Passive Analysis of Anycast Routing and the Impact of Remote Peering

Towards Passive Analysis of Anycast in Global Routing: Unintended Impact of Remote Peering,
Rui Bian, Shuai Hao, Haining Wang, Amogh Dhamdere, Alberto Dainotti, and Chase Cotton.
ACM SIGCOMM Computer Communication Review (CCR), Volume 49 Issue 3, July 2019.

We developed a passive method to study IP anycast by utilizing BGP data. We proposed a set of BGP-related features to classify anycast/unicast prefixes by using the BGP data (RouteViews and RIPE RIS) and evaluated the effectiveness of our approach by using the active measurement results as a near-ground-truth. The results show that our approach achieves high classification accuracy, about 90% for anycast and 99% for unicast.

While further delving into the causes of inaccuracy, we found that remote peering has an unintended impact on anycast routing, due to its invisibility at layer-3, breaking the assumption that the peered autonomous systems are physically close and provide a short path. In our study, 19.2% of anycast prefixes are sensitive to remote peering and around 40% of such prefixes are further confirmed to be impacted by remote peering via traceroute measurements. [paper]

scripts

• construct_datasets.py: construct datasets for classification from full_results_anycast.txt and full_results_unicast.txt and datasets contain five features N, P1, P2, MD, ML, details in paper §3.2
• extract_asrelation_distance.py: process raw BGP file (BGP data from RouteViews and RIPE RIS from June 1st 00:00 UTC to June 1st 23:59 UTC) and extract the distance of ASes
• process_anycast_to_get_complete_data.py: process BGP data (BGP data from RouteViews and RIPE RIS from June 1st 00:00 UTC to June 1st 23:59 UTC) and extract BGP features
• process_data_mulway_ris.py: process raw BGP data, extract origin/upstream ASes, and the distance of upstream ASes for each prefix
• process_unicast_to_get_complete_data.py: process unicast prefix BGP data (BGP data from RouteViews and RIPE RIS from June 1st 00:00 UTC to June 1st 23:59 UTC) and extract BGP features
• decision_tree_save_results.py: classification using decision tree algorithm
• randomforest_save_results: classification using random forest algorithm

data

Passive Anycast Detection

• anycast_2017_uniq.txt: (near-)Ground truth of anycast prefixes from Anycast enumeration (CoNEXT'15 paper)
• as_distance: AS distance from RIPE RIS and RouteViews data
• datasets_prefix_N_P1_P2_MD_ML.txt: feature data for classification
• full_results_anycast.txt: anycast prefix feature data, the format is

prefix in groundtruth|prefix in BGP|origin AS|number of orgin AS|number of upstream AS (N)|number of collectors|number of upstream AS pairs|number of upstream AS pairs whose distance above 1|number of upstream AS pairs whose distance above 2| Percentage of upstream AS pairs whose distance is more than 1 (P1)|Percentage of upstream AS pairs whose distance is more than 2 (P2)|maximum distance between upstream ASes (MD)| minimum of distance between upstream ASes| mean of distance| standard deviation of distance|maximum length of AS paths (ML)|minimum of length of AS paths|mean of length of AS paths| standard deviation of length of AS paths

• full_results_unicast.txt: unicast prefixes' feature data, format is same as full_results_anycast.txt
• full_results_all.txt: all prefixes' feature data constructed from raw BGP data (BGP data from RouteViews and RIPE RIS from June 1st 00:00 UTC to June 1st 23:59 UTC), format is same as full_results_anycast.txt (except no groundtruth prefixes, because they are constructed from Raw BGP data)
• res_DT_5.txt: results of decision tree classification
• res_RF_5.txt: results of random forest classification

Remote Peering Data/Experiments

• Anycast-data-UMD: Anycast experiment data from University of Maryland's D-Root Nameserver (SIGCOMM'18 paper)
• asn_ixp: IXP member ASNs and remote peering ASes (IXP member ASN collected from IXP websites; remote peering AS data from IMC'18 remote peering paper)
• aspair_ixp: AS pair of IXP member and remote peering IP (constructed from IXP_members and asn_ixp)
• rp_in_ixp: BGP searching results of remote peering (using AS pairs to search in anycast prefix BGP records)
• anycast_2017_uniq.txt: (near-)Ground truth of anycast prefixes from Anycast enumeration
• atlas_exp_prefix_msmid.txt: Measurement ID of RIPE Atlas
• inferences_all.txt: information including local and remote peering from IMC'18 remote peering paper
• inferences_remote.txt: remote peering information from IMC'18 remote peering paper
• traceroute_prefix_ip_path.txt: traceroute results (IP paths) from RIPE Atlas experiments
• traceroute_prefix_asn_pyasn.txt: ASN paths from RIPE Atlas experiments
• path collection: path collection (details in paper §5.2) results from RIPE Atlas traceroute experiments

Due to limit of size, we didn't include large files like BGP data, which can be downloaded from RouteView/RIPE NCC, or retrieved from CAIDA's BGPStream. If you need those files to reproduce your results, we have included the method in our paper and feel free to contact us by email (bianrui@udel.edu).