Documente Academic
Documente Profesional
Documente Cultură
architectures
BRKSPG-2904
Agenda
Introduction
Architecture of SW and HW forwarding
Fabric
NPU
Local Packet Transport Services/LPTS
Software forwarding/handling
BRKSPG-2904
Cisco Public
Acknowledgements
With contributions from
YOU
Thanks for all the feedback on the support forums and last years CL!
BRKSPG-2904
Cisco Public
Introduction
Hardware and Software Forwarding architectures
Getting started
With packet forwarding troubleshooting
One common architecture between all platforms
IOS XRv
XR virtualization
nV Satellite
9000v,901,903
BRKSPG-2904
9001
9001-S
Cisco Public
9904
6
9006
9010
9912
9922
Forwarding paths
In principal there are 2 forwarding handlers
Hardware based forwarding from Linecard to Linecard over the fabric
Software based
This is not truly forwarding as XR has no true software forwarding path
This is for local packet consumption, eg for me packets
These are either handled by
LC CPU
RP CPU
BRKSPG-2904
Cisco Public
CPU
Software forwarding
Calls a receive interrupt (RX interrupt) to the CPU
to handle the packet
Interfaces
PA
PA
PA
process
BGP
OSPF
jump
Jump procs
Cisco Public
Interupt
handler
Hardware Forwarding
Transit
Ingress PHY receives frame
Central fabric
If RSP2/RSP440 this is on the RSP
If 9922/9910 this is the separate fabric cards
Egress LC fabric and FIA hand over packet to egress NPU for egress feature
processing
Egress NPU hands over to PHY and transmit
BRKSPG-2904
Cisco Public
Software forwarding
The punt path
If ingress NPU detects this packet is for me.
How does it determine it is for me?
The iFIB (internal FIB) portion of the LPTS (local packet transport services)
After the CPU receives the packet after an RX interrupt the packet is handled by
SPP, and potentially the
NETIO chain
BRKSPG-2904
Cisco Public
10
iFIB
Policing
MEM
CPU
SDD
3x 10G
3x10GE
SFP +
Punt
FPGA
Typhoon
3x 10G
3x10GE
SFP +
Typhoon
CPU
FIA
3x 10G
Typhoon
3x 10G
3x10GE
SFP +
Typhoon
FIA
3x 10G
3x10GE
SFP +
Typhoon
FIA
3x 10G
3x10GE
SFP +
Typhoon
Switch Fabric
ASIC
3x10GE
SFP +
Switch
Fabric
USB
Disk0/1
FIA
I/O FPGA
Typhoon
NVRAM
Boot Flash
3x 10G
3x10GE
SFP +
Typhoon
3x10GE
SFP +
BRKSPG-2904
CPU Complex
FIA
3x 10G
Typhoon
Cisco Public
11
Trace-points
Typhoon
FIA
3x 10G
3x10GE
SFP +
Typhoon
FIA
3x 10G
Typhoon
Typhoon
FIA
3x 10G
3x10GE
SFP +
Typhoon
FIA
3x 10G
3x10GE
SFP +
Typhoon
Switch
Fabric
Switch Fabric
ASIC
3x 10G
3x10GE
SFP +
100GE
MAC/PHY
Switch Fabric
ASIC
3x10GE
SFP +
Ingress 100G
Typhoon
FIA
Egress 100G
Typhoon
FIA
Ingress
Typhoon 100G
100GE
MAC/PHY
3x 10G
3x10GE
SFP +
Typhoon
FIA
3x 10G
3x10GE
SFP +
BRKSPG-2904
Switch
Fabric
Typhoon
Cisco Public
13
FIA
Egress
Typhoon 100G
Physically separated from LC. Resides on RSP or dedicated card (9912, 9922)
Logically separated from LC and RSP
All fabric ASICs run in active mode regardless of RSP Redundancy status
Extra fabric bandwidth and instant fabric switch over
If the FAB has been previously initiated then even with RP in rommon FABRIC IS
ACTIVE!
BRKSPG-2904
Cisco Public
14
Fabric Arbitration
RSP0
Crossbar
Fabric
ASIC
1: Fabric Request
5: credit return
Crossbar
Fabric
ASIC
Arbitration
Fabric Interface
and VOQ
2: Arbitration
Crossbar
Fabric
ASIC
3: Fabric Grant
Crossbar
Fabric
ASIC
4: load-balanced
transmission across
fabric links
Arbitration
RSP1
BRKSPG-2904
Cisco Public
15
Fabric Interface
and VOQ
RSP0
Fabric Interface
and VOQ
Crossbar
Fabric
ASIC
Crossbar
Fabric
ASIC
Arbitration
RSP1
Unicast traffic sent across first available fabric link to destination (maximizes efficiency)
Cisco Public
16
RSP0
Crossbar
Fabric
ASIC
Arbitration
RSP1
Cisco Public
C1
B2
A3
B1
A2
A1
Crossbar
Fabric
ASIC
Fabric Interface
and VOQ
17
Packet 2
Packet 1
Max
Super-frame
Min
Super-frame
Packet 1
No super-framing
Packet 1
Min reached
Packet 1
Max reached
Jumbo
0 (Empty)
Note that fabric counters are showing super frames not individual packets!!
(show controller fabric fia stats loc 0/X/CPU0)
BRKSPG-2904
Cisco Public
18
BRKSPG-2904
Cisco Public
19
9922,
how many RSPs does this system
have?
