Frequently Asked Questions About the Link Services Interface

Use answers to the following questions to solve configuration problems on a link services interface:

Which CoS Components Are Applied to the Constituent Links?
What Causes Jitter and Latency on the Multilink Bundle?
Are LFI and Load Balancing Working Correctly?
Why Are Packets Dropped on a PVC Between a J Series Device and Another Vendor?

Which CoS Components Are Applied to the Constituent Links?

Problem—I have configured a multilink bundle, but I also have traffic without MLPPP encapsulation passing through constituent links of the multilink bundle. Do I apply all CoS components to the constituent links, or is applying them to the multilink bundle enough?

Solution—On a J Series device you can apply a scheduler map to the multilink bundle and its constituent links. Although you can apply several CoS components with the scheduler map, configure only the ones that are required. We recommend that you keep the configuration on the constituent links simple to avoid unnecessary delay in transmission.

Table 87 shows the CoS components to be applied on a multilink bundle and its constituent links. For more information, see the Junos Class of Service Configuration Guide.

Table 87: CoS Components Applied on Multilink Bundles and Constituent Links

Cos Component	Multilink Bundle	Constituent Links	Explanation
Classifier	Yes	No	CoS classification takes place on the incoming side of the interface, not on the transmitting side, so no classifiers are needed on constituent links.
Forwarding class	Yes	No	Forwarding class is associated with a queue, and the queue is applied to the interface by a scheduler map. The queue assignment is predetermined on the constituent links. All packets from Q2 of the multilink bundle are assigned to Q2 of the constituent link, and packets from all the other queues are queued to Q0 of the constituent link.
Scheduler map	Yes	Yes	Apply scheduler maps on the multilink bundle and the constituent links, as follows: Transmit rate—Make sure that the relative order of the transmit rate configured on Q0 and Q2 is the same on the constituent links as on the multilink bundle. Scheduler priority—Make sure that the relative order of the scheduler priority configured on Q0 and Q2 is the same on the constituent links as on the multilink bundle. Buffer size—Because all non-LFI packets from the multilink bundle transit Q0 of constituent links, make sure that the buffer size on Q0 of the constituent links is large enough. RED drop profile—Configure a RED drop profile on the multilink bundle only. Configuring the RED drop profile on the constituent links applies a back pressure mechanism that changes the buffer size and introduces variation. Because this behavior might cause fragment drops on the constituent links, make sure to leave the RED drop profile at the default settings on the constituent links.
Shaping rate for a per-unit scheduler or an interface-level scheduler	No	Yes	Because per-unit scheduling is applied only at the end point, apply this shaping rate to the constituent links only. Any configuration applied earlier is overwritten by the constituent link configuration.
Transmit-rate exact or queue-level shaping	Yes	No	The interface-level shaping applied on the constituent links overrides any shaping on the queue. Thus apply transmit-rate exact shaping on the multilink bundle only.
Rewrite rules	Yes	No	Rewrite bits are copied from the packet into the fragments automatically during fragmentation. Thus what you configure on the multilink bundle is carried on the fragments to the constituent links.
Virtual channel group	Yes	No	Virtual channel groups are identified through firewall filter rules that are applied on packets only before the multilink bundle. Thus you do not need to apply the virtual channel group configuration to the constituent links.

What Causes Jitter and Latency on the Multilink Bundle?

Problem—To test jitter and latency on a J Series device, I sent three streams of IP packets. All packets have the same IP precedence settings. After I configured LFI and CRTP, the latency increased even over a non-congested link. How can I reduce jitter and latency?

Solution—To reduce jitter and latency, do the following:

Make sure that you have configured a shaping rate on each constituent link.
Make sure that you have not configured a shaping rate on the link services interface.
Make sure that the configured shaping rate value is equal to the physical interface bandwidth. For more information, see Applying Shaping Rates to Interfaces.
If shaping rates are configured correctly, and jitter still persists, contact the Juniper Networks Technical Assistance Center (JTAC).

Are LFI and Load Balancing Working Correctly?

Problem—I have a single network that supports multiple services. My network transmits data and delay-sensitive voice traffic. I configured MLPPP and LFI to make sure that voice packets are transmitted across the network with very little delay and jitter. How can I find out if voice packets are being treated as LFI packets and load balancing is performed correctly?

