vMX

Expand all | Collapse all

Container LSP and auto bw

Jump to Best Answer
  • 1.  Container LSP and auto bw

    Posted 06-28-2020 10:40

    Hi everyone,

    Please consider the following example:

    ####################################################

    EXAMPLE.PNG

    ##############################################################

     

    1)What is the minimum BW a child LSP could have? is it 2m because merging-bandwidth is 2m or is it 1k because minimum-bw is 1k as defined under auto-bw?

    2)What is maximum BW a child LSP could have? is it 4m because splitting-bandwidth is 4m? or is it 10M becasue max-bw is 10M as defined under auto-bw?

     

    Thanks and have a good weekend!!



  • 2.  RE: Container LSP and auto bw
    Best Answer

    Posted 06-28-2020 23:10

    Hello,

     


    @sarahr202 wrote:

     

    1)What is the minimum BW a child LSP could have? is it 2m because merging-bandwidth is 2m or is it 1k because minimum-bw is 1k as defined under auto-bw?

     


     

    This container LSP does not have "minimum-signaling-bandwidth" knob configured, so 1K.

    "merging-bandwidth" takes effect only when merging, i.e. when sum of all member LSP utilizations is 2M or below and JUNOS needs to reduce the count of member LSPs. Container LSP cannot have less than 1 member.

     

     


    @sarahr202 wrote:

     

    2)What is maximum BW a child LSP could have? is it 4m because splitting-bandwidth is 4m? or is it 10M becasue max-bw is 10M as defined under auto-bw?

     

     

     

    Please define "have". If "have" means :

    a/ the instantaneous load a member LSP can carry - then up to the smallest link BW along the path,

    b/ measured (==measured by JUNOS AutoBW feature) member LSP load && if "maximum-member-lsp" is _NOT_ exceeded - it is 4M, because on exceeding measured member LSP load above 4M, this member LSP will be split.   

    c/ measured (==measured by JUNOS AutoBW feature) member LSP load && if "maximum-member-lsp"  _IS_ exceeded - it is 10M, because on exceeding measured member LSP load above 4M, this member LSP cannot be split anymore.   

     

    HTH

    Thx

    Alex



  • 3.  RE: Container LSP and auto bw

    Posted 07-01-2020 18:44

    Thanks Alex,

     

    @sarahr202 wrote:

     

    1)What is the minimum BW a child LSP could have? is it 2m because merging-bandwidth is 2m or is it 1k because minimum-bw is 1k as defined under auto-bw?

    This container LSP does not have "minimum-signaling-bandwidth" knob configured, so 1K.

    "merging-bandwidth" takes effect only when merging, i.e. when sum of all member LSP utilizations is 2M or below and JUNOS needs to reduce the count of member LSPs. Container LSP cannot have less than 1 member.

    ########################################################################

    It looks like this book is probably wrong ( MPLS in SDN era, the only book I could find about Container LSP)

    Page 569:

    protocols mpls {
    2 label-switched-path LSP-TEMPLATE {
    3 template;
    4 least-fill;
    5 adaptive;
    6 auto-bandwidth {
    7 adjust-interval 600;
    8 adjust-threshold 10;
    9 minimum-bandwidth 1k;
    10 maximum-bandwidth 10m;
    11 adjust-threshold-overflow-limit 3;
    12 resignal-minimum-bandwidth;
    13 }
    14 }
    15 container-label-switched-path PE1--->PE3 {
    16 label-switched-path-template LSP-TEMPLATE;
    17 to 172.16.0.33;
    18 splitting-merging {
    19 maximum-member-lsps 4;
    20 minimum-member-lsps 1;
    21 splitting-bandwidth 4m;
    22 merging-bandwidth 2m;
    23 normalization normalize-interval 1200;

    example continue to page 571

    Page 571:

    "Now imagine that the traffic generator brings down the traffic rate from 8 Mbps to
    7.5 Mbps. After the next normalization interval (2 minutes) fully expires, the maximum
    bandwidth sample is around 7.5 Mbps and there are two options to distribute
    this traffic:
    • 3 (member LSPs) x 2.50 Mbps (each member LSP)
    • 2 (member LSPs) x 3.75 Mbps (each member LSP)"

    ###############################################################

    Here is my thought

    Whethet to merge LSP  ( reduce the current number of LSP) is determined by "merging-bandwidth 2m" knob as configured in above excerpt.  Based on this , no merging should be happening as the sum of LSP utilization is not equal or less than 2m but then contradicts author 's statements ( as highlighted above).

    Am i missing something ?

    Thanks

     

     

     

     

     

     

     

     

     

     



  • 4.  RE: Container LSP and auto bw

    Posted 07-01-2020 23:15

    Hello,

     


    @sarahr202 wrote:

    MPLS in SDN era, the only book I could find about Container LSP)

     

    There is a free TE++ whitepaper on Juniper.net as well https://www.juniper.net/assets/us/en/local/pdf/whitepapers/2000587-en.pdf

     

     


    @sarahr202 wrote:

     

    Page 571:

    "Now imagine that the traffic generator brings down the traffic rate from 8 Mbps to
    7.5 Mbps. After the next normalization interval (2 minutes) fully expires, the maximum
    bandwidth sample is around 7.5 Mbps and there are two options to distribute
    this traffic:
    • 3 (member LSPs) x 2.50 Mbps (each member LSP)
    • 2 (member LSPs) x 3.75 Mbps (each member LSP)"

    ###############################################################

    Here is my thought

    Whethet to merge LSP  ( reduce the current number of LSP) is determined by "merging-bandwidth 2m" knob as configured in above excerpt.  Based on this , no merging should be happening as the sum of LSP utilization is not equal or less than 2m but then contradicts author 's statements ( as highlighted above).