BRKSPG-2904
Cisco Public
20
The FIA ingress drop stats offsets (0-3) represent XBAR link stats and correspond
to:
BRKSPG-2904
Cisco Public
21
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
8
0
2
3
0
3
0
2
3
0
Cisco Public
22
Priority
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
**********
BRKSPG-2904 FIA-1 **********
2014 Cisco and/or its affiliates. All rights reserved.
Uc dq pkt-len-crc/RO-seq/len error
Mc rf crc
Mc vl0 src0 buffer full
Mc vl1 src0 buffer full
Mc vl2 src0 buffer full
Mc vl3 src0 buffer full
Mc vl0 src1 buffer full
Mc vl1 src1 buffer full
Mc vl2 src1 buffer full
Mc vl3 src1 buffer full
23
drp
drp
drp
drp
drp
drp
drp
drp
drp
drp
0
0
0
0
0
0
0
0
0
0
: 12782
: 9778
Cisco Public
24
Parse
L2/L3 header
packet parsing in
TCAM
Builds keys for
ingress ACL,
QoS and
forwarding
lookups (uCode)
BRKSPG-2904
Search
Resolve
Modify
Performs QoS
and ACL
lookups in
TCAM tables
Performs L2
and L3 lookups
in RLDRAM
Processes Search
results:
ACL filtering
Ingress QoS
classification and
policing
Forwarding (egress
SFP determined)
Performs L2 MAC
learning
Adds internal
system headers
Egress Control
Header (ECH)
Switch Fabric
Header (SFH)
Cisco Public
26
Queueing
Scheduling
Queuing,
Shaping and
Scheduling
functions
All packets go
through this
stage
Troubleshooting NP Forwarding
1. Identify interface in question with problem.
2. Identify the mapping from interface to NPU.
3. Examine NP counters.
4. Look for rate counters that match lost traffic rate.
If none of the counters match the expect traffic, check drops at interface
controller
Cisco Public
27
Interface to NP mapping
RP/0/RSP0/CPU0:A9K-BNG#show controller np ports all loc 0/0/cpU0
Node: 0/0/CPU0:
---------------------------------------------------------------NP
-0
1
2
3
Bridge
----------
BRKSPG-2904
Fia
--0
1
2
3
Ports
--------------------------------------------------GigabitEthernet0/0/0/0 - GigabitEthernet0/0/0/9
GigabitEthernet0/0/0/10 - GigabitEthernet0/0/0/19
TenGigE0/0/1/0
TenGigE0/0/1/1
Cisco Public
28
Examine NP Counters
RP/0/RSP0/CPU0:A9K-2#show controller np counters np0 loc 0/0/CPU0
Node: 0/0/CPU0:
---------------------------------------------------------------Show global stats counters for NP0, revision v2
Read 57 non-zero NP counters:
Offset Counter
FrameValue Rate (pps)
------------------------------------------------------------------------------16 MDF_TX_LC_CPU
22755787
6
17 MDF_TX_WIRE
1614696
0
21 MDF_TX_FABRIC
1530106
0
29 PARSE_FAB_RECEIVE_CNT
1555034
0
33 PARSE_INTR_RECEIVE_CNT
22026578
6
37 PARSE_INJ_RECEIVE_CNT
335774
0
41 PARSE_ENET_RECEIVE_CNT
2115361
1
45 PARSE_TM_LOOP_RECEIVE_CNT
17539300
5
Packets received
MDF=Modify
TX transmit
WIRE to the wire
= egress
If delta between received from Fab to TX-wire is almost 0, then everything forwarded
and not punted. If not, we dropped packets, could be ACL, QOS, or for other
reasons (eg PUNT) or large portion of injected traffic.
BRKSPG-2904
Cisco Public
29
NP Counter Description
RP/0/RSP0/CPU0:A9K-2#sh controllers np descriptions location 0/0/CPU0
Counter
------------------------------------------------------------------------------<...>
PARSE_DROP_IPV4_LENGTH_ERROR
Drop L3
IPv4 packets dropped on receipt
from Ethernet due to IP packet length field inconsistent with L2 frame length
<...>
Look up controller np counters via google OR they can be seen in
show controller np description
BRKSPG-2904
Cisco Public
30
= 985886125
= 985886125
= 8343919
= 6957281
= 1386636
=2
Cisco Public
31
There is also a specific counter from which port index these drops came from:
UIDB_TCAM_MISS_DROP_1
This means that the second port (starting count from zero) on that NPU
experienced that drop.
So if your show controller np ports tells us that ports X Y and Z are connected to
this NPU, and the drop index is _1, then port Y is the culprit.
BRKSPG-2904
Cisco Public
32
You can monitor any counter in the NPU on Typhoon generation line cards
Captured packets are always dropped
Exists automatically after capturing <N> packets or when timeout is reached
NPU is reset upon exit (~50ms forwarding stop)
This will be enhanced later
Cisco Public
33
Cisco Public
34
STOP
What is LPTS?
Local packet transport services handling 4 key functions:
DIRECT
POLICE
CATEGORIZE
RP
Bgp-known
Packet from configured neighbor
MPP denied packet to control
Plane, eg telnet from not
bgp-cfg
known station
LC
BRKSPG-2904
Cisco Public
36
LPTS characteristics
Classification and Policing inside NPU
An LPTS flow is policed at a certain rate
If you have eg multiple BGP neighbors on this NPU they share the same police
rate
For instance:
All our established neighbors are categorized as bgp-known
The bgp-known flow is policed at 10,000 pps
Police rates can be adjusted on a per LC basis, or globally applicable to all LCs.
