ASR9000/XR: How to use Port Spanning or Port Mirroring

[xthuijs]

Introduction

This document provides some extra documentation and use cases on the use of port spanning or port mirroring.

You can monitor traffic passing in & out of a set of L2 or L3 Ethernet interfaces (including bundle-Ether).

Core Issue

ASR 9000 is the only platform implementing SPAN on XR (Only support on ethernet linecards, not on SIP-700.)

You can use SPAN/Mirror in the follow scenarios

- L2 & L3 interfaces.
- Local,  R-SPAN, and PW-SPAN only (no ER SPAN.)
- Scale limits:
    8 monitor sessions
    800 total source ports
    1.5 Gig bidirectional replication limit toward fabric for bundle interfaces and 10 Gig ports.
    Guideline:  ~ 10% - 15% total bandwidth can be mirrored system-wide
- Source ports:  Physical, EFPs, and bundles interfaces (L2 & L3)
- Destination ports:  Ethernet interfaces, EFPs, and PW-SPAN. (No bundle) [ only L2 transport interfaces are supported as destination ports]

- Ability to use ACL's to define which traffic is to be captured

- Capture multicast traffic is possible

Note: some of the functionality mentioned are enhancements to the XR 4.0.1 release, this document assumes you are using this release or later.

A good reference on the terminology of SPAN/Mirror can be found here:

http://www.cisco.com/en/US/docs/switches/lan/catalyst6500/ios/12.2SX/configuration/guide/span.pdf

SPAN order of operation

SPAN mirrors what is on the wire
For ingress, this means packets are mirrored before QOS, ACL, and encapsulation rewrite operations.
For egress, this means packets are mirrored after QOS, ACL, and encapsulation rewrite operations.

Partial Packet Mirroring

User can configure to mirror first 64 upto 256 bytes of the packet.
Note: The actual mirrored packet will be the configured size plus 4-byte trailling CRC.

Sample config:

interface GigabitEthernet0/6/0/20 l2transport
  monitor-session PW
  mirror first 100  <==  valid range: [64, 256], inclusively
  !
!

Note:  The mirrored packet received at sniffer will have the size of 104
               (4-byte of trailing CRC added by transmit MAC layer.)

ACL based Mirroring

“permit/deny” determines the behavior of the regular traffic (forwarded or dropped)
“capture” determines whether the packet is mirrored to the SPAN destination.

On SPAN: mirror traffic on the wire (regardless with or without ACL.)

      ACL on ingress direction:
           SPAN will mirror traffic even regular traffic dropped by ACL:  Always mirror!
     ACL on egress direction
          Will mirror if regular traffic is forwarded (Permit)
          Will not mirror if regular traffic is dropped (Deny.)

Inconsistent configurations:
“acl” is configured on SPAN source port but
   ACL has no “capture” keyword:
    No traffic gets mirrored. 
“acl” is NOT configured on SPAN source port but
   ACL has “capture” keyword:
    Mirroring traffic as normal, no ACL performed.

The ACL can also be an L2 ACL :

ethernet-services access-list esacl_t2
10 deny 1234.5678.90ab 0000.0000.0000 any capture

L3 Spanning Example

monitor-session TEST
destination interface GigabitEthernet0/1/0/2 (<<<< this is NP3)
!
interface GigabitEthernet0/1/0/14  (<<<< this is NP2)
ipv4 address 5.5.1.1 255.255.255.0
monitor-session TEST
  acl
!
load-interval 30
ipv4 access-group span ingress
!
ipv4 access-list span
10 permit ipv4 any host 1.1.1.10 capture
15 permit ipv4 any host 239.1.1.1 capture
20 permit ipv4 any host 2.2.2.100
30 permit ipv4 any any

Sample TRAFFIC GEN: (sending multicast in this example)
tgn rate 1000
L2-dest-addr 0100.5E01.0101
L2-src-addr 0003.A0FD.28A8
L3-src-addr 5.5.1.2
L3-dest-addr 239.1.1.1

Checking NP2: (the port that we are spanning)
Show global stats counters for NP2, revision v3

