2b:null 2c:["$","title",null,{"children":"Extending RIPE Atlas in an unbiased way | Pavlos Sermpezis | Observable"}] 31:T41f,## Recommendations Data Table: ASNs, their characteristics, and bias difference scores The following table shows for the ASNs that do not host any RIPE Atlas probe the following data - **_Bias difference score_:** That is the difference in bias they can bring if added to the RIPE Atlas infrastructure (i.e., deploy at least on probe). Negative values mean that the bias is decreased, and positive values that the bias is increased. The smaller the value the more the bias is decreased (e.g., -1% is better than -0.5%). - **_Network characteristics_:** Some characteristics of the ASNs, e.g., size, customer cone, neighbors, network type, peering policies, connectivity to IXPs, etc. The table is by default sorted by the TOTAL bias difference score, with first appearing the ASNs with the smaller bias difference scores (i.e. those that would be most preferred). You can sort the table by other columns and by ascending/descending order by clicking on the headers. Every time a filter option or a bias weight changes (see below), the table is updated.32:T706,viewof table = Inputs.table(filtered_data, { sort: "TOTAL bias (selected)", columns: [ "ASN", "RIR region", // "Location (continent)", "Location (country)", "TOTAL bias (selected)", "Location bias", "Network size bias", "Topology bias", "IXP related bias", "Network type bias", // "TOTAL bias", "Customer cone (#ASNs)", "Customer cone (#prefixes)", // "AS hegemony", "#neighbors (total)", // "#neighbors (customers)", // "#neighbors (peers)", // "#neighbors (providers)", // "AS_rel_degree", "Scope (PeeringDB)", "Network type (PeeringDB)", "Peering policy (PeeringDB)", "#IXPs (PeeringDB)", "#facilities (PeeringDB)", // "is_in_peeringDB", // "nb_atlas_probes_v4", // "nb_atlas_probes_v6", // "is_ris_peer_v4", // "is_ris_peer_v6", // "is_routeviews_peer", // 'min_dist', // 'Improvement proximity' ], header: { "RIR region": "RIR", "Location (country)": "Country", "#neighbors (total)": "#neighbors", "#neighbors (customers)": "#customers", "#neighbors (peers)": "#peers", "#neighbors (providers)": "#providers", "Scope (PeeringDB)": "Scope", "Network type (PeeringDB)": "Net. type", "Peering policy (PeeringDB)": "Peering policy", "#IXPs (PeeringDB)": "#IXPs", "#facilities (PeeringDB)": "#facilities", "min_dist": "min dist. from RIS", "Improvement proximity": "Improvement" }, format: { "ASN": x => "AS"+x.toString(), "Location bias": x => x+"%", "Network size bias": x => x+"%", "Topology bias": x => x+"%", "IXP related bias": x => x+"%", "Network type bias": x => x+"%", // "TOTAL bias": x => x+"%", "TOTAL bias (selected)": x => x.toFixed(2)+"%", } })33:T54b,## Filtering and bias calculation options We provide a number of filtering and modifications options that enable to adapt the recommendations based on specific needs, constraints, etc. For example, we may want to take into account only location bias or we may want to consider only ASNs with a PeeringDB record. Using these options, the [recommendations data table](https://observablehq.com/@pavlos/ai4netmon-ripe-atlas-extension#cell-595) is updated automatically. - The bias (score) is calculated based on many different characteristics (location, network size/type, connectivity, etc.). We may not want to take into account all these dimensions, or