Each NPU has that configured police rate (rate not shared between all npus on LC)
BRKSPG-2904
Cisco Public
37
CPU
CPU
CLI
TCAM
3x 10G
3x10GE
SFP +
Typhoon
Neighbor 1.2.3.4
FIA
3x 10G
3x10GE
SFP +
Typhoon
Typhoon
FIA
3x 10G
3x10GE
SFP +
Typhoon
FIA
3x 10G
3x10GE
SFP +
Typhoon
TCP
3x 10G
3x10GE
SFP +
Socket
Switch Fabric
ASIC
Typhoon
Established TCP
Creates TCP
3x 10G
3x10GE
SFP +
BGP
LPTS
3x 10G
3x10GE
SFP +
Typhoon
3x10GE
SFP +
BRKSPG-2904
CPU Complex
FIA
3x 10G
Typhoon
Cisco Public
38
LPTS
packets in
App 1
good packets
for-us packets
LPTS Internal FIB
(IFIB)
FIB
DCoPP
Dynamic Control
Plane Policing
bad packets
BRKSPG-2904
App 2
RP
LC
transit
packets out
RP
Cisco Public
39
Local
Stacks
LC
Router bgp
neighbor 202.4.48.99
ttl_security
! mpls ldp
port
Rate
ICMP
ANY
ANY
1000
Priority
low
any
179
any
any
100
medium
any
179
202.4.48.99
any
1000
medium
202.4.48.1
179
202.4.48.99
2223
10000
medium
200.200.0.2
13232
200.200.0.1
646
100
medium
ttl
255
Socket
port
Any
LPTS
Local
BGP
LDP
SSH
Cisco Public
40
Verification of filtering
Example, config used:
ntp
server 3.0.0.1
!
LPTS entries:
RP/0/RSP0/CPU0:A9K-BNG#show lpts pifib hardware entry brief location 0/0/cPU0 | i 123
Wed Apr 2 07:54:53.996 EDT
32 IPV4 default
UDP any
LU(30) any,123 3.0.0.1,any
<< yes we like that
204 IPV4 default
UDP any
LU(30) any,123 any,any
<< crap this we need to close of!
BRKSPG-2904
Cisco Public
43
Verification of policing
RP/0/RSP0/CPU0:A9K-BNG#show lpts pifib hardware entry loc 0/0/cPU0 | be 3.0.0.1
Cisco Public
44
Flow
FLOW NAME
curr.rate
| i NTP
accepted
dropped
default rate
NTPdefault
126
NTP-known
Static
180
200
Static
200
200
200
01234567
Configuration
lpts pifib hardware police
flow ntp default rate 1
This adjusts the policer rate to 1pps. (value of 0 is to be fixed up) Explanation of burst!
BRKSPG-2904
Cisco Public
45
01234567
Memory
CPU
ICMP BGP
CDP
0
1
2
1
2
FTP
HDLC
0
1
3
4
5
raw UDP
TCP
0
1
6
7
8
9
10
11
12
BFD
Netflow
NetIO
CpuCntrl Queues
From high/medium/critical etc
13
14
15
16
SPP
BRKSPG-2904
Cisco Public
47
# spp_ui
spp-ui>
spp-ui> trace filter node client/punt
Node "client/punt" set for trace filtering. Index: 11
Decoder https://scripts.cisco.com/ui/use/xr_spp_ui_to_pcap
BRKSPG-2904
Cisco Public
48
Static
RSVP-TE
BGP
ISIS
LSD
OSPF
EIGRP
RIB
RSP CPU
RSP
LC
ARP/NDP
HW FIB
SW FIB
Adjacency
LC NP
AIB
AIB: Adjacency Information Base
RIB: Routing Information Base
FIB: Forwarding Information Base
LSD: Label Switch Database
LC CPU
BRKSPG-2904
Cisco Public
50
Egress LC
ARP/NDP
SW FIB
SW FIB
AIB
LC CPU
Recursions
resolved
Packet
HW FIB
LC CPU
Packet
HW FIB
Adjacency
LC NP
LC NP
Ingress Lookup:
Find egress NP +
egress interface ID
BRKSPG-2904
Egress Lookup:
Find output interface
+ do L2 rewrite
Cisco Public
51
Packet
Adjacency
stored only for
local and virtual
interfaces
L3 NP FIB Architecture
Adjacency
pointer(s) 1xLDI
Up to 32 ways
Non-Recursive
Prefix Leaf
NR NR
LDI LDI
NR
LDI
Up to 8 or 32 ways
Recursive
Prefix Leaf
R R R R
LDI LDI LDI LDI
Cisco Public
OIF
Adj
52
Adj
LAG (64
members)
OIF
Protected TE
Adj
LAG
OIF
Backup TE
Adj
OIF
L3
L4
payload
CRC32
32 bits
HASH
8 bits selected
8 bits selected
(3 drawn) 256 buckets
Path (ECMP)
path2
path1
BRKSPG-2904
path2
Member (LAG)
path2
path1
path1
Cisco Public
53
(1) L2 access bundle uses access S/D-MAC + RID, OR L3 if configured (under l2vpn)
(2) L2 access AC to PW over mpls enabled core facing bundle uses PW label (not FAT-PW label even if configured)
-
BRKSPG-2904
Cisco Public
54
Load-balancing scenarios
45 for ipv4
BRKSPG-2904
Cisco Public
55
This means that if you have a P router that has no knowledge about the MPLS service of the packet,
that nibble can either mean the IP version (in MPLS/IP) or it can be the DMAC (in EoMPLS).
RULE: If you have EoMPLS services AND macs are starting with a 4 or 6. You HAVE to use ControlWord
L2
MPLS
MPLS
45 (ipv4)
0000 (CW)
41-22-33 (mac)
4111.0000.