Read 12 non-zero NP counters:
Offset  Counter                                         FrameValue   Rate (pps)
-------------------------------------------------------------------------------
  22  PARSE_ENET_RECEIVE_CNT                                  5478        1001
  31  PARSE_INGRESS_DROP_CNT                                     3           1
  33  RESOLVE_INGRESS_DROP_CNT                                5474        1000
(there is no mcast recipient for this mcast addr, but we’re still replicating, see red line)
  40  PARSE_INGRESS_PUNT_CNT                                     1           0
  50  MODIFY_RX_SPAN_CNT                                      5475        1000
  54  MODIFY_FRAMES_PADDED_CNT                                5475        1000
  68  RESOLVE_INGRESS_L3_PUNT_CNT                                1           0
104  LOOP                                                       1           0
224  PUNT_STATISTICS                                            9           2
480  RESOLVE_IPM4_ING_RTE_DROP_CNT                           5475        1000
565  UIDB_TCAM_MISS_AGG_DROP                                    3           1
570  UIDB_TCAM_MISS_PORT4_DROP_FOR_HOST                         3           0

NP3 is the span monitor interface:
Show global stats counters for NP3, revision v3

Read 16 non-zero NP counters:
Offset  Counter                                         FrameValue   Rate (pps)
-------------------------------------------------------------------------------
  22  PARSE_ENET_RECEIVE_CNT                                    36           0
  23  PARSE_FABRIC_RECEIVE_CNT                               79656        1000
  30  MODIFY_ENET_TRANSMIT_CNT                               79655        1000

Packets received from fabric and sent off to the Ethernet on the span port!

PW SPAN example

For PW span to work, you need to define a local monitor session with a destination pseudo wire. You apply that span session to the interface of interest and define an xconnect group that also leverages that span session as one of the pw ends.

On the remote side where the PW terminates, you just configure regular VPWS.

Here an example:

On the Local Side, besides my Span configuration, there is also a local cross connect between the interested session we want to span over the PW

l2vpn

xconnect group TEST
  p2p TEST
   interface GigabitEthernet0/1/0/39

   ! port 39 is the port where we apply the span on.
   interface GigabitEthernet0/1/0/20.100
  ! this is just a random AC to have traffic flowing between the spanned port.
!

AC configuration:

interface GigabitEthernet0/1/0/20.100 l2transport
encapsulation dot1q 100
rewrite ingress tag pop 1 symmetric
! the tag is popped because the other XCON end is a plain ethernet without vlan. The explanation and use cases of tag popping can be found a related

! Tech note article.

Configuration on the remote side:

Regular VPWS configuration:

RP/0/RSP0/CPU0:A9K-TOP#sh run l2vpn
l2vpn
xconnect group PW-SPAN
  p2p PW-SPAN_1
   interface GigabitEthernet0/0/0/39
   neighbor 2.2.2.2 pw-id 1
   !
  !
!
interface GigabitEthernet0/0/0/39
load-interval 30
transceiver permit pid all
l2transport
!
!

the neighbor in the l2vpn configuration is the LDP neighbor ID
between which the PW is built.

Show on remote side:
RP/0/RSP0/CPU0:A9K-TOP#show l2vpn xcon group PW-SPAN det

Group PW-SPAN, XC PW-SPAN_1, state is up; Interworking none
  AC: GigabitEthernet0/0/0/39, state is up
    Type Ethernet
    MTU 1500; XC ID 0x4000a; interworking none
    Statistics:
      packets: received 0, sent 16570475
      bytes: received 0, sent 994228500

! packets received from the PW are sent out hte Attachment circuit's interface. The analyzer is connected to G0/0/0/39
  PW: neighbor 2.2.2.2, PW ID 1000, state is up ( established )
    PW class not set, XC ID 0x4000a
    Encapsulation MPLS, protocol LDP
    PW type Ethernet, control word disabled, interworking none
    PW backup disable delay 0 sec
    Sequencing not set