     


     

     

    This example is valid for a scenario where "merging-bandwidth" and "splitting-bandwidth" are not configured at all, therefore using defaults of 1bps and 1bps respectively

    https://www.juniper.net/documentation/en_US/junos/topics/reference/configuration-statement/splitting-merging-edit-protocols-mpls-container-label-switched-path.html

     

    The whole purpose of "merging-bandwidth" and "splitting-bandwidth" knobs is to reduce the volume of (re)signaling when the number of active member LSPs changes.

     

    I believe the author forgot to mention this in his book.

     

     HTH

    Thx

    Alex

     

     



  • 5.  RE: Container LSP and auto bw

    Posted 07-03-2020 14:57

    Thanks for the white paper and clarification.

     



  • 6.  RE: Container LSP and auto bw

    Posted 07-04-2020 20:24

    Hi Alex,

    The white paper link you provided has some contradictory info:

     

    Page#7:

    The ingress computes the number of LSPs to signal by dividing the aggregate demand by the maximum signaling
    bandwidth. The ingress’ attempt to bring up an LSP starts with this number of members and maximum signaling
    bandwidth per member. If, after a normalization event, the ingress determines this setup hasn’t been successful, TE++
    increases the number of LSPs by one and derives the appropriate per-member bandwidth to meet the aggregate demand

     

    Page#8

    TE+++ in Action

    Example:

    Figures 3 through 5 illustrate the splitting and merging processes of Juniper TE++. Minimum-signaling bandwidth and
    maximum-signaling bandwidth are two thresholds used during normalization to determine how many LSPs there should
    be and the bandwidth associated with each member. Let’s assume the following configuration:
    • Minimum signaling bandwidth: 2 Gbps
    • Maximum signaling bandwidth: 8 Gbps
    • Merging bandwidth: 2 Gbps
    • Splitting bandwidth: 9 Gbps
    In each of the figures below, dashed lines indicate the old instance of an LSP that is in the process of getting removed.
    In Figure 3, a container LSP creates a minimum number (1) LSP (A-E-1) with a minimum signaling bandwidth of 2 Gbps.
    As traffic begins moving over the LSP, Ingress A will compute the aggregate bandwidth from the samples. Let’s say the
    LSP grows to 7 Gbps owing to auto-bandwidth adjustment.

    Capture-1.PNG

    In Figure 4, the LSP container traffic surges to 10 Gbps. Normalization could be triggered as a result of failure to reserve
    a 10 Gbps member LSP when failover normalization is configured, or it could be triggered due to normalization timer
    expiry. The computation would decide to split LSPs as the per-member utilization has crossed the splitting-bandwidth
    threshold of 9 Gbps. As a result of splitting, one new member LSP A-E-2 is created with 5 GB of bandwidth and the
    existing member LSP A-E-1 is re-signaled with 5 GB of bandwidth in a make-before-break way.

     

    Capture-2.PNG

     

    ##################################################################################

     Highlighted lines above contradict each other mentioned below again  for easy reference:

    The ingress’ attempt to bring up an LSP starts with this number of members and maximum signaling
    bandwidth per member.

    Contradicts:

    As a result of splitting, one new member LSP A-E-2 is created with 5 GB of bandwidth and the
    existing member LSP A-E-1 is re-signaled with 5 GB of bandwidth in a make-before-break way.

     

    The maxium signalling BW is 8Gbs, so two  member LSPs should be signaled with 8Gbs.

    What am I missing here again?

     

    Thanks!!

     



  • 7.  RE: Container LSP and auto bw

    Posted 07-05-2020 04:56

    Hello,

    I don't see any controversy here.  If You think of "maximum-*" as a ceiling, and not a target, then everything makes sense.

     


    @sarahr202 wrote:

     

    The ingress’ attempt to bring up an LSP starts with this number of members and maximum signaling
    bandwidth per member.

     

    The algorithm to determine number of LSP and actual signaling BW does not stop here.

     

     


    @sarahr202 wrote:

     

    Contradicts:

    As a result of splitting, one new member LSP A-E-2 is created with 5 GB of bandwidth and the
    existing member LSP A-E-1 is re-signaled with 5 GB of bandwidth in a make-before-break way.

     

    The maxium signalling BW is 8Gbs, so two  member LSPs should be signaled with 8Gbs.

     

    "Maximum-sgnaling-BW" is the ceiling, the actual signaling BW is taken from last AutoBW measurement/determined by AutoBW LSP utilization. It may be actually unequal per member if the paths to tailend have differing BW, my customer has such setup in production.  

     

    HTH

    Thx

    Alex