Control Word inserts additional zeros after the inner label showing the P nodes to go for label based
balancing.
In EoMPLS, the inner label is VC label. So LB per VC then. More granular spread for EoMPLS can be achieved with
FAT PW (label based on FLOW inserted by the PE device who owns the service).
Take note of the knob to change the code: PW label code 0x11 (17 dec, as per draft specification). (IANA assignment is 0x17)
BRKSPG-2904
Cisco Public
56
Adj nodes compute the same hash, with little variety if the RID is close
This can result in north bound or south bound routing.
Hash shift makes the nodes look at complete different bits and provide more spread.
Trial and error (4 way shift trident, 32 way typhoon, values of >5 on trident result in modulo)
BRKSPG-2904
Cisco Public
57
L3
L4
payload
Hash shift 8
Y HASH
HASH
8 bits selected
8 bits selected
(3 drawn) 256 buckets
Path (ECMP)
path2
path1
BRKSPG-2904
path2
Member (LAG)
path2
path1
path1
Cisco Public
58
BRKSPG-2904
Cisco Public
59
router static
address-family ipv4 unicast
10.1.1.1/32 TenGigE0/1/1/0
TenGigE0/1/1/1 10.2.2.1
SW FIB
AIB
LC CPU
TenGigE0/1/1/0
10.2.2.2/24
LC CPU
HW FIB
HW FIB
TenGigE0/0/0/0
10.1.1.2/24
Adjacency
TenGigE0/1/1/1
BRKSPG-2904
LC NP
Cisco Public
LC NP
60
TenGigE0/0/0/1
Local
exists
(arp learnt 10.1.1.1 on te 0/0/0/0
Localadjacency
adjacency
exists
router static
address-family ipv4 unicast
10.1.1.1/32 TenGigE0/1/1/0
TenGigE0/1/1/1 10.2.2.1
SW FIB
AIB
LC CPU
TenGigE0/1/1/0
10.2.2.2/24
LC CPU
HW FIB
HW FIB
TenGigE0/0/0/0
10.1.1.2/24
Adjacency
Packet
BRKSPG-2904
TenGigE0/1/1/1
LC NP
Switch
Fabric
Packet w/ DestIP
10.1.1.1
61
Cisco Public
LC NP
TenGigE0/0/0/1
Packet
BRKSPG-2904
Cisco Public
62
BRKSPG-2904
Cisco Public
63
Troubleshooting L3 Forwarding
Identify the input interface, slot and NP
Walk the SW+HW control plane to confirm the expected forwarding decision
RIB
SW FIB on ingress LC
Ingress HW FIB on ingress LC
Egress HW FIB on egress LC
BRKSPG-2904
Cisco Public
64
BRKSPG-2904
Cisco Public
65
BRKSPG-2904
Cisco Public
66
BRKSPG-2904
Cisco Public
67
BRKSPG-2904
Cisco Public
68
Block
Description
SW FIB Entry
HW Leaf
NR-LDI
Always included
Displays the egress NP and egress interface ID
Cisco Public
69
Cisco Public
70
BRKSPG-2904
Cisco Public
71
BRKSPG-2904
Cisco Public
72
BRKSPG-2904
Cisco Public
73
BRKSPG-2904
Cisco Public
75
RP
Source
(S, G) Register
(S, G) Joins
Shared Tree
Source Tree
(unicast)
BRKSPG-2904
Cisco Public
76
Non-Default ssm:
50.0.0.10
239.x.x.x
RP
A
CPU
Receiver 1
BRKSPG-2904
Cisco Public
77
Out-of-band
source directory,
example: web server
Workaround
Source Mapping
Router maps IGMPv2 Joins in SSM range to well-known sources via DNS or static
configuration
BRKSPG-2904
Cisco Public
78
PHY
NP0
PHY
NP1
PHY
NP2
PHY
NP3
4 MFIB
5
B0
CP
U
PI
M
2
MRIB
FIA
B1
CrossbarCrossbar
Switch
Fabric Fabric
ASIC
ASIC
Fabric
2.
3.
MRIB sends the multicast state information to MFIB process on all LCs
4.
MFIB program HW forwarding tables in NP, Bridge FPGA and Fabric interface
ASIC
5.
Software switched multicast packet or data packet for protocol signaling is sent
to local line card CPU
2014 Cisco and/or its affiliates. All rights reserved.
Cisco Public
79
CPU
BRKSPG-2904
IGMP
MFIB
CPU
MGID
IGMP joins
PHY
PHY
IGMP joins
Multicast
Source
NP1
PHY
NP2
PHY
NP3
FIA
B0
FIA
FGID/
FPOE
Switch
Fabric
B1
BRKSPG-2904
LC3
Cisco Public
B03
NP1
PHY
B1
NP2
PHY
NP3
PHY
NP0
PHY
B0
NP1
PHY
B1
NP2
PHY
NP3
PHY
CPU
PHY
LC2
FIA
IGMP joins
NP04
MGID
CPU
NP0
BRKSPG-2904
Cisco Public
FGID (Slotmask)
FGIDs: 10 Slot Chassis
LC 0
LC 1
LC 2
LC 3
RSP 1
RSP 0
LC 4
LC 5
LC 6
LC 7
Logical
Slot
Logical
Slot
LC 3
LC 2
LC 1
LC 0
RSP 1
RSP 0
Slot
Slot
Logical
LC7
Slot Mask
Physical
9
Binary
1000000000
Hex
0x0200
LC6
0100000000
0x0100
LC5
0010000000
0x0080
LC4
0001000000
0x0040
RSP0
0000100000
0x0020
RSP1
0000010000
0x0010
LC3
0000001000
0x0008
LC2
0000000100
0x0004
0000000010
LC1
LC0
BRKSPG-2904
0 Cisco Systems,0000000001
2006
Inc. All rights reserved.
