MPLS Traffic Engineering with Auto-Bandwidth: Operational Experience and Lessons Learned. Mohan Nanduri Mark Kasten Naoki Kitajima

Transcription

1 MPLS Traffic Engineering with Auto-Bandwidth: Operational Experience and Lessons Learned Mohan Nanduri Mark Kasten Naoki Kitajima

2 Agenda Auto-Bandwidth basics and overview Operational Experience Lessons Learned Configuration examples Show command output examples

3 MPLS Auto-Bandwidth Basic and Overview Automates the process of monitoring and adjusting LSP bandwidth MPLS auto-bandwidth uses traffic flows through the LSP and automatically adjusts the bandwidth based on the measured traffic flowing through the LSP LSP bandwidth is adjusted on a per-lsp based on configurable values Automatic bandwidth adjustment feature treats each LSP independently to adjust bandwidth according to the adjustment frequency configured for the LSP without regard for any adjustments previously made or pending for other LSPs

4 MPLS Auto-Bandwidth Basic and Overview Auto-Bandwidth Calculation mechanism: Uses average per-lsp traffic utilization samples and picks the highest measured bandwidth value observed during the configured interval At the end of each interval the highest measured bandwidth value is compared with the existing LSP bandwidth value If the delta is between existing LSP bandwidth and highest measured bandwidth value chosen is equivalent or greater than adjust-threshold value, the LSP tries to re-signal with the new bandwidth, otherwise the LSP bandwidth remains the same Make-before-break mechanism kicks into place and LSP re-signals for the new bandwidth. Once it s successful in setting up the LSP, the maximum bandwidth value is reset to zero. The above process is repeated for the next adjust-interval Auto-Bandwidth uses two mechanisms to adjust the bandwidth Adjust-Interval Overflow/underflow

5 MPLS Auto-Bandwidth With Adjust- Interval The highest measured bandwidth value is 100 Mbps for a given adjust-interval period, the LSP is signaled with 100 Mbps and its reset to zero. As the new adjust-interval starts, the highest measured bandwidth value is 50 Mbps. LSP is signaled with 50 Mbps if the adjust-threshold value is met. If for any reason the LSP cannot signal 50 Mbps, the LSP s bandwidth will remain at 100 Mbps.

6 MPLS Auto-Bandwidth With Overflow If the highest measured bandwidth value is higher than the current LSP signaled bandwidth value. The delta between highest measured bandwidth value and current LSP signaled bandwidth value is greater than or equal to configured adjust-threshold value. The number of occurrences must reach the configured overflow count consecutively. Overflow counter resets to the configured overflow value if the bandwidth decreases during one the sampling cycles. LSP is signaled with the new bandwidth value before the configured adjustinterval. Adjust-interval timer is reset to configured value.

7 Benefits of Auto-Bandwidth Network can react faster to sudden burst of traffic in near real-time and not rely on manually intervention Effective use of bandwidth resources by minimizing the over-subscription/padding of LSP bandwidth Maximizes the usage of available bandwidth and optimizes the network effectively to signal better paths

8 Why auto-bandwidth for us? Extensive use of traffic engineering using offline bandwidth calculation and adjusting bandwidth based on historical traffic data Lagging behind bandwidth adjustments Not reacting to sudden spikes in traffic shifts

9 Operational Experience Auto-bandwidth deployed since early 2009 globally Some skepticism initially letting the device automatically adjust bandwidth Engineers who came from ISP background and were used to offline bandwidth calculation Worked fairly well so far Ran into some issues early on Smear timers not spacing LSPs properly Bloated traffic stats used to setup LSPs Phased rollout of auto-bandwidth globally Extensive analysis was performed on network after auto-bandwidth was introduced to optimize Significant churn by smaller sized LSPs Tiered LSP model was introduce to reduce churn RSVP interface subscriptions were modified to account spike in traffic Sudden spike in traffic was still a issue Automatic LSP creation and deletion using even scripts

10 Lessons Learned Significant amount of time was spent to understand internals, was tough to get info initially from vendor Many hours were spent testing the feature, it was a big change on the network, test all the available knob thoroughly, ran into multiple issues Thorough planning and training made the deployment very smooth Correlation scripts to check traffic stats and MPLS bandwidth setup Show commands lacks historical bandwidth adjustment information, have to glean information from multiple log files Multiple enhancements have been submitted to display info to make troubleshooting easier, to name a few: show previous bandwidth adjusted values, maximum bandwidth ever used and counter to keep track of bandwidth adjustments.

11 Configuration groups { autobw_lsp_setup { protocols { mpls { label-switched-path <*> { ldp-tunneling; soft-preemption; admin-group include-any [ gold]; adaptive; auto-bandwidth { adjust-interval 900; adjust-threshold 10; minimum-bandwidth 1m; maximum-bandwidth 2g; autobw_lsp_setup_1m_profile { protocols { mpls { label-switched-path <*> { ldp-tunneling; admin-group include-any [ gold ]; adaptive; auto-bandwidth { adjust-interval 900; adjust-threshold 50; minimum-bandwidth 1m; maximum-bandwidth 50m; label-switched-path rtr1-to-rtr2 { apply-groups autobw_lsp_setup; to ; admin-group exclude [bad]; label-switched-path rtr1-to-rtr3 { apply-groups autobw_lsp_setup_1m; to ; admin-group exclude [bad]; mpls { statistics { file mpls_stats.log size 1m files 10; interval 60; auto-bandwidth;

12 Show Commands show mpls lsp ingress extensive Ingress LSP: 119 sessions From: , State: Up, ActiveRoute: 86, LSPname: rtr1-to-rtr2 ActivePath: standard_path (primary) LSPtype: Static Configured LoadBalance: Random Autobandwidth MinBW: 5Mbps MaxBW: 250Mbps AdjustTimer: 900 secs AdjustThreshold: 50% Max AvgBW util: kbps, Bandwidth Adjustment in 806 second(s). Overflow limit: 0, Overflow sample count: 0 Encoding type: Packet, Switching type: Packet, GPID: IPv4 *Primary standard_path State: Up Priorities: 3 3 Bandwidth: Mbps OptimizeTimer: 900 SmartOptimizeTimer: 180 Include Any: transpacific favored core Exclude: transatlantic Reoptimization in 226 second(s). mohan@sat-96c-poc-netwk-1a# run show log mpls-new.stats Jan 2 04:53:33 trace_on: Tracing to "/var/log/mpls-new.stats" started HiPri-96c-to-16c (LSP ID 2, Tunnel ID 23040) pkt Byte pps Bps Reserved Bw 0 Bps 96c-to-16c-1 (LSP ID 2, Tunnel ID 23044) pkt Byte 0 pps 0 Bps Reserved Bw 0 Bps Jan 2 04:54:33 Total 34 sessions: 7 success, 0 fail, 27 ignored HiPri-96c-to-16c (LSP ID 2, Tunnel ID 23040) pkt Byte pps Bps Reserved Bw 0 Bps 96c-to-16c-1 (LSP ID 2, Tunnel ID 23044) pkt Byte 0 pps 0 Bps Reserved Bw 0 Bps