########################################################################################## ## HotMobile'20 Dataset Release # # MI-LAB: http://http://milab.cs.purdue.edu/ # # ########################################################################################## This README is used to introduce our released datasets by our HotMobile'20 work: “Unveiling the Missed 4.5G Performance In the Wild”. We have conducted a city-scale measurement study in West Lafayette, Indiana, US with four top-tier US carriers(AT&T, Sprint, Verizon, T-Mobile). We run our experiments through MI-LAB testbed (http://milab.cs.purdue.edu/). In particular, we run two distinct types of experiments: (I) let the mobile devices run heavy traffic flows (here, downloading a 500MB file from the lab server), and (II) let mobile devices run mice flows (here, ping google every second). The type-I experiments are performed via the App2Go task in MI-LAB (with specific configuration for 500MB file downloading) and was primarily performed from Aug 25 to Oct 15, 2019 (Later more runs are performed, but not included in HotMobile’20). We conducted these experimetns in both both static tests at selected locations and driving tests across the whole city. The type-II experiments are run via MMLABv2 task in MI-LAB. It is mainly performed in the driving tests, before Oct 2019. As a result, for the city of West Lafayette, we have three datasets: D1) type-I experiments in the static tests (heavy load tasks via App2Go); D2) type-I experiments in the driving tests (heavy load tasks via App2Go); and D3) type-II experiments in the driving tests (light load tasks via MMLABv2). Three datasets (D1,D2 and D3) in our HotMobile’20 dataset contains logs collected in the experiments for 740 hours and over 8,756KM. After Oct 2019, we continued to run more experiments to extend this work. As of January 15, 2020, the new datasets (released here) are collected in the experiments for 809 hours and over 10,190 km. 1) Structure of files ├── heavy_load │ ├── driving_experiment │ │ ├── data_310120.csv │ │ ├── data_310260.csv │ │ ├── data_310410.csv │ │ └── data_311480.csv │ │ │ └── static_experiment │ ├── ATT │ │ ├── location_** │ │ │ ├── data.csv │ │ │ ├── Sprint │ │ ├── location_** │ │ │ ├── data.csv │ │ │ │ │ ├.... │ ├── T-Mobile │ │ ├── location_** │ │ │ ├── data.csv │ │ │ │ │ ├── ... │ └── Verizon │ ├── location_*** │ │ ├── data.csv │ │ │ ├── ... ├── low_load │ ├── data.csv 2) Dataset release and its Descriptions (data*.csv) We release intermediate results (data*.csv) to reproduce our HotMobile'20 work which is performed in the city of West Lafayette, IN, USA. We have three datasets: D1) heavy-load experiments in the static tests. D2) heavy-load experiments in the driving tests, D3) low-load experiments in the driving tests -------------------------------------------- Dataset 1 heavy_load: static_experiment -------------------------------------------- This is the dataset for static tests. When the mobile device stays at a fixed location, it keeps downloading a large file from our lab server. We record its serving cells set and instant throughput (per second). The experiments are performed at the following 15 locations. Location GPS: name lat lon type Location01 40.42475 -86.91925 campus Location02 40.46125 -86.92475 suburban Location03 40.46725 -86.92325 suburban Location04 40.43075 -86.91375 campus Location05 40.4326167 -86.90517928 campus Location06 40.4310398 -86.9190431 campus Location07 40.4250299 -86.9274937 campus Location08 40.4242872 -86.9069842 campus Location09 40.424986 -86.9194766 campus Location10 40.4381676 -86.9424108 rural Location11 40.4249385 -86.9373129 rural Location12 40.4563182 -86.9476907 rural Location13 40.4638752 -86.9207795 suburban Location14 40.461058 -86.908222 suburban Location15 40.432538 -86.952462 rural Data file format: pcell_cid: physical cell ID of PCell pcell_freq: downlink earfcn of PCell scell_1_cid: physical cell ID of SCell1 scell_1_freq: downlink earfcn of SCell1 scell_2_cid: physical cell ID of SCell2 scell_2_freq: downlink earfcn of SCell2 scell_3_cid: physical cell ID of SCell3 scell_3_freq: downlink earfcn of SCell3 seconds_since_epoch: datetime in epoch format throughput: bits per secound past_seconds_since_rrc_request: past seconds since device receives an RRC request -------------------------------------------- Dataset 2 heavy_load: driving_experiment -------------------------------------------- This is the dataset for performance measurement in the driving tests. The only difference from Dataset 1 is that the GPS changes over time. In addition to the data recorded in Dataset 1, it also recordds the GPS of each location (grid). The whole area is divided to multiple grid (each approximately 55m x 42m). Data file format: grid_lat: latitude of grid (precision 0.0005) grid_lon: longitude of grid (precision 0.0005) pcell_cid: physical cell ID of PCell pcell_freq: downlink earfcn of PCell scell_1_cid: physical cell ID of SCell1 scell_1_freq: downlink earfcn of SCell1 scell_2_cid: physical cell ID of SCell2 scell_2_freq: downlink earfcn of SCell2 scell_3_cid: physical cell ID of SCell3 scell_3_freq: downlink earfcn of SCell3 seconds_since_epoch: datetime in epoch format throughput: bits per secound past_seconds_since_rrc_request: past seconds since device receives an RRC request -------------------------------------------- Dataset 3 low_load experiments -------------------------------------------- This is the dataset to learn the serving cell set in the drive tests. It runs a mice flow (keeps pinging Google), rather than the heavy file downloading to save the data usage. All the recorded data is the same as D2, except that no throughput is recorded. Data file format: grid_lat: latitude of grid (precision 0.0005) grid_lon: longitude of grid (precision 0.0005) pcell_cid: physical cell ID of PCell pcell_freq: downlink earfcn of PCell scell_1_cid: physical cell ID of SCell1 scell_1_freq: downlink earfcn of SCell1 scell_2_cid: physical cell ID of SCell2 scell_2_freq: downlink earfcn of SCell2 scell_3_cid: physical cell ID of SCell3 scell_3_freq: downlink earfcn of SCell3 seconds_since_epoch: datetime in epoch format past_seconds_since_rrc_request: past seconds since device receives an RRC request