we may want to put more weight on some of them. To this end, we provide [below](https://observablehq.com/@pavlos/ai4netmon-ripe-atlas-extension#cell-4) the option for the user to select a weight (from 0 to 1) for each group of dimensions. The _TOTAL_ bias difference score is calculated as the weighted sum of the bias difference scores per dimension group. - We may want to exclude some ASes (e.g., edge ASes, small ASes, not connected in IXPs) for feasibility, policy, regulations, or other reasons. To this end, we provide [below](https://observablehq.com/@pavlos/ai4netmon-ripe-atlas-extension#cell-631) different filtering options (i.e., to filter out some ASes with certain characteristics).34:T534,filtered_data = { let arr = searched_data // filter based on checkboxes: PDB, Atlas, RouteViews if (checkboxes_conditions.includes("Is in PeeringDB")) { arr = arr.filter(e => e["is_in_peeringDB"] == 1) } if (checkboxes_conditions.includes("Peers at IXP(s)")) { arr = arr.filter(e => e["#IXPs (PeeringDB)"] > 0 ) } // filter based on peering policy checkboxes // if (checkboxes_peer_policy.length > 0) { // if (checkboxes_peer_policy.includes('N/A')) { // arr = arr.filter(e => (checkboxes_peer_policy.includes(e["Peering policy (PeeringDB)"])) || (e["Peering policy (PeeringDB)"] == null)) // } else { // arr = arr.filter(e => checkboxes_peer_policy.includes(e["Peering policy (PeeringDB)"])) // } // } // filter by value on nb neighbors, customer cones, improvement const dict_colname_form = { "#neighbors (total)": form_nb_neighbors, "Customer cone (#ASNs)": form_cone_asns, "Customer cone (#prefixes)": form_cone_pfxs } for(var cn in dict_colname_form) { if (dict_colname_form[cn]["option"]==="larger than") { arr = arr.filter(e => e[cn] >= dict_colname_form[cn]["range"]) } if (dict_colname_form[cn]["option"]==="smaller than") { arr = arr.filter(e => e[cn] <= dict_colname_form[cn]["range"]) } } return arr; }35:[{"id":"4848089afe788f956458a1c762b1dcd5069d02866e403a63e7e5fca284f6408e9083c3eee78c2fb65cf48d8cf1444e0c0b69336ec7e5092b123785ba56dadd8a","href":"https://static.observableusercontent.com/files/4848089afe788f956458a1c762b1dcd5069d02866e403a63e7e5fca284f6408e9083c3eee78c2fb65cf48d8cf1444e0c0b69336ec7e5092b123785ba56dadd8a","name":"bias_df__no_stubs_Atlas_groups.csv","type":"text/csv","lastModified":1660822891123.897,"size":3392422}] 2f:{"id":"3039563c662f7ec0","slug":"ai4netmon-ripe-atlas-extension","title":"Extending RIPE Atlas in an unbiased way","thumbnail":"5cf0a5ed8aa5076f443b941f5ea972295d86d7ad5123bd5f7ef7e15e7bcc24bc","access_level":"public","create_time":"$D2022-08-17T13:37:19.213Z","update_time":"$D2023-06-29T12:53:55.347Z","view_time":null,"first_public_time":"$D2022-08-17T13:38:03.471Z","fork":{"id":"ed189c82955484d2","slug":"ai4netmon-ripe-ris-extension","title":"Extending RIPE RIS in an unbiased way","owner":{"id":"54a924218005a3d6","login":"pavlos","name":"Pavlos Sermpezis"}},"stars":0,"starred":false,"owner":{"type":"team","id":"54a924218005a3d6","login":"pavlos","name":"Pavlos Sermpezis","avatar_url":"https://avatars-next.observableusercontent.com/200aa95291d6d5ebe777baf6ba8e2f53b577798e1ae41ebc39d70811739e73d6","tier":"starter_2024"},"version":803,"publish_version":null,"role":null,"authored":false,"authors":[{"type":"individual","id":"0742c07d75b1a2ea","login":"pavlos","name":"Pavlos