0x0002
Cisco0x0001
Confidential
Cisco Public
Slot Mask
Logical
Physical
Binary
Hex
LC3
0000100000
0x0020
LC2
0000010000
0x0010
LC1
0000001000
0x0008
LC0
0000000100
0x0004
RSP1
0000000010
0x0002
FGID Calculation
0
0000000001
Examples
0x0001
RSP0
Target Linecards
LC6
0x0100
LC1 + LC5
MGID Tables
MGID Bitmasks
MGID
FIA
MGID
Bit 1
Bit 0
Bridge1
NP3
BRKSPG-2904
MGID
Bridge0
NP2
Cisco Public
NP1
83
Bit 1
Bit 1
NP0
Bit 0
Bit 0
SSM-MAP
Provides the mapping for non v3 capable receivers to map the mcast group to a specific
source (whatever source address that is).
DNS mapping: allow a name instead of an address for the dynamic mapping of groups
to sources
BRKSPG-2904
Cisco Public
84
Cisco Public
igmp v2 Receiver
0/1/0/1.1501
ASR9K-1
(51.2.1.2, 239.1.1.10)
Source Gig0/0/0/4.300
SSM MAP
If not configured mcast groups outside default range are perceived to be v2 and forwarded to the
Route Processor and might get dropped (if these are actually SSM mcast adds) or sent to the
Rendez-Vous point
BRKSPG-2904
Cisco Public
In the presence of receivers, last hop router builds (S,G) towards source.
Due to (S,G) from last-hop router, traffic get forwarded correctly.
This scenario is applicable when first hop router is enabled for both PIM-SM and
PIM-SSM.
If PIM-SM RP is not configured and if there are no receivers, packets gets dropped
in NP; Example output below:
P/0/RSP0/CPU0:ASR9K-1#show controllers np counters np0 location 0/1/CPU0
41 PARSE_ENET_RECEIVE_CNT
81743
781
332 RSV_DROP_IPM4_ING_RTE_DROP
80396
777
Seen when there is no RendezVous Point configured and there is no registered v3 receiver
BRKSPG-2904
Cisco Public
(52.0.0.2,239.1.1.2) RPF nbr: 52.0.0.2 Flags: Group is not part of default SSM group
Up: 00:00:15
Incoming Interface List
GigabitEthernet0/1/0/15 Flags: A, Up: 00:00:15
RP/0/RSP0/CPU0:EAST-PE-ASR9K-1#
Seen when RP is configured, we are sending the traffic to the fabric; forwarding decision is made before we know the
receivers in the NP. The fabric receives erroneous traffic in this case which will get dropped.
BRKSPG-2904
Cisco Public
In the absence of SSM range config, packets get punted to RP with FGID Null.
With FGID null, packets get dropped in Crossbar: [Port NP FIA Fabric Crossbar]
RP/0/RSP0/CPU0:ASR9K-1#show mrib route detail
(51.2.1.2,239.1.1.10) Ver: 0x659d RPF nbr: 51.2.1.2 Flags:,
FGID zero
MGID: 16918
Up: 00:25:27
Incoming Interface List
GigabitEthernet0/0/0/4.300 Flags: A NS, Up: 00:01:46 Source
RP/0/RSP0/CPU0:ASR9K-1#show controllers fabric crossbar statistics instance 0 loc 0/rsp0/cpu0
Port statistics for xbar:0 port:12
Low priority stats (multicast)
===========================
NULL FPOE Drop Count
: 71740 = Packets get dropped in RSP cross-bar
==============================
RP/0/RSP0/CPU0:ASR9K-1#
PIM periodically, sends null registers to PIM-RP to check if there are new receivers but data packets
are not really sent to RP. (few packets leaked from hw to CPU for this purpose)
RP/0/RSP0/CPU0:EAST-PE-ASR9K-1#show pim traffic
PIM Traffic Counters
Elapsed time since counters cleared: 01:03:25
Received
Null Register
0
Register Stop
69
BRKSPG-2904
Cisco Public
Sent
67
0
Sent to RP
Received from RP
This is the situation where you want to be, dropped in HW, no punting. If there is a receiver
(either
v2 via ssm-map
or a direct v3 then hw Cisco
gets
programmed to forward)
BRKSPG-2904
2014 Cisco and/or its affiliates. All rights reserved.
Public
Best-Practice
(51.2.1.2, 239.1.1.10)
Source Gig0/0/0/4.300
Cisco Public
multicast-routing
address-family ipv4
ssm range SSM-groups
interface all enable
RP/0/RSP0/CPU0:ASR9K-1#show run ipv4 access-list
SSM-group
ipv4 access-list SSM-groups
10 permit ipv4 239.0.0.0 0.255.255.255 any
20 permit ipv4 232.0.0.0 0.255.255.255 any
RP/0/RSP0/CPU0:ASR9K-1#
Satellite
Satellite Basics
Satellite Access port
ASR9k
(Host)
Satellite
Ingress: ASR9K uses the SAT-VLAN id and the
incoming uplink to determine the satellite port. Rest of
the packet processing is identical to packets received at
local ports
Ingress
MAC DA
MAC SA
VLANs(opt)
Payload/FCS
MAC DA
MAC SA SAT-VLAN
VLANs(opt)
Payload/FCS
Egress
Egress: Satellite box maps the incoming SAT-VLAN to
its front port, strips the SAT-VLAN and forwards the
packet.
