microsoft / usbroadbandusagepercentages Goto Github PK

The zip code to county matching in broadband_data_zipcode.csv seems a bit erroneous. There are duplicates of zip code by county which initially seems ok because zip codes don't follow county boundaries. But, for example, 56318 shows to be in 27153 (Todd County) and 27111 (also Todd County but should be Otter Tail County) but in reality it is in 27153 (Todd County) and 27097 (Morrison County). Perhaps your lookup is coming from a kind of geoip location but sometimes the county seems to be wrong for a zip code. Perhaps you meant this to be a distinct list of postal codes and the duplicates are unintended. They only exist in AK, MN, and SD.

Please add "n", the number of samples/tests, for each line in the data.

The paper suggests that the test is able to estimate aggregate connection speeds but is silent on factors influencing & constraining the measurements. For example, does the test sense whether other users are concurrently using the connection and thereby reducing available per-user or per-device capacity (and not record a result in the presence of competing traffic)? Does the test record whether the device was on WiFi or ethernet and can the dataset be separated based on connection type? For WiFi connections, is the 802.11 protocol, WiFi SNR or number of other clients on the WiFi network recorded in the test, potentially enabling filter out older/slow WiFi standards, devices at the edge of coverage, etc.? What about device type and device capabilities (e.g. do you record if it is a 15 year old laptop that cannot exceed 10 Mbps)?

Reference: https://cacm.acm.org/magazines/2020/12/248801-measuring-internet-speed

Microsoft's Delivery Optimization feature allows users to "Limit how much bandwidth is used for downloading updates". Users can limit "Absolute bandwidth" used for downloads or they can limit the "Percentage of measured bandwidth" used. Is Microsoft able to discern which measurements included within its data pool were constrained by user self-imposed Delivery Optimization throttling, which is facilitated by Microsoft itself, and is that erroneous and misleading data removed from the data pool?

https://docs.microsoft.com/en-us/windows/deployment/update/waas-delivery-optimization

Does Microsoft have sufficient server capacity to deliver content that saturates every single access line of connections that it is severing software updates to?
Are Microsoft's servers geographically distributed to optimize update distribution?
Is routing optimized for end-user experience and performance or to minimize Microsoft's costs (since this is a cost center)?
Does Microsoft have sufficient interconnection capacity to support surges in software update demands?

The paper does not mention the number of TCP connections used in conducting the tests. It is well-known that using a single connection will tend to underestimate aggregate connection speed so it seems to be important to share more details on how the underlying test works.