      MPLS         Local                          Remote
      ------------ ------------------------------ -----------------------------
      Label        16002                          16027
      Group ID     0xa40                          0x2
      Interface    GigabitEthernet0/0/0/39        PW/TM/MS
      MTU          1500                           1500
      Control word disabled                       disabled
      PW type      Ethernet                       Ethernet
      VCCV CV type 0x2                            0x2
                   (LSP ping verification)        (LSP ping verification)
      VCCV CC type 0x6                            0x6
                   (router alert label)           (router alert label)
                   (TTL expiry)                   (TTL expiry)
      ------------ ------------------------------ -----------------------------
    MIB cpwVcIndex: 4294705162
    Create time: 04/04/2011 14:36:42 (00:20:07 ago)
    Last time status changed: 04/04/2011 14:36:42 (00:20:07 ago)
    Statistics:
      packets: received 16570475, sent 0
      bytes: received 994228500, sent 0

! Packets received on the Pseudo Wire from the SPAN port

NOTE: Pseudo Wire counters on the span side are not incrementing.That is the XCON group "cisco" in this picture config example.

This is intentional. You can review the SPANNING also with this command:

RP/0/RSP1/CPU0:A9K-BOTTOM#sh monitor-session counters

Monitor-session PW_TM_MS
  GigabitEthernet0/1/0/39
    Rx replicated: 58488205 packets, 3743245120 octets
    Tx replicated: 58488206 packets, 3743245184 octets
    Non-replicated: 0 packets, 0 octets

R-SPAN configuration:

R-SPAN is natively support with the capability of ASR9000 to do vlan imposition:

monitor-session MS2

destination interface gig0/2/0/19.10

!

interface gig0/2/0/12.10 l2transport

encapsulation dot1q 10 <<< Monitoring vlan 10 traffic

monitor-session MS2

!

interface gig0/2/0/19.10 l2transport (*)

encapsulation dot1q 100 <<< VLAN 100 will get imposed.

!

(*) Monitor destination could be any supported destination interface regardless of monitor source

Related Information

n/a

Xander Thuijs, CCIE #6775

Sr. Tech Lead ASR9000

[Deniz AYDIN]

Thanks a lot for quick response. This may sound stupid question but  always confuse this calculation (for bandwidth:) so i want to clarify this. if I have traffic on the same port like 1Mpps in 1Mpps out, this requires 2Mpps on the NP?

[xthuijs]

Aha I see, here is the deal on that Deniz. If we classify a forwarding power in bps (bandwidth) then it omits one important piece of the puzzle. It is like giving the top speed of the car without giving you an indication of torque.

So the top speed of the car is 250Mph, but it requires 10 minutes, with the pedal floored to reach that speed and hill down. Torque matters here how fast it can pull to this speed.

Same with bandwidth (speed) and torque (pps).

Having 10G with 1000 byte packets would only require 1342177 pps

Having 10G at 64 byte packets would require a whole: 20971520 pps

so that is a substantial increase there in terms of pps.

the typhoon does 45Mpps per direction. so 90 total.

and if you have 1 in each direction then yes you'd be using 2 total and if everything is spanned then it is 4 total.

cheers!

xander

[Evan Roggenkamp]

Am I correct in the analysis that you can only monitor traffic going in ONE direction, either INGRESS or IN to the interface or EGRESS our OUT of the interface using ACL mirror?

[xthuijs]

Hi Evan,

ah I see why you'd think that, but this is just the example in my case. If I apply an ACL ingress and egress and both ACL's have the capture keyword in them I can span in both directions no problem. Obviously the traffic load will be higher on the span destination port (as more traffic gets spanned). I think for ease of troubleshooting, you'd want to do this per direction, otherwise your span port sees everything and it is hard to unravle the traffic from which direction it came from...

cheers!

xander

[slicerpro]

Hi, i have my monitor session set up thus:

Monitor-session monitor_alu_nrt3_epc
Destination interface GigabitEthernet0/7/0/37
================================================================================
Source Interface      Dir   Status
--------------------- ----  ----------------------------------------------------
Gi0/6/0/25.1408       Both  Operational
Gi0/6/0/7             Both  Operational
Gi0/6/0/9             Both  Operational
Gi0/6/0/17            Both  Operational
Gi0/6/0/19            Both  Operational
Gi0/6/0/21            Both  Operational
Gi0/6/0/22.1034       Both  Operational
Gi0/6/0/22.1408       Both  Operational
Gi0/6/0/23.1036       Both  Operational
Gi0/6/0/23.1416       Both  Operational
Gi0/6/0/23.1420       Both  Operational

I know that there is traffic going through Gi0/6/0/23.1420 because connectivity is up. However, i cannot capture anything. Any ideas...?