BRKSPG-2904
Cisco Public
Satellite Bringup
Discovery Protocol
Operates at Layer 2.
Provides the bootstrap mechanism by which a Satellite and Host begin communication.
Enables Host to become aware of a reachable Satellite device, and exchange sufficient
information to set up a full Control session.
The Host device initiates discovery. Satellite devices are factory-shipped to listen for
incoming Discovery probe packets.
Control Protocol
Makes use of the connectivity set up by the discovery protocol to provide a reliable and
extensible mechanism for the Host device and each satellite device to exchange the
required configuration, state, etc.
BRKSPG-2904
Cisco Public
94
Discovery Protocol
Discovery Protocol states
Stopped The link is down. Unshut the satellite Fabric link or the member links incase
Bundle is used as ICL.
Probing for satellites / Discovering The Discovery protocol has sent probe packets
to the satellite. It is waiting for a response from the satellite.
Configuring satellite The Discovery protocol has sent configuring packets to the
satellite. It is waiting for a response from the satellite.
Ready Discovery protocol has completed. Control protocol bring-up can now start.
Cisco Public
95
Host A
Satellite
Host A
Satellite
VLAN-B
Satellite
Host B
CFM
Satellite
fabric extension
ICCP
Host B
Satellite
Host
Cascading
BRKSPG-2904
Cisco Public
96
Please use following steps to upgrade the image on satellite incase the image on
satellite is not latest: (install nv satellite)
BRKSPG-2904
Cisco Public
97
BRKSPG-2904
Cisco Public
98
Contd.
If Host is not sending the discovery probes or if the host is dropping the discovery
replies:
Check show int <ICL> output on host, input/output counter would show SDAC discovery
input/output packets.
Check the NP drop counters of the ICL port.
If IN_SATELLITE_DISCOVERY_DISCARD NP counter is increasing:
Check if icl bit is programmed in uidb for icl interface
show uidb data location 0/0/CPU0 tenGigE 0/0/1/0 ingress
Satellite IC interface
0x1 <<<
If the ICL bit is not set in UIDB, we need to check the vlan ea db
sh ethernet infra internal ea trunks tenGigE 0/14/0/3 location 0/14/cPU0 --> main interfaces
sh ethernet infra internal ea subs tenGigE 0/14/0/3.1 location 0/14/cPU0 --> for icl which is a sub interface
(incase of L2 Fab)
is_in_icl_mode: 1
If the bit is not set in vlan ea then we need to check the vlan ma
sh ethernet infra internal ether-ma trunks tenGigE 0/14/0/3 location 0/14/cPU0 (use subs for sub interfaces)
is_in_icl_mode: 1
BRKSPG-2904
Cisco Public
99
BRKSPG-2904
Cisco Public
100
BRKSPG-2904
Cisco Public
101
Host A
E-ICCP
Satellite
Host B
BRKSPG-2904
Cisco Public
102
Check if TCP session is UP. show tcp brief should show the state as ESTAB
BRKSPG-2904
Cisco Public
103
BRKSPG-2904
Cisco Public
104
Check if the router has tftp homedir configured on the ASR9000 host.
tftp vrf default ipv4 server homedir disk0:
If the tftp transfer requests from satellite comes on the default vrf [through manual IP configuration] and tftp
home directory configured on the host is disk0: then image transfer request will fail as the tftp_fs will try to
read the disk0:/ path.
Please remove the tftp homedir and retry the transfer.
BRKSPG-2904
Cisco Public
105
default
v4
tftp
udp
69
unlimited
wait
Option
<<<<<<<<<<<<<<<< ICPE
Cisco Public
106
Cisco Public
107
If ARP is not resolved on the host check if the ARP is resolved on the other end
connected to satellite. If ARP is resolved on the other end, send ping packets with
size 1400 and trace the packet path same as above on the opposite direction.
If ARP is not resolved on either end, then initiate ping from the host and host will send
ARP requests, check the qos tm counters to see if packets go out of host.
Configure static arp on the host and initiate ping, now the host will send normal ping
packets(not arp packets), check the counters on the ICL link and the satellite as
above.
BRKSPG-2904
Cisco Public
108
Troubleshooting QoS
PHY
NP0
PHY
NP1
PHY
NP2
PHY
NP31
Egress side of LC
CPU
Ingress (sub-)interface
QoS Queues
FIA
FIA
Switch
Fabric
Virtual
Output
Queues
Configure with
Ingress MQC 4-layer hierarchy
Two strict high priority + Normal priority
BRKSPG-2904
NP0
CPU
Implicit Configuration
Two strict high priority +
Normal priority
Cisco Public
110
Egress FIA
Queues
PHY
NP1
PHY
NP2
PHY
NP3
PHY
Egress (sub-)interface
QoS Queues
Crossbar
Fabric
ASIC
1: Fabric Request
Crossbar
Fabric
ASIC
Ingress LC
Arbiter
FIA
2: Arbitration
Crossbar
Fabric
ASIC
3: Fabric Grant
Crossbar
Fabric
ASIC
4: load-balanced
transmission across
fabric links
Arbiter
RSP1
BRKSPG-2904
Cisco Public
111
5: credit return
Egress LC
FIA
BRKSPG-2904
Cisco Public
112
On Typhoon LCs, NP is designed for multiple 10G, 40G, and 100G ports.
