Overview
Convergence Overview
- For a network to be converged, all nodes must have the same information
- Must not contradict each other
- 4 main factors that effect convergence time:
- Failure detection time
- The time it takes to detect a failure
- Event propagation time
- The time it take for me to update everyone
- The time it takes to recalculate the new topology
- Forwarding table update time
- The time it takes to install the changes
Modifying the Convergence Time
Modifying the convergence time
- Reactive methods
- Fast Hellos and BFD
- OSPF LSA and SPF Pacing Timers
- FIB Prefix prioritisation
- Proactive methods
- Used to forward through failures
- Pre-calculated backup paths
- EIGRP Feasible Successors
- OSPF Loop Free Alternative (LFA)
- BGP Prefix Independant Convergence (PIC)
- MPLS TE Fast Reroute (TE FRR)
Routing Protocol Failure Detection
Routing Protocol Failure Detection
- Link failure events
- Based on Layer 1 issues
- Time to detect failure based on
- Link debounce for switches
- Carrier delay for routers
- Default 2 seconds
- Can be adjusted manually
- Command:
- (config-if)#carrier-delay [msec] <time>
- Can be adjusted to suppress short link flaps
- Not Layer 1 adjacent
- Link up/down unreliable
- Use upper layer timers
- OSPF/EIGRP Hello/Dead time
- OSPF Supports sub-seconds Hello timers
- Not recommended because of CPU load
- eBGP uses fast fall-over
- Can support per-peer fall-over
- Per-peer fall-over allows conditional checking
- Uses route-map with prefix-list checking if peer is reachable by a route
IP Event Dampening
IP Event Dampening
- Used to punish links that keep flapping
- Causes adjacencies to come up and go down
- Protocol independent
- Uses a configurable penalty method
- Links exceeding the penalty are removed from RIB
- Configured at the interface level
- Command
- (config-if)#dampening <half-life> <value to start reusing> <value to start suppressing> <max duration to suppress>
- To display dampening information for interfaces
- #show interfaces dampening
Event Propagation Time
Event Propagation Time
- The length of time taken to inform all routers in the topology
- Most configurable option out of all 4 factors that affect convergence
EIGRP
- Bounded by QUERY and REPLY process
- Smaller QUERY domain, less QUERY packets
- Improved by
- Feasible Successors
- Stub Routers
- Command:
- (config-router)#eigrp stub
- Summarization
- Command:
- (config-if)#ip summary-address eigrp <AS> <network> <mask>
[_/su_spoiler]
OSPF
- LSA Flooding process
- Less devices in the Area, less need to be informed
- Improved by:
- Areas
- Stub Areas
- Incremental SPF (iSPF)
- Runs a partial SPF only on areas of the SPT that were affected
- Should be enabled for all devices in domain
- Command:
- Prefix Suppression
- Prevents router advertising all prefixes except:
- Loopback addresses
- Secondary IP addresses
- Passive interfaces
- Command:
- (config-if)#ip ospf prefix-suppression
- (config-router)#prefix-suppression
- SPF and LSA Throttling Pacing timers
- Control consecutive SPF calculation events
- Command:
- (config-router)#timers throttle spf <spf-start> <spf-hold> <spf-max-wait>
- Control LSA Flooding events
- Command:
- (config-router)#timers throttle lsa [all] <start-interval> <hold-interval> <max-interval>
- Minimum interval at which same LSA can be accepted from neighbors
- Command:
- (config-router)#timers lsa arrival <milliseconds>
- Control the interval rate between a group of LSAs being refreshed
- Longer timer means OSPF tries to group more updates together to send rather than send more individually
- Command:
- (config-router)#timers pacing lsa-group <seconds>
- Control minimum interval of consecutive LSA updates in OSPF transmission queue
- Command:
- (config-router)#timers pacing flood <milliseconds>
- Time interval to retransmit lost LSAs
- Command:
- (config-router)#timers pacing retransmission <milliseconds>
[_/su_spoiler]
BGP
- UPDATE/WITHDRAW process
- Unbounded
- Internet Routing Table never converges
- TCP based protocol, therefore improvements must be done to the TCP stack to improve BGP
- Improved by
- MTU Size
- Larger MTU means more updates in individual packets
- Set to Jumbo MTU 9216
- Configured at interface level
- TCP Maximum Segment Size
- Provides similar benefits to MTU
- Configured globally
- Default MSS is 1460 bytes on a local LAN, else it is 536 bytes
- Command:
- (config)#ip tcp mss <68-10000>
- TCP Window Size
- Defines the number of packets that can be sent without receiving an acknowledgement
- Configured globally
- Default set to 4128 when scaling not enabled
- If only one neighbor is configured for scaling, default window size is 65535
- Command:
- (config)#ip tcp window-size <68-1073741823>
- TCP Acknowledgements
- Acknowledges received packets
- If packets are lost from one window of data, an aggressive sender could re-send packets early even though later sequence packets might have already been received
- This can be adjusted using the Selective Acknowledgement configuration
- Returns selective acknowledgements to sender, informing what has been received
- Sender only re-sends the missing data
- Acknowledgements not necessarily sequential
- Improves overall performance
- Enabled globally
- Command:
- (config)#ip tcp selective-ack
[_/su_spoiler]
EIGRP FRR and OSPF LFA
EIGRP FRR and OSPF LFA
EIGRP Fast Re-Route
- Feasible Successor already provides a loop free alternate path
- FRR Installs a backup path to destination prefix
- Not used for ECMP or UCMP
- Displays as a "Repair path" in the RIB and FIB
- If there are no Feasible Successors then FRR doesn't do anything
- Provides an Active/Standby configuration for prefixes
- When Successor fails, Feasible Successor copied from EIGRP Topology table to RIB
- RIB writes to FIB
- FIB writes to TCAM
- Repair "Backup" Path are already pre-written into hardware with FRR
- Configured under topology in named mode or in classic mode under the global process
- Can be configured for all prefixes, although not recommended.
- Command:
- (config-router)#fast-reroute per-prefix all
- (config-router-af-topology)#fast-reroute per-prefix all
- Can be configured for individual prefixes using a route-map
- Command:
- (config-router)#fast-reroute per-prefix route-map <route-map-name>
- (config-router-af-topology)#fast-reroute per-prefix route-map <route-map-name>
[_/su_spoiler]
OSPF Loop Free Alternative
- End intention is to provide a function the same as EIGRP Feasible Sucessor
- Calculate a backup path (repair-path) for the primary
- Have an Active/Standby prefix configuration
- Traditionally OSPF only calculates a single loop-free path from itself to each node
- LFA runs the SPF tree for all its neighbors once own local loop-free path calculated, to calculate backup paths
- The end result gets written to the RIB as a Repair-Path
- If a failure occurs within the SPT;
- LSA flooding and recalculation still occurs but the result is already pre-written in the FIB
- Could cause scalability issues depending on Area size
- Not recommended to run on all prefixes
- Limit to selected prefixes
- Configured under the OSPF process
- Enabled on Area or individual prefixes
- Area configuration
- Command:
- (config-router)#fast-reroute per-prefix enable area <area-id>
- Individual Prefix configuration
- Command:
- (config-router)#fast-reroute per-prefix enable prefix-priority <low | high>
- Low keyword indicates to calculate LFA for ALL prefixes not just high prefixes
- High keyword indicates to calculate LFA for HIGH prefixes only
- To configure specific prefixes as high priority
- Command:
- (config-router)#prefix-priority high route-map <route-map-name>
- Directly connected prefixes can be disabled from LFA
- Command:
- (config-router)#ip ospf fast-reroute per-prefix protection disable
- All neighbors can be excluded from LFA
- Command:
- (config-router)#ip ospf fast-reroute per-prefix candidate disable
- Verification:
- Command:
- #show ip ospf neighbors fast-reroute
- #show ip route repair-paths
[_/su_spoiler]
BGP Prefix Independent Convergence (PIC)
BGP Prefix Independent Convergence Edge (PIC)
- Used with MPLS L3 VPN
- Normally used for dual-homed/dual PE connections
- Same as OSPF LFA and EIGRP FRR
- Installs a repair path in the FIB
- Next best path
- If the primary path fails, the repair path immediately takes over
- This process is unique for each node
- Can select individual prefixes to create a repair path
- BFD required for fast failure detection of neighbors, otherwise still waiting for update messages
- Enabled manually under the BGP AF or SAFI (Subsequeny Address Family Identifier)
- Command:
- (config-router-af)#bgp additional-paths install
- More preferred option to enable for specific prefixes, otherwise could reach limitations of hardware quickly e.g. TCAM
- Command:
- (config-router-af)#bgp additional-paths select [all | backup | best | best-external | group-best]
- Can view the repair path:
- #show ip cef vrf <vrf-name> <prefix> detail
- Detail word required, normal CEF view won't show repair path
[_/su_spoiler]
Troubleshooting Commands
- #show ip route repair-paths
[_/su_spoiler]