Solution—When LFI is enabled, data (non-LFI) packets are encapsulated with an MLPPP header and fragmented to packets of a specified size. The delay-sensitive, voice (LFI) packets are PPP-encapsulated and interleaved between data packet fragments. Queuing and load balancing are performed differently for LFI and non-LFI packets.

To verify that LFI is performed correctly, determine that packets are fragmented and encapsulated as configured. After you know whether a packet is treated as an LFI packet or a non-LFI packet, you can confirm whether the load balancing is performed correctly.

Solution Scenario—Suppose two J Series devices R0 and R1 are connected by a multilink bundle lsq-0/0/0.0 that aggregates two serial links, se-1/0/0 and se-1/0/1. On R0 and R1, MLPPP and LFI are enabled on the link services interface and the fragmentation threshold is set to 128 bytes.

In this example, we used a packet generator to generate voice and data streams. You can use the packet capture feature to capture and analyze the packets on the incoming interface. For more information, see the Junos OS Administration Guide for Security Devices.

The following two data streams were sent on the multilink bundle:

100 data packets of 200 bytes (larger than the fragmentation threshold)
500 data packets of 60 bytes (smaller than the fragmentation threshold)

The following two voice streams were sent on the multilink bundle:

100 voice packets of 200 bytes from source port 100
300 voice packets of 200 bytes from source port 200

To confirm that LFI and load balancing are performed correctly, first verify that the link services interface is performing packet fragmentation as configured. Second, verify that the interface is encapsulating packets as configured. Finally, use the results to verify load balancing.

Note: Only the significant portions of command output are displayed and described in this example. For more information, see Verifying the Link Services intelligent queuing Interface Configuration.

Step 1: Verifying Packet Fragmentation

From the CLI, enter the show interfaces lsq-0/0/0 command, to check that large packets are fragmented correctly.

user@R0#> show interfaces lsq-0/0/0

Physical interface: lsq-0/0/0, Enabled, Physical link is Up
  Interface index: 136, SNMP ifIndex: 29
  Link-level type: LinkService, MTU: 1504
  Device flags   : Present Running
  Interface flags: Point-To-Point SNMP-Traps
  Last flapped   : 2006-08-01 10:45:13 PDT (2w0d 06:06 ago)
  Input rate     : 0 bps (0 pps)
  Output rate    : 0 bps (0 pps)

  Logical interface lsq-0/0/0.0 (Index 69) (SNMP ifIndex 42) 
    Flags: Point-To-Point SNMP-Traps 0x4000 Encapsulation: Multilink-PPP
    Bandwidth: 16mbps
    Statistics         Frames      fps       Bytes        bps
    Bundle:
      Fragments:
        Input :             0        0           0          0
        Output:          1100        0      118800          0
      Packets:
        Input :             0        0           0          0
        Output:          1000        0      112000          0
   ...
    Protocol inet, MTU: 1500
      Flags: None
      Addresses, Flags: Is-Preferred Is-Primary
        Destination: 9.9.9/24, Local: 9.9.9.10

What It Means—The output shows a summary of packets transiting the device on the multilink bundle. Verify the following information on the multilink bundle:

The total number of transiting packets = 1000
The total number of transiting fragments=1100
The number of data packets that were fragmented =100

The total number of packets sent (600 + 400) on the multilink bundle match the number of transiting packets (1000), indicating that no packets were dropped.

The number of transiting fragments exceeds the number of transiting packets by 100, indicating that 100 large data packets were correctly fragmented.

Corrective Action—If the packets are not fragmented correctly, check your fragmentation threshold configuration. Packets smaller than the specified fragmentation threshold are not fragmented. For information about configuring the fragmentation threshold, see Configuring the Link Services Interface with a Configuration Editor.

Step 2: Verifying Packet Encapsulation

To find out whether a packet is treated as an LFI or non-LFI packet, determine its encapsulation type. LFI packets are PPP encapsulated, and non-LFI packets are encapsulated with both PPP and MLPPP. PPP and MLPPP encapsulations have different overheads resulting in different-sized packets. You can compare packet sizes to determine the encapsulation type.

A small unfragmented data packet contains a PPP header and a single MLPPP header. In a large fragmented data packet, the first fragment contains a PPP header and an MLPPP header, but the consecutive fragments contain only an MLPPP header.

PPP and MLPPP encapsulations add the following number of bytes to a packet:

PPP encapsulation adds 7 bytes:
4 bytes of header+2 bytes of frame check sequence (FCS)+1 byte that is idle or contains a flag
MLPPP encapsulation adds between 6 and 8 bytes:
4 bytes of PPP header+2 to 4 bytes of multilink header

Figure 32 shows the overhead added to PPP and MLPPP headers.

Figure 32: PPP and MLPPP Headers

Image g016709.gif

For CRTP packets, the encapsulation overhead and packet size are even smaller than for an LFI packet. For more information, see Configuring CRTP.

Table 88 shows the encapsulation overhead for a data packet and a voice packet of 70 bytes each. After encapsulation, the size of the data packet is larger than the size of the voice packet.

Table 88: PPP and MLPPP Encapsulation Overhead

Packet Type	Encapsulation	Initial Packet Size	Encapsulation Overhead	Packet Size after Encapsulation
Voice packet (LFI)	PPP	70 bytes	4 + 2 + 1 = 7 bytes	77 bytes
Data fragment (non-LFI) with short sequence	MLPPP	70 bytes	4 + 2 + 1 + 4 + 2 = 13 bytes	83 bytes
Data fragment (non-LFI) with long sequence	MLPPP	70 bytes	4 + 2 + 1 + 4 + 4 = 15 bytes	85 bytes

From the CLI, enter the show interfaces queue command to display the size of transmitted packet on each queue. Divide the number of bytes transmitted by the number of packets to obtain the size of the packets and determine the encapsulation type.

Step 3: Verifying Load Balancing

From the CLI, enter the show interfaces queue command on the multilink bundle and its constituent links to confirm whether load balancing is performed accordingly on the packets.

user@R0> show interfaces queue lsq-0/0/0

Physical interface: lsq-0/0/0, Enabled, Physical link is Up
  Interface index: 136, SNMP ifIndex: 29
Forwarding classes: 8 supported, 8 in use
Egress queues: 8 supported, 8 in use
Queue: 0, Forwarding classes: DATA 
  Queued:
    Packets              :                600                  0 pps
    Bytes                :              44800                  0 bps
  Transmitted:
    Packets              :                600                  0 pps
    Bytes                :              44800                  0 bps
    Tail-dropped packets :                  0                  0 pps
    RED-dropped packets  :                  0                  0 pps
    …
Queue: 1, Forwarding classes: expedited-forwarding 
  Queued:
    Packets              :                  0                  0 pps
    Bytes                :                  0                  0 bps
  …
Queue: 2, Forwarding classes: VOICE 
  Queued:
    Packets              :                400                  0 pps
    Bytes                :              61344                  0 bps
  Transmitted:
    Packets              :                400                  0 pps
    Bytes                :              61344                  0 bps
    …
Queue: 3, Forwarding classes: NC 
  Queued:
    Packets              :                  0                  0 pps
    Bytes                :                  0                  0 bps
 …

user@R0> show interfaces queue se-1/0/0

Physical interface: se-1/0/0, Enabled, Physical link is Up
  Interface index: 141, SNMP ifIndex: 35
Forwarding classes: 8 supported, 8 in use
Egress queues: 8 supported, 8 in use
Queue: 0, Forwarding classes: DATA 
  Queued:
    Packets              :                350                 0 pps
    Bytes                :              24350                 0 bps
  Transmitted:
    Packets              :                350                 0 pps
    Bytes                :              24350                 0 bps
   ...
Queue: 1, Forwarding classes: expedited-forwarding 
  Queued:
    Packets              :                  0                 0 pps
    Bytes                :                  0                 0 bps
  …
Queue: 2, Forwarding classes: VOICE 
  Queued:
    Packets              :                100                 0 pps
    Bytes                :              15272                 0 bps
  Transmitted:
    Packets              :                100                 0 pps
    Bytes                :              15272                 0 bps
    …
Queue: 3, Forwarding classes: NC 
  Queued:
    Packets              :                 19                 0 pps
    Bytes                :                247                 0 bps
  Transmitted:
    Packets              :                 19                 0 pps
    Bytes                :                247                 0 bps
    …