vqi
voq
vqi
BRKSPG-2904
Cisco Public
113
NP0
PHY
NP1
PHY
NP2
PHY
NP3
5Gbps
BRKSPG-2904
CPU
CPU
FIA
FIA
Switch
Fabric
Cisco Public
114
NP0
PHY
NP1
PHY
NP2
PHY
NP3
PHY
Trident
Egress (NP->Line)
Ingress (NP->Fabric)
Egress (NP->Line)
Cisco Public
115
interface Gig0/0/1/8
Cisco Public
116
Note: The marking is implicit once you assign a packet into a given queue on
ingress; its sets the fabric header priority bits onto the packet.
no specific set action is required
BRKSPG-2904
Cisco Public
117
From wire
IFIB action
I/F
classification
ACL
classification
QOS
classification
Fwd lookup
QoS action
L2 rewrite
ACL action
IFIB lookup
To fabric
From fabric
egress linecard
ACL action
QoS action
BRKSPG-2904
QOS
classification
ACL
classification
To wire
Cisco Public
118
L2 rewrite
Fwd lookup
IFIB action
classification
classification
classification
QoS action
L2 rewrite
ACL action
To fabric
IFIB lookup
ACL action
QOS
classification
Police
QoS action
BRKSPG-2904
ACL
classification
Mark
To wire
Cisco Public
119
L2Queue/sha
rewrite
pe/WRED
Fwd lookup
Injected packets
In general are injected to-wire (same as Pak Priority in IOS)
Means that all features are bypassed.
Including QOS
Few exceptions
ICMP
Netflow
BRKSPG-2904
Cisco Public
120
BRKSPG-2904
Cisco Public
121
Higher scale
Higher queue and policer scale
More granular bandwidth control for both policing and queuing
Higher buffer size
No input shaping on high-NP loading configs (36x10G, 8x10 MPA, 40G MPA)
Cisco Public
122
BRKSPG-2904
Cisco Public
123
Higher scale
Higher queue and policer scale
More granular bandwidth control for both policing and queuing
Higher buffer size
Cisco Public
124
BRKSPG-2904
Cisco Public
125
Inner/outer cos
Inner/Outer vlan *
DEI*
Outer EXP
Dscp/Tos
TTL, TCP flags, source/destination L4 ports
Protocol
Source/Destination IPv4
Source/Destination MAC address*
Discard-class
Qos-group
match all/match any
Note:
Not all fields are supported on L3 interfaces*
Some fields dont make sense on ingress (e.g. discard-class, qos-group)
MPLS classification is based on EXP only (note in 530 we will be able to apply QOS matching (tentative) and
ACL matching on MPLS labeled packets)
BRKSPG-2904
Cisco Public
126
Fields
supported
BRKSPG-2904
Format 0
Format 1
Format 2
Format 3
Outer
VLAN/COS/DEI
Inner VLAN/COS
IPV4 Source
address
(Specific/Range)
IP DSCP / TOS /
Precedence
QOS-group (output
policy only)
Discard-class
(output policy only)
Outer
VLAN/COS/DEI
Inner VLAN/COS
IPV4 Destination
address
(Specific/Range)
IP DSCP / TOS /
Precedence
QOS-group (output
policy only)
Discard-class
(output policy only)
Outer
VLAN/COS/DEI
Inner VLAN/COS
MAC Destination
address
MAC source
address
QOS-group (output
policy only)
Discard-class
(output policy only)
Cisco Public
127
dscp/precedence
EXP imposition
EXP topmost
cos inner/outer
qos-group
discard-class
ASR9K supports maximum of 2 fields per class-map. The same 2 fields can be
placed in any combination below
BRKSPG-2904
Cisco Public
128
Policy-map parent
Class class-default
Police rate 10 Mbps peak-rate 20 mbps
conform-action set dscp af12
conform-action set cos 2
exceed-action set dscp af13
exceed-action set cos 3
transmit
drop
Cisco Public
129
policy-map child
class class1
police rate 20 mbps peak-rate 50 mbps
class class2
police rate 30 mbps peak-rate 60 mbps
policy-map parent
class class-default
service-policy child
police rate 60 mbps
child-conform-aware
BRKSPG-2904
Cisco Public
130
BRKSPG-2904
Cisco Public
131
Use bandwidth (minBw) for a CIR guarantee relative to the parent hierarchy level
Min. RATE: 64kbps (8k granularity)
Two parameter scheduler support at class level and subscriber group level (L4, L2):
Shape & BwR (ratio / percent)
Shape & MinBw (absolute / percent)
Not supported: BwR & MinBw on the same class
BRKSPG-2904
Cisco Public
132
WRED configuration unit options are: bytes, kbytes, mbytes, us, ms, packets
These values will be rounded up to a set of pre-defined profiles ranging from 8 kB to
262144 kB
The actual implementation uses 512 byte buffer particles
BRKSPG-2904
Cisco Public
133
Absolute vs Percentage
All relevant policy actions support both, absolute and percentage based
configuration:
shape
bandwidth
Police
bandwidth remaining*
For tri-rate Copper SFPs (10/100/1000) percentage based QOS will be adjusted
automatically based on the selected rate
Cisco Public
134
Display HA Chkpt related info for all the chkpt tables for QoS EA
BRKSPG-2904
Cisco Public
136
137
Parent
BW
is
zero
or
64k,
only
applicable
in
oversubscription
at
sum
of parent levels
Shape Profile: 1 CIR: 64 kbps CBS: 10240 bytes PIR: 499968 kbps PBS: 6291456 bytes
WFQ Profile: 4 Committed Weight: 1 Excess Weight: 1
Bandwidth: 0 kbps, BW sum for Level 0: 0 kbps, Excess Ratio: 1
policy-map dummy
---------------------------------------------------------------------class class-default
Level: 1 Policy: dummy Class: class-default
bandwidth percent 100
Parent Policy: p-map Class: class-default
!