[fcotofaj]

Hi Xander,

Can you confirm that 1.5Gbps is the limit for the entire box or is it per bundle interface?

1.5 Gig bidirectional replication limit toward fabric for bundle interfaces and 10 Gig ports

Thanks,

Fede

[xthuijs]

Hi Fede,

it is really on a per NP to FIA link bases. So on the 40G Trident cards (40GE, 2T20G, 8T/4, 4T)

that is where you have a hard limit of that 1.5G on the card itself, regardless of where the span dest is. On the 80G cards, that is the 8T and the 16T/8 you have 2 FIA's so the 2 port groups served by each FIA can both do 1.5G each.

Realistically, trident is a bit dated now, so if you want to do something fancy with span, either use a bridge domain and disable mac learning to rely on the TM replication to the two ports. One of which is the "span" port. Or consider using the Typhoon (or Tomahawk) linecards.

cheers

xander

[fcotofaj]

Thank you Alexander, that means that on the 24x10G Typhoon which has 4 FIAs, we should be able to reach 6Gs (1.5G x4)?

Thanks,

Fede

[xthuijs]

hi fede,

the typhoon dont have that bw limitation. they are only bound by a pps limitation, so as far as your NPU can go, it can continue to replicate.

spanned pakcets are subject to a "recirculation" so when they enter the PARSE stage for a 2nd round, they "compete" with the fabric and wire packets. however span is the lowest priority. so if the NPU runs out of steam pps wise, span is the first that gets dropped, not wire received (ingress) or fabric received (egress).

cheers!

xander

[slicerpro]

I figured out that the expected traffic was actually not coming through. The customer was sending traffic through the standby router despite his repeated denials. After I temporarily shut down the interface on the standby router, the traffic started arriving and being captured.

[xthuijs]

I totally missed your note from months ago I see now, apologies!

but great to hear that you figured it out! great to read how span was able to prove a good technical point! :)

cheers

xander

[Evan Roggenkamp]

Hi Xander, 

I have another question for you. I am trying to monitor like this:

interface TenGigE0/6/0/6.1604
 mtu 9000
 ipv4 address 10.80.0.133 255.255.255.252
 monitor-session DDOS direction rx-only
 !
 encapsulation dot1q 1604

It seems when I try to do this on the parent interface, I get nothing...

I have tried using interface TenGigE0/5/0/0 with or without l2transport as the mirror destination. 

It appears operational:

Monitor-session DDOS
Destination interface TenGigE0/5/0/0
================================================================================
Source Interface      Dir   Status
--------------------- ----  ----------------------------------------------------
Te0/6/0/6.1604        Rx    Operational

But there should be over 4Gbps of traffic on this port RX, whilst the monitor destination appears to note around 30Mbps. 

I am not sure if this is something I am doing wrong or something I am just missing. Thanks!

[xthuijs]

hi evan, the monitor destination must be an l2transport, otherwise egress (l2) headers on the span port may get messed up.

not seeing all traffic can be because:

- replication limitation due to the monitored interface's npu is running out of steam.

- incorrect application of span, eg applied to main interface not seeing subinterface traffic (in that case span needs to be applied to those interfaces dedicated you want to monitor for)

- bw limitation on the replication point.

you could check show controller np counter np <of spanned interface> and see if the packet rate of ENET_RECEIVE is showing similar counts for MDF_SPAN

you could check show controller np tm counters to see if the traffic manager is hitting a bw limitation for the span operation.

Or if there are any early fast discards on the npu counters that signify that the npu is running out of steam.

cheers

xander

[Evan Roggenkamp]

Xander

TAC helped me figure out the rest. It would appear with our linecards we have hit a limit. Thanks for the help.

[xthuijs]

Ok thanks for the update Evan!

xander