user@R0> show interfaces queue se-1/0/1

Physical interface: se-1/0/1, Enabled, Physical link is Up
  Interface index: 142, SNMP ifIndex: 38
Forwarding classes: 8 supported, 8 in use
Egress queues: 8 supported, 8 in use
Queue: 0, Forwarding classes: DATA 
  Queued:
    Packets              :                350                 0 pps
    Bytes                :              24350                 0 bps
  Transmitted:
    Packets              :                350                 0 pps
    Bytes                :              24350                 0 bps
    …
Queue: 1, Forwarding classes: expedited-forwarding 
  Queued:
    Packets              :                  0                 0 pps
    Bytes                :                  0                 0 bps
  …
Queue: 2, Forwarding classes: VOICE 
  Queued:
    Packets              :                300                 0 pps
    Bytes                :              45672                 0 bps
  Transmitted:
    Packets              :                300                 0 pps
    Bytes                :              45672                 0 bps
    …
Queue: 3, Forwarding classes: NC 
  Queued:
    Packets              :                 18                 0 pps
    Bytes                :                234                 0 bps
  Transmitted:
    Packets              :                 18                 0 pps
    Bytes                :                234                 0 bps

What It Means—The output from these commands shows the packets transmitted and queued on each queue of the link services interface and its constituent links. Table 89 shows a summary of these values. (Because the number of transmitted packets equaled the number of queued packets on all the links, this table shows only the queued packets.)

Table 89: Number of Packets Transmitted on a Queue

Packets Queued	Bundle lsq-0/0/0.0	Constituent Link se-1/0/0	Constituent Link se-1/0/1	Explanation
Packets on Q0	600	350	350	The total number of packets transiting the constituent links (350+350 = 700) exceeded the number of packets queued (600) on the multilink bundle.
Packets on Q2	400	100	300	The total number of packets transiting the constituent links equaled the number of packets on the bundle.
Packets on Q3	0	19	18	The packets transiting Q3 of the constituent links are for keepalive messages exchanged between constituent links. Thus no packets were counted on Q3 of the bundle.

On the multilink bundle, verify the following:

The number of packets queued matches the number transmitted. If the numbers match, no packets were dropped. If more packets were queued than were transmitted, packets were dropped because the buffer was too small. The buffer size on the constituent links controls congestion at the output stage. To correct this problem, increase the buffer size on the constituent links. For more information, see Defining and Applying Scheduler Maps.
The number of packets transiting Q0 (600) matches the number of large and small data packets received (100+500) on the multilink bundle. If the numbers match, all data packets correctly transited Q0.
The number of packets transiting Q2 on the multilink bundle (400) matches the number of voice packets received on the multilink bundle. If the numbers match, all voice LFI packets correctly transited Q2.

On the constituent links, verify the following:

The total number of packets transiting Q0 (350+350) matches the number of data packets and data fragments (500+200). If the numbers match, all the data packets after fragmentation correctly transited Q0 of the constituent links.
Packets transited both constituent links, indicating that load balancing was correctly performed on non-LFI packets.
The total number of packets transiting Q2 (300+100) on constituent links matches the number of voice packets received (400) on the multilink bundle. If the numbers match, all voice LFI packets correctly transited Q2.
LFI packets from source port 100 transited se-1/0/0, and LFI packets from source port 200 transited se-1/0/1. Thus all LFI (Q2) packets were hashed based on the source port and correctly transited both constituent links.

Corrective Action—If the packets transited only one link, take the following steps to resolve the problem:

Determine whether the physical link is up (operational) or down (unavailable). An unavailable link indicates a problem with the PIM, interface port, or physical connection (link-layer errors). If the link is operational, move to the next step.
Verify that the classifiers are correctly defined for non-LFI packets. Make sure that non-LFI packets are not configured to be queued to Q2. All packets queued to Q2 are treated as LFI packets.
Verify that at least one of the following values is different in the LFI packets: source address, destination address, IP protocol, source port, or destination port. If the same values are configured for all LFI packets, the packets are all hashed to the same flow and transit the same link.

Why Are Packets Dropped on a PVC Between a J Series Device and Another Vendor?

Problem—I configured a permanent virtual circuit (PVC) between T1, E1, T3, or E3 interfaces on my Juniper Networks device and another vendor's device, and packets are being dropped and ping fails.

Solution—If the other vendor's device does not have the same FRF.12 support as the J Series device or supports FRF.12 in a different way, the J Series interface on the PVC might discard a fragmented packet containing FRF.12 headers and count it as a "Policed Discard." As a workaround for this problem, configure multilink bundles on both peers, and configure fragmentation thresholds on the multilink bundles.