QueueID: 642 (Priority Normal)
end-policy-map
Queue Limit: 8388 kbytes Profile: 2 Scale Profile: 2
!
WFQ Profile: 5 Committed Weight: 1020 Excess Weight: 1020
Bandwidth: 500000 kbps, BW sum for Level 1: 500000 kbps, Excess Ratio: 1
---------------------------------------------------------------------BRKSPG-2904
2014 Cisco and/or its affiliates. All rights reserved.
Cisco Public
138
Cisco Public
139
policy-map p-map
class class-default
service-policy dummy
shape average 500 mbps
!
end-policy-map
!
policy-map dummy
class class-default
bandwidth percent 100
!
end-policy-map
!
140
Cisco Public
141
linerate
PBS
PIR
CIR
RP/0/RSP0/CPU0:A9K-BNG#show policy-map int g 0/0/0/0 | i Queue
Queueing statistics
Queue ID
Queue(conform)
Queue(exceed)
BRKSPG-2904
Cisco Public
142
:
:
: 136
0/0
0/0
0
0
QOS summary
All Ethernet linecards support Queuing, Marking and Policing.
Some high speed linecards do not support ingress Queuing (but support policing
and marking).
Because their ingress TM (Traffic Manager) is disabled
To guarantee priority end to end, make sure high priority traffic is marked on
ingress (This will not burn a queue)
https://supportforums.cisco.com/docs/DOC-15592
BRKSPG-2904
Cisco Public
143
Better quality
Better identification of what should be smud
Cisco Software Manager (in PRIME now too). V2 just released
Better DDTS info
Cisco Public
145
Smart Reload
Minimizing Upgrade Time
4.2.1
Benefits
Faster image upgrades
Faster reload SMU installation and activation
Significantly reduced traffic loss due to
unplanned events (e.g. power outages)
Reduces maintenance window duration
Supports any-to-any image upgrade
Improved TFTP image download time
Improved traffic convergence time by
optimizing prefix download and programming
time.
Early availability of IOS-XR prompt, Line
Card services and resources
BRKSPG-2904
Cisco Public
146
34 Minutes
4.3.0
21 Minutes
4.3.1
19 Minutes
5.1.1
11 Minutes
Machine to Machine
BRKSPG-2904
Cisco Public
147
Machine to Machine
BRKSPG-2904
Cisco Public
148
XR Software Manager
Next Enhancements
SMU ETA visibility
Auto SW download(BSD)
PDF format conformance report
Machine to Machine
BRKSPG-2904
Cisco Public
149
ISSU SMU
Differentiation
Up to 60% of
reload SMUs can
be converted to
ISSU SMUs
Reduce packet
loss to <60 sec**
from 20-30
Cisco Public
How do we Measure ?
Current 4.3.2 reload SMUs stands at 0%
Goal: Reload SMUs cannot be more then 20%
BRKSPG-2904
Cisco Public
151
XR Service Packs
Minimizing System Reloads
Service Packs on XR provide packaging of SMUs, and
reduce the number of reboots for software updates
between releases
Benefits
Highlights
BRKSPG-2904
Cisco Public
152
IPV6-Etherconsole:
Customers dont want to use serial (RS232) console.
RS232 to Ethernet conversion using External Dongle (e.g. Raspberry Pi) is not a viable
solution.
Avoid the need to use the serial (RS232) console by ensuring that the router will always
be accessible over IPV6 via Management Ethernet interfaces.
Hardware Solution:
Dedicated physical port in next-Gen RSP front Panel
Goal is to achieve true emulation of console port
Software Solution:
BRKSPG-2904
Cisco Public
153
BRKSPG-2904
group G-ISIS-INTERFACE
router isis .*
interface Gig.*
lsp-interval 20
hello-interval 40
address-family ipv4 unicast
metric 10
!
config
router isis green
apply-group G-ISIS-INTERFACE
Cisco Public
154
Accelerated Upgrade
Python based Auto upgrade tool
Architecture support Plugins to expand and
customize to any platform
Cisco Public
155
Feature Highlights
Supported on all XR platforms
Can perform install upgrade or turboboot
Its open sourced and official, anywan can modify
script as needed
Customer can share AU with Cisco or community
Availability: NOW
156
BRKSPG-2904
Cisco Public
Roadmap highlights
Integrate AU with CSM
Add support for NG XR
Simple script to support reading config file
Support setting ROMMON variable from file for
turboboot
Availability: NOW
157
BRKSPG-2904
Cisco Public
Accelerated Upgrade
Syntax
DEMO
Pre-Upgrade Check
Plugins
172.28.98.3
= ASR9001
158
BRKSPG-2904
Cisco Public
Easy Upgrades
Released: 5.1.1
Feature Highlights
Fewer Commands
Simplified Upgrade
On RSP2
Cisco Public
Whats next?
Continued focus on Support forums
Read the blog for useful announcements
Q&A
Tech docs
BRKSPG-2904
Cisco Public
160
You can submit an entry for more than one of your favorite speakers
Dont forget to follow @CiscoLive and @CiscoPress
View the official rules at http://bit.ly/CLUSwin
BRKSPG-2904
Cisco Public
162
BRKSPG-2904
Cisco Public
163
BRKSPG-2904
Cisco Public
164