
  including an examination of many definitions
  covering different technologies, and a
  communication from SG 15. The consensus is to use
  a long-standing ITU-T definition as a basis for
  the definition on Ethernet Service Availability.
  The contributions and presentations that led to
  this consensus are available below (from our
  recent Expert's meeting in Red Bank, NJ)
• In summary, we looked at most/all the existing
  definitions, including MEF's, before adopting the
  current definition. The main technical merits are
  simplicity, familiarity, and consistency with
  other ITU-T Definitions.

22
Answers to Questions from previous meetings

• Directionality of availability
• Loss/SES unidirectional
• Availability calculated as unidirectional
• Because Ethernet service is bidirectional,
  Ethernet network is in the unavailable state if
  either one, or both directions, are in the
  unavailable state.
• Performance parameters in the unavailable state
• The speed, accuracy, and dependability
  parameters of this Recommendation are intended to
  characterize Ethernet service in the available
  state.
• However, no measurements of these parameters
  made during periods of unavailable time are ever
  used for determining long-term frame transfer
  performance and mechanisms must be established to
  exclude all performance measurement results
  collected during unavailable periods from any
  estimations of long-term Ethernet frame
  performance parameters.
• Loss threshold
• A provisional value s1 of 0.5 is proposed, and
  different values may also be chosen depending on
  the Class of Service (CoS).
• Note that provisional values are subject to
  change following additional study and real
  network experience.
• Availability as a multipoint parameter
• The Percent meeting objective (PMO) approach
  can be used to interpret availability for a
  multipoint scenario
• Scheduled uptime/downtime
• Availability categorized without considering
  scheduled up/down time
• PEU/ PEA only scheduled time is taken into
  account

23
Universal definition
24
Comparison
25
Old proposal for Y.ETHperf

• Notes
• It was an old proposal from Ericsson to Y.1563
• Was too complex
• Showed how could ITU-T be compatible with MEF
• Focus was on ITU-T
• Something similar to MEF new Option 3 (Number of
  consecutive ?t for assessing an indicator) was
  not considered
• Option was included for sliding window
• Current conclusion
• Its easier for MEF to achieve compatibility with
  ITU-T than for ITU-T to be compatible with MEF

26
Old proposal text for Y.ETHperf

• A severe loss block (SLB) outcome occurs for a
  block of packets observed during time interval
  Tlb at ingress MP0 when the ratio of lost packets
  at egress MPi to total packets in the block
  exceeds s1.
• Evaluation of successive blocks (time intervals)
  should be non-overlapping.
• The minimum number of packets that should be used
  in evaluating the severe loss block outcome is
  Mlb.
• A severe non-loss block (SNLB) outcome occurs for
  a block of packets observed during time interval
  Tlb at ingress MP0 when the ratio of lost packets
  at egress MPi to total packets in the block not
  more than s2.
• Evaluation of successive blocks (time intervals)
  should be non-overlapping.
• The minimum number of packets that should be used
  in evaluating the severe non-loss block outcome
  is Mlb.
• Ethernet service availability function
• The onset of unavailability begins with the
  occurrence of n consecutive SLBs. These n times
  Tlb time intervals are part of unavailable time.
• A period of unavailability ends with the
  occurrence of n consecutive SNLBs. These n times
  Tlb time intervals are part of available time.
• The n times Tlb criteria are supported depending
  on the value of SW using a
• Non-sliding window (0) with nTlb granularity or
  a
• Sliding window (1) with Tlb granularity.

27
(No Transcript)