Network Anomography

Network Anomography Matthew Roughan matthew.roughan@adelaide.edu.au Joint work withZihui Ge, Albert Greenberg, Yin Zhang AusCTW, Feb 2007

Network Anomaly Detection • Is the network experiencing unusual conditions? • Call these conditions anomalies • Anomalies often indicate network problems • DDoS, worms, flash crowds, outages, misconfigurations … • Need rapid detection and diagnosis • Want to fix the problem quickly • Questions of interest • Detection • Is there an unusual event? • Identification • What’s the best explanation? • Quantification • How serious is the problem?

B C A Network Anomography • What we want • Volume anomalies [Lakhina04]Significant changes in an Origin-Destination flow, i.e., traffic matrix element • What we have • Link traffic measurements • It is difficult to measure traffic matrix directly • Network Anomography • Anomography = anomalies + tomography • Infer volume anomalies from link traffic measurements

An Illustration OD flow I-b OD flow I-b Link c-b Link d-c Link f-d Link l-f Sat Fri Sun Courtesy: Anukool Lakhina [Lakhina04]

Mathematical Formulation Only measure at links 1 route 3 link 1 router 2 route 2 link 2 route 1 3 link 3 bt = At xt (t=1,…,T) Typically massively under-constrained!

Static Network Anomography Only measure at links 1 route 3 link 1 router 2 route 2 link 2 route 1 3 link 3 B = AX Time-invariantAt (= A), B=[b1…bT], X=[x1…xT]

Anomography Strategies • Early Inverse • Inversion • Infer OD flows X by solving bt=Axt • Anomaly extraction • Extract volume anomalies X from inferred X Drawback: errors in step 1 may contaminate step 2 • Late Inverse • Anomaly extraction • Extract link traffic anomalies B from B • Inversion • Infer volume anomalies X by solving bt=Axt Idea: defer “lossy” inference to the last step     

 Extracting Link Anomalies B  • Temporal Anomography: B = BT • ARIMA modeling • Diff: ft = bt-1 bt = bt – ft • EWMA: ft = (1-) ft-1 +  bt-1 bt = bt – ft • Fourier / wavelet analysis • Link anomalies = the high frequency components • Temporal PCA • PCA = Principal Component Analysis • Project columns onto principal link column vectors • Spatial Anomography: B = TB • Spatial PCA [Lakhina04] • Project rows onto principal link row vectors   

 Extracting Link Anomalies B  • Temporal Anomography: B = BT • Self-consistent • Tomography equation: B = AX • Post-multiply by T: BT = AXT B = AX • Spatial Anomography: B = TB • No longer self-consistent   

  Solving bt = A xt   • Pseudoinverse: xt = pinv(A) bt • Shortest minimal L2-norm solution • Minimize |xt|2subject to |bt – A xt|2 is minimal • Maximize sparsity (i.e. minimize |xt|0) • L0-norm is not convex  hard to minimize • Greedy heuristic • Greedily add non-zero elements to xt • Minimize |bt – A xt|2 with given |xt|0 • L1-norm approximation • Minimize |xt|1 (can be solved via LP) • With noise  minimize |xt|1 +  |bt-Axt|1            

Dynamic Network Anomography • Time-varying At is common • Routing changes • Missing data • Missing traffic measurement on a link setting the corresponding row of At to 0 in bt=Atxt • Solution • Early inverse: Directly applicable • Late inverse: Can extend ARIMA modeling • L1-norm minimization subject to link constraints • minimize |xt|1subject to xt = xt – xt-1, bt=Atxt, bt-1=At-1xt-1 • Reduce problem size by eliminating redundancy  

Performance Evaluation: Inversion • Fix one anomaly extraction method • Compare “real” and “inferred” anomalies • “real” anomalies: directly from OD flow data • “inferred” anomalies: from link data • Order them by size • Compare the size • How many of the top N do we find • Gives detection rate: | top Nreal top Ninferred | / N

Inference Accuracy  Tier-1 ISP (10/6/04 – 10/12/04) Diff (bt = bt = bt – bt-1) Sparsity-L1 works best among all inference techniques

Inference Accuracy  Tier-1 ISP (10/6/04 – 10/12/04) Diff (bt = bt = bt – bt-1) detection rate = | top Nreal top Ninferred | / N Sparsity-L1 works best among all inference techniques

Impact of Routing Changes  Tier-1 ISP (10/6/04 – 10/12/04) Diff (bt = bt = bt – bt-1) Late inverse (sparsity-L1) beats early inverse (tomogravity)

Impact of Missing Data  Tier-1 ISP (10/6/04 – 10/12/04) Diff (bt = bt = bt – bt-1) Late inverse (sparsity-L1) beats early inverse (tomogravity)

Performance Evaluation: Anomography • Hard to compare performance • Lack ground-truth: what is an anomaly? • So compare events from different methods • Compute top M “benchmark” anomalies • Apply an anomaly extraction method directly on OD flow data • Compute top N “inferred” anomalied • Apply another anomography method on link data • Report min(M,N) - | top Mbenchmarktop Ninferred | • M < N  “false negatives” # big “benchmark” anomalies not considered big by anomography • M > N  “false positives” # big “inferred” anomalies not considered big by benchmark method • Choose M, N similar to numbers of anomalies a provider is willing to investigate, e.g. 30-50 per week

Anomography: “False Negatives” • Diff/EWMA/H.-W./ARIMA/Fourier/Wavelet all largely consistent • PCA methods not consistent (even with each other) • - PCA cannot detect anomalies in the “normal” subspace • - PCA insensitive to reordering of [b1…bT]  cannot utilize all temporal info • Spatial methods (e.g. spatial PCA) are not self-consistent

Anomography: “False Positives” • Diff/EWMA/H.-W./ARIMA/Fourier/Wavelet all largely consistent • PCA methods not consistent (even with each other) • - PCA cannot detect anomalies in the “normal” subspace • - PCA insensitive to reordering of [b1…bT]  cannot utilize all temporal info • Spatial methods (e.g. spatial PCA) are not self-consistent

Summary of Results • Inversion methods • Sparsity-L1 beats Pseudoinverse and Sparsity-Greedy • Late-inverse beats early-inverse • Anomography methods • Diff/EWMA/H-W/ARIMA/Fourier/Wavelet all largely consistent • PCA methods not consistent (even with each other) • PCA methods cannot detect anomalies in “normal” subspace • PCA methods cannot fully exploit temporal information in {xt} • Reordering of [b1…bT] doesn’t change results! • Spatial methods (e.g. spatial PCA) are not self-consistent • Temporal methods are • The method of choice: ARIMA + Sparsity-L1 • Accurate, consistent with Fourier/Wavelet • Robust against measurement noise, insensitive to choice of • Works well in the presence of missing data, routing changes • Supports both online and offline analysis

Conclusions • Network Anomography • Find anomalies in {xt} given bt=Atxt (t=1,…,T) • Contributions • A general framework for anomography methods • Decouple anomaly extraction and inference components • A number of novel algorithms • Taking advantage of the range of choices for anomaly extraction and inference components • Choosing between spatial vs. temporal approaches • The first algorithm for dynamic anomography • Extensive evaluation on real traffic data • 6-month Abilene and 1-month Tier-1 ISP • The method of choice: ARIMA + Sparsity-L1

Future Work • Correlate traffic with other types of data • BGP routing events • Router CPU utilization • Anomography for performance diagnosis • Inference of link performance based on end-to-end measurements can be formulated as bt=Axt • Beyond networking • Detecting anomalies in other inverse problems • Are we just reinventing the wheel

Thank you !

Network Anomography