############################################################################### # iCellSpeed Dataset Release (MobiCom'20) # # MI-LAB: http://http://milab.cs.purdue.edu/ # ############################################################################### This README is used to introduce our released datasets by our MobiCom'20 work: “iCellSpeed: Increasing Cellular Data Speed with Device-Assisted Cell Selection”. We have conducted measurement study in several cities including Lafayette/West Lafayette (IN), Austin (TX) and Los Angeles (CA) with four top-tier US carriers: AT&T, T-Mobile, Verizon and Sprint. We run our experiments through MI-LAB testbed (http://milab.cs.purdue.edu/). In particular, we run three distinct types of experiments for three purposes: (I) Network downlink speed test at static locations and while driving along specific routes. We pick about 30 locations and 10 routes at different cities and let the mobile devices run heavy traffic flows (here, downloading a 500MB file from the lab server). (II) Network cell deployement measurements. We drove along all main roads across West Lafayette, IN and let mobile devices run light traffic flows (Ping Google every second). (III) Evaluation of iCellSpeed. We let mobile devices run file downloading, DASH video streaming and Zoom video conferencing. Both type-I static and type-I driving experiments are performed via the ApplicationOnTheGo task in MI-LAB (with specific configuration for 500MB file downloading) and was primarily performed from Aug 25 2019 to March 13, 2020. For type-I static, at every static location spot, we conducted experiment in both default networking settings and controlled networking settings with iCellSpeed or swiching airplane mode for comparison. For type-I driving, most experiment traces are collected under default networking settings except particular routes with iCellSpeed enabled or switching airplane mode. For type-II experiment, we release the extra experiment collected on March, 2020 using Sprint at Lafayette, IN. The driving experiments are performed vai the MMLabv2 task in MI-LAB (with continously sending PING packet to Google server every second). Note, we have already released results colected before 2020 in our HotMobile2020 dataset release (http://milab.cs.purdue.edu/hotmobile2020_release/). For type-III experiment, we evaluate the application performance with/without iCellSpeed. The experiments are performed in June 2020. File downloading: download 500 MB file from our server. DASH video streaming: https://bitmovin.com/demos/stream-test?format=hls&manifest=https%3A%2F%2Fcdn.bitmovin.com%2Fcontent%2Fdemos%2F4k%2F38e843e0-1998-11e9-8a92-c734cd79b4dc%2Fmanifest.m3u8. Video conferencing: using Zoom desktop client with USB tethered cellular network. As a result, we have three datasets: D1) type-I experiments in the static/driving tests (heavy load tasks via ApplicationOnTheGo). D2) type-II experiments in the driving tests (light load tasks via MMLabv2). D3) type-II evluation experiments (file downloading, DASH video streaming and Zoom video conferencing). Two datasets (D1 and D2) in our MobiCom’20 dataset contains logs collected in the experiments for 921.4 hours and over 11,066 Km. 1) Structure of files ├── D1_driving │ ├── West Lafayette │ │ ├── data_310120.csv │ │ ├── data_310260.csv │ │ ├── data_310410.csv │ │ └── data_311480.csv │ ├── Lafayette │ │ ├.... │ ├── LosAngeles │ │ ├.... │ ├── Austin │ ├.... │ ├── D1_static │ ├── West Lafayette │ │ ├── ATT │ │ │ ├── location_01 │ │ │ │ ├── default │ │ │ │ │ ├── data_***_001.csv │ │ │ │ │ ├── data_***_002.csv │ │ │ │ │ ├.... │ │ │ │ │ │ │ │ │ ├── control │ │ │ │ │ ├── data_***_001.csv │ │ │ │ │ ├── data_***_002.csv │ │ │ │ │ ├.... │ │ │ ├.... │ │ │ │ │ ├── T-Mobile │ │ │ │ ├.... │ │ └── Verizon │ │ ├── ... │ ├── Lafayette │ │ ├.... │ ├── LosAngeles │ │ ├.... │ ├── Austin │ ├.... ├── D2_light │ ├── Lafayette │ ├─── data_310120.csv ├── D3_evaluation │ ├── eval_file_download │ │ ├── ATT │ │ │ ├── location_01 │ │ │ │ ├── default │ │ │ │ │ ├── ***.csv │ │ │ │ │ │ │ │ │ ├── control │ │ │ │ │ ├── ***.csv │ │ │ ├.... │ │ │ │ ├── eval_dash │ │ ├── ATT │ │ │ ├── location_01 │ │ │ │ ├── vbr_default │ │ │ │ │ ├── ***.csv │ │ │ │ │ │ │ │ │ ├── vbr_control │ │ │ │ │ ├── ***.csv │ │ │ │ │ │ │ │ │ ├── cbr_4K_default │ │ │ │ │ ├── ***.csv │ │ │ │ │ │ │ │ │ ├── cbr_4K_control │ │ │ │ │ ├── ***.csv │ │ │ │ │ │ │ │ ├.... │ │ │ │ ├── eval_zoom │ │ ├── ATT │ │ │ ├── location_17 │ │ │ │ ├── default │ │ │ │ │ ├── ***.csv │ │ │ │ │ │ │ │ │ ├── control │ │ │ │ │ ├── ***.csv 2) Dataset release and its Descriptions (data*.csv) ------------------------------------------------------------------------------- Dataset 1 heavy_load: static/driving experiment ------------------------------------------------------------------------------- This is the dataset for driving/static tests. The mobile device keeps downloading a large file from our lab server. We record its serving cells set and instant throughput (per second). 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 2 light_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 ------------------------------------------------------------------------------- Dataset 3 evaluation experiments ------------------------------------------------------------------------------- This is the dataset to evaluate the performance improvement of iCellSpeed. Eval_DASH_CBR file format: Stall/Play: Ratio of stall/play during video playing per pause (when player buffer is empty) Eval_DASH_VBR file format: Video bit rate: Video bit rate (kbps) selected by Bitmovin Player (608, 878, 1228, 1628, 2128, 3128, 4728, 6128, 8728, 11728, 50128) Eval_Zoom Data file format: send latency: latency of video frames sent out receive latency: latency of video frames received send jitter: jitter of video frames sent out receive jitter: jitter of video frames received send resolution: resolution of video frames sent out receive resolution: resolution of video frames received