Skip to content

Latest commit

 

History

History
119 lines (85 loc) · 3.88 KB

speedtest.md

File metadata and controls

119 lines (85 loc) · 3.88 KB
Raw

threefive + Python3 VS. threefive + pypy3 VS. tsduck in C++ VS. cuei in Go.

Q. Why only use threefive, tsduck and cuei? What about other SCTE-35 parsers?

A. To be honest, they are very few other parsers that will parse SCTE-35 from MPEGTS, here are three that were excluded and why.

  1. lib BitStream from the VLC folks. Fast parser, but incomplete SCTE-35 Data last time I used it, and it requires writing about 500 lines of code to parse SCTE-35.
  2. gstreamer I believe it will parse SCTE-35, but I spent a few hours trying to figure out how to do it and I have no idea how to even run gstreamer, much less parse SCTE-35 with it.
  3. Comcast GOTS has about 13,000 lines of code, and requires you write about 300 more to parse SCTE-35, and I'm not exactly sure how to use it.

The Contestants

  1. the threefive cli tool running on python3.
  2. the threeefive cli tool running on pypy3.
  3. the tsduck cli using the SpliceMonitor plugin.
  4. the cuei library and ten lines of code to make a cli tool.

The File

a@slow:~/cuei$ ls -alh ../v2.ts
-rw-r--r-- 1 a a 4.9G Mar  8 04:03 ../v2.ts
  • I used SuperKabuki to insert 14,862 SCTE-35 Time Signals

Test 1: threefive with Python 3.11.2

a@slow:~/cuei$ time threefive  ../v2.ts 2>&1 | grep sap_type | wc -l
14862

real    0m8.600s
user    0m8.026s
sys     0m0.778s
  • We grep 'sap_type' because it appears in every SCTE-35 Cue, and only once, so we can count the number of cues detected.
  • 'sap_type' is proper var naming in python.

Test 2: threefive with PyPy 7.3.11

a@slow:~/cuei$ time threefivepypy3  ../v2.ts 2>&1 | grep sap_type | wc -l
14862

real    0m3.543s
user    0m2.880s
sys     0m0.846s
  • We grep 'sap_type' because it appears in every SCTE-35 Cue, and only once, so we can count the number of cues detected.

Test 3: tsduck 3.36-3528 written in C++

a@slow:~/cuei$ time  tsp -I file ../v2.ts -P splicemonitor -a -O drop | grep "Command type:" | wc -l
14862

real    0m1.435s
user    0m2.006s
sys     0m0.861s
  • We grep 'Command type' because tsduck is not up to date with the specification and doesn't have sap type yet. 'Command type' appears in every tsduck SCTE-35 Cue, and only once, so we can count the number of cues detected.

Test 4: cuei v1.1.93 "Junior" written in Go

a@slow:~/cuei$ time ./cuei ../v2.ts | grep SapType | wc -l
14862

real    0m1.048s
user    0m0.605s
sys     0m0.509s
  • The code used for the test.
package main                          // 1

import (                              // 2
        "os"                          // 3    
        "github.com/futzu/cuei"       // 4
)                                     // 5

func main(){                          // 6
        arg := os.Args[1]             // 7
        stream := cuei.NewStream()    // 8
        stream.Decode(arg)            // 9
}                                     // 10  lines
  • We grep 'SapType' because it appears in every SCTE-35 Cue, and only once, so we can count the nunmber of cues detected.
  • 'SapType' is proper Go var naming.

Results:

tool Cues Detected GB/second
threefive w/ python3 14862 0.5697674
threefive w/ pypy3 14862 1.3830087
tsduck 14862 3.4146341
cuei 14862 4.6755725

Winner: cuei v1.1.93 "Junior"

parsing over 4.6 GigaBytes per second. I haven't even optimized cuei for speed yet, it will go even faster.

ADDENDUM:

threefive didn't win, but even at the slowest, it's parsing over 500 MegaBytes a second, high quality 4K video is about 11 MegaBytes a second.