<html>
<head>
<meta http-equiv="content-type" content="text/html; charset=ISO-8859-1">
</head>
<body bgcolor="#FFFFFF" text="#000000">
<pre>ref <b>Ulrik Dickow</b> <b> log</b>
> 150s to download 118s with BUFX enabled (many backwards jumps)
> 41s to download 99s with BUFX disabled (-R, no jumps at all)
41s for 99s file seems according to "old rule of thumb" of getting 2x speed "at full speed".
I assume the servers are-were set up for 2x "bursts" by simply limiting max feed to 2x?
However, most CSPAN servers and other "BBC-like" can do, and do, 4-8x speed (with the BUFX hack), have been doing that for some years.
Without jumpbacks it means one gets a 100s clip in 25 or less seconds
Also seems most servers (like CSPAN) now are starting with some higher initial Nx speed-max "to fill buffers" but then slow down to more nominal.
Anyway, my point is that with CSPAN, without -R, ie with jumpbacks, I get a file which can be 2-3x in SIZE!!
I assume that is also (at least) the amount of (excessive) traffic.
Goes somewhat well with my observation that I get a jump-back of appr 2.75/2s every 4x5.5s (with server willing to do maybe 4x).
Often even longer, repeated jump-backs, jump-jump-jump-backs
I.m fairly sure I have observed (and watched videos with) "multiple jump-backs inside each other".
But that is probably because the BUFX hack "remembers a timestamp" from an already jumped-back videochunk">
That is, the "odd" things I observed by following only the EOFs and 20k-packets with BUFX enabled.
It looked like BUFX first "asked" for the actual latest timestamp (but got an earlier) and then, often very soon, "in pairs",for an even earlier timestamp.
ANyway, my point that the amount of "excessive traffic" can be very high, although in theory only..hmm..
2.75/2s every 5.5s? 25%?
But additionally
- a 2x speed seems "OK" for the (old)servers.
- more than that probably only uses, or abuses, this (new?) "intro-super-speed" to fill buffers.
It seems possible that abusing this 4-8x intro-speed wld start a "rtmpdump-war" from the "server-people"
That is, I would suggest some kind of "semi-friendly" function, "adaptive Nx speed", where N is max 2-3?
(