search

hongzimao / pensieve

Neural Adaptive Video Streaming with Pensieve (SIGCOMM '17)
http://web.mit.edu/pensieve/
MIT License
516 stars 279 forks source link

Out of bounds error while implementing dp.cc #132

Closed naman-32 closed 2 years ago

naman-32 commented 3 years ago

Hi,

First of all great work! I really enjoyed the paper. And thanks for providing the implementation details publicly!

I am trying to reuse your dp.cc implementation in a slightly different context in project where format of video metadata file is different. So, I have tweaked dp.cc a bit to parse video metadata and hence populate the data pretty much in the same way as in original implementation.

The script successfully performs offline optimal simulation correctly on most of the cases, but for certain combination of video and trace file , I got a segmentation error at download_time = all_download_time[n][t][nm];. t is this case was detected to be t_max_idx +1 (or 2). I would be grateful if you could please spare some time to indicate towards possible sources of this error. Please comment if some other details are required from my end.

P.S. :- The error was temporarily resolved by setting _t_portion_ to be 1. And, I have matched all data units to the original script.

Thanks in advance, Naman

hongzimao commented 3 years ago

I think the segmentation fault means t overflows the total simulation duration. There might be some hardcoded horizon for the dp (we were manually setting it as the max across all traces we dealt with). When using a new trace, you may need to update those max (or truncating your trace files).

I think _t_portion_ have to do with when a video chunk is downloaded within a 2 or 4 seconds interval. Setting it to 1 may make the simulation jump in the trace file (e.g., some portion of the interval isn't used by the corresponding bandwidth).

Hope these help.

joybi531 commented 3 years ago

@naman-32 Hi, I was trying to do this experiment but unfortunately, I am not being able to get access to any of the raw data (zip file) and cooked_trace data. Do you have any of these two or both of them? I would really appreciate any support or information on where I can find the dataset.

Note: I have checked on the fcc.gov website but couldn't download the data due to some error.

naman-32 commented 2 years ago

Hi,

The cooked_trace data samples doesn't seem to be accessible now on link provided in the repository. I was working on different private cooked_trace dataset, which I cannot share due to NDA.

Best Naman