Sermpezis","avatar_url":"https://avatars-next.observableusercontent.com/200aa95291d6d5ebe777baf6ba8e2f53b577798e1ae41ebc39d70811739e73d6","tier":null}],"members":[{"type":"individual","id":"0742c07d75b1a2ea","login":"pavlos","name":"Pavlos Sermpezis","avatar_url":"https://avatars-next.observableusercontent.com/200aa95291d6d5ebe777baf6ba8e2f53b577798e1ae41ebc39d70811739e73d6","tier":null,"role":"editor"},{"type":"individual","id":"9b5b479f2484d7a6","login":"sof","name":"Sofia Kostoglou","avatar_url":"https://avatars-next.observableusercontent.com/74cd6d6a06ddc4b648f41ec6af8a76ed2bb5f51746c5d2d9715133b26167e46d","tier":null,"role":"editor"}],"comments":[],"custom_thumbnail":false,"trashed":false,"lapsed":null,"subscribed":null,"forks":1,"collection":{"id":"9a9ee52de9bfd299","slug":"ai4netmon","title":"AI4NetMon","owner":{"id":"54a924218005a3d6","login":"pavlos","name":"Pavlos Sermpezis"}},"type":"notebook","body":{"title":"Extending RIPE Atlas in an unbiased way","cells":[{"id":0,"value":"# Extending RIPE Atlas in an unbiased way\n\n\n_(For a similar notebook for the **RIPE RIS** case, see [here](https://observablehq.com/@pavlos/ai4netmon-ripe-ris-extension))_\n\n\nThis notebook provides recommendations for extending the RIPE Atlas infrastructure (i.e., selecting ASes to deploy RIPE Atlas probes) in an \"unbiased way\", based on the results and analyses of the [AI4NetMon](https://sermpezis.github.io/ai4netmon/) project. For an introduction on the bias of Internet measurement platforms see the [RIPE Labs article](https://labs.ripe.net/author/pavlos_sermpezis/bias-in-internet-measurement-infrastructure/).","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":769,"value":"**Quick view: top-5 recommendations** (see below the complete data & options) ","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":771,"value":"viewof short_table = Inputs.table(filtered_data.sort((a, b) => a[\"TOTAL bias (selected)\"] - b[\"TOTAL bias (selected)\"]).slice(0,5), {\n \n sort: \"TOTAL bias (selected)\",\n \n columns: [\n \"ASN\",\n \"RIR region\",\n \"Location (country)\",\n \"TOTAL bias (selected)\",\n ],\n \n header: {\n \"RIR region\": \"RIR\",\n \"Location (country)\": \"Country\",\n \"TOTAL bias (selected)\": \"TOTAL bias\"\n },\n\n format: {\n \"ASN\": x => \"AS\"+x.toString(),\n \"TOTAL bias (selected)\": x => x.toFixed(2)+\"%\",\n },\n \n width: {\n \"ASN\": 80,\n \"RIR region\": 50,\n \"Location (country)\": 50,\n \"TOTAL bias (selected)\": 60\n },\n maxWidth: 380, \n})","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":765,"value":"The recommendations are provided as follows: For every AS that does *not* host any RIPE Atlas probe, we calculate what would be the difference in bias if we deploy at least one probe in it. Hence, we calculate a _**bias difference score**_. Adding an AS can decrease or increase bias. We would prefer to decrease the bias, thus, we would prefer to add (if possible) the ASes that have the *most negative bias difference score*. \n\nThe table below provides the bias difference scores per AS (along with other characteristics of the ASes); for advanced filtering and bias calculation options see [below](https://observablehq.com/@pavlos/ai4netmon-ripe-atlas-extension#cell-706).","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":595,"value":"$31","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":781,"value":"

Do you see your ASN in the top positions of the following list?
