180 likes | 512 Views
Verification of RSTP Convergence and Scalability by Measurements and Simulations. I stv án Moldován, Saad Abuguba, Csaba Lukovszki 1 1-14 Decembe r 2006. Background. Metro Ethernet, Ethernet Access, Ethernet aggregation All using Spanning Tree Protocol In the LAN
E N D
Verification of RSTP Convergence and Scalability by Measurements and Simulations István Moldován, Saad Abuguba, Csaba Lukovszki 11-14December 2006
Background • Metro Ethernet, Ethernet Access, Ethernet aggregation • All using Spanning Tree Protocol • In the LAN • No special requirements, just simplicity • In the provider’s network • Carrier grade requirements! • Carrier Grade Questions: • How fast is the restoration? • Does it scale?
STP – Why needed? • Ethernet does not have TTL field • Loop protection is required • Spanning Tree Protocol reduces the active topology to a tree • Alternatives • Disable STP – not needed on tree topologies • VLAN based trees • VLAN topologies configured as p2p or p2mp connections • management is responsible for loop control • a configuration error becomes FATAL
Spanning Tree standards, problems • Basic STP – IEEE 802.1D – 1998 • timer based operation • slow (~1minute restoration) • does not scale • Rapid STP – IEEE 802.1w - 2001 – IEEE 802.1D - 2004 • much faster • but still no upper bound on convergence • several problems revealed (like count to infinity) • Scalability? • Multiple STP – IEEE 802.1s - 2002 • multiple regions • multiple trees within regions • scalable
Root R R R D D D A A A B B B D Spanning Tree Basics Loop-free Connectivity • One root bridge per network • One root port per non-root bridge • One designated port per segment • Non-designated ports are blocked Root Port (Fwd): Port receiving the best BPDU for the bridge – shortest path to the Root in terms of path cost Designated Port (Fwd): Port sending the best BPDU on a segment Alternate Port (Disc): Port blocked by BPDUs from a different bridge – redundant path to the Root Backup Port (Disc):Port blocked by BPDUs sent from the same bridge – redundant path to a segment
Proposal Agreement Proposal Proposal Agreement Agreement IEEE 802.1w sequence of events • Receive a proposal • Block all other non-edge ports • Send an agreement back • Put the new root port to forwarding • Send out proposals on other ports • Receive agreement from others • Put ports into forwarding Forward Edge port Forward Forward Block Block
Simulations on convergence • We used OPNET Modeler 10.5 simulation tool • RSTP supported by default • RSTP standard compliance verified (bug removed) • Simulations on different resilient topologies • Dual Homing topologies • deep: multiple levels • wide: high aggregation • Ring topologies • Mixed ring and dual homing • Simulation objective: restoration time after failure
Dual-homing scenarios • Dual homing connections for resilience • Simulations with increasing number of layers • Traffic between node_5 and node_8 • primary path fails • we measure restoration time Recovery happens almost instantaneously by accepting a proposal sent earlier on the other link We have found similar results for all investigated dual-homing topologies -no matter of the width, depth and failure location, since there is always an alternate port to the root
RSTP restoration on ring topology • Different ring sizes • from 6-14 • Different bridge BPDU processing times • 10-5000 BPDU/s • Recovery time observed • From port state changes Basically limited by the bridge BPDU processing time Measured BPDU processing time on real bridge: - 0.025s, std. deviation of 0.001s
Hold Timer • TxHoldCount • At least one BPDU per HelloTime interval, and not more than (TxHoldCount + 1) BPDUs in one second • By Standard selectable • Between 1-10 • The value of 1 introduces delays • Hello BPDU + other BPDU can not go in the same second
Count to infinity problem • Documented behavior • On failure of root long failover times • Several seconds! • Reason: • Old information persists and circulates • Until • message ages out or • Root path costs reaches maximum value Root
RSTP scalability issues • Big ring • BPDU information ages • ports where BPDU ages out becomes forwarding (standard says so) • Two spanning trees formed! • 2 roots LOOP CONDITION!
Maximum Bridge Diameter • IEEE 802.1D – 2004 does not contain this statement • Diameter limit is given by the value of Maximum Age value • 20 by default, up to 40 • Thus theoretically topologies with diameter up to 40 hops can be created • Diameter distance from root! • Diameter = the maximum length path in the network • Important since in case of failure longer paths may be formed • 2 trees, possible loops
Conclusions • RSTP convergence is limited by BPDU processing speed • Fast convergence in typical topologies (below 1 second) • No upper limit on convergence • Depends on topology and bridge BPDU processing speed • Vulnerable to count to infinity problem • Limited scalability • By default up to diameters of 20 • Can be tuned up to 40 (theoretical limit)
Thank you for your attention! Questions?