Apply for hosting a probe in your network [https://atlas.ripe.net/get-involved/become-a-host/](https://atlas.ripe.net/get-involved/become-a-host/)

","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":51,"value":"viewof searched_data = Inputs.search(data_with_sum_bias, {label: 'Filter by: ASN, RIR, or Country'})","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":34,"value":"$32","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":798,"value":"downloadbuttonCSV(table)","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":706,"value":"$33","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":4,"value":"### Select the way you calculate the bias score [optional]\n\nFor each group of dimensions, select the weight (a value from 0 to 1). The TOTAL bias difference score is calculated as \n\n``TOTAL bias = weight_location * location_bias + weight_net_size * net_size_bias + ...``\n\nBy default we take into account all dimensions with the same weight (equal to 1). If you want to neglect a dimension, assign its weight to 0. ","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":436,"value":"viewof form_weights_bias = Inputs.form({\n \"Location bias\": Inputs.range([0, 1], {label: \"Location bias\", step: 0.1, value: 1}),\n \"Network size bias\": Inputs.range([0, 1], {label: \"Network size bias\", step: 0.1, value: 1}),\n \"Topology bias\": Inputs.range([0, 1], {label: \"Topology bias\", step: 0.1, value: 1}),\n \"IXP-related bias\": Inputs.range([0, 1], {label: \"IXP-related bias\", step: 0.1, value: 1}),\n \"Network type bias\": Inputs.range([0, 1], {label: \"Network type bias\", step: 0.1, value: 1}) \n})","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":631,"value":"### Filtering options [optional]\n\nBelow you can select different options to filter out some ASes from the table, by selecting those that \n- have a characteristic with larger/smaller value than a threshold; by default this filter is not activated (\"None\"). The characteristics are\n - #neighbors, or #ASNs/#prefixes in customer cone\n- have a specific characteristic (e.g., are in PeeringDB); filter out them by selecting the respective checkbox\n- belong to a given RIR region or a country (or select by AS number)","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":228,"value":"viewof form_nb_neighbors = my_range_form(\"#neighbors (total)\", 1)","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":382,"value":"viewof form_cone_asns = my_range_form(\"Customer cone (#ASNs)\", 1)","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":565,"value":"viewof form_cone_pfxs = my_range_form(\"Customer cone (#prefixes)\", 1)","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":290,"value":"viewof checkboxes_conditions = Inputs.checkbox([\"Is in PeeringDB\", \"Peers at IXP(s)\"], {label: \"Filter by condition:\"})","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":698,"value":"## Appendix","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":626,"value":"### Table with summary statistics for bias difference scores per dimension\n\nThe following table shows the distribution (histogram) of the bias difference scores among the considered (filtered) ASes. It can help to understand the values, such as, what is the room for improvement of bias (e.g., min/max values, or number of ASes with values less than threshold), or whether a dimension is more dominant that others (e.g., to weight it), etc.","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":485,"value":"viewof summary_data = SummaryTable(filtered_data_only_bias_columns, {label: \"Summary statistics for bias scores\", format: d3.format(\",.4f\")})","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":643,"value":"## Code appendix\nThe following sections correspond to the code that is needed to calculate/present the above results. If you do not want to change it, no need to look at it.","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":510,"value":"// Create a view of the summary statistics\nfiltered_data_only_bias_columns = {\n let arr = [];\n for (let j=0; j v[colname]));\n let minv = Math.min.apply(Math, data.map(v => v[colname]));\n return [minv, maxv];\n}\n","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":457,"value":"calculateSumBiasSelectedDimensions = function(data_obj) {\n let current_sum = 0;\n for (let cn in data_obj) {\n if (cn in form_weights_bias) {\n current_sum = current_sum + form_weights_bias[cn]*data_obj[cn]; \n }\n }\n return current_sum;\n}","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":45,"value":"data = FileAttachment(\"bias_df__no_stubs_Atlas_groups.csv\").csv({typed: true})\n// data = FileAttachment(\"bias_df__no_stubs_RIS_groups@6.csv\").csv({typed: true})","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":796,"value":"import {downloadbuttonCSV} from \"@pavlos/download-buttons\"","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":489,"value":"import { SummaryTable } from \"@observablehq/summary-table\"","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"}],"files":"$W35","stdlib":"1","titling":"auto"}} 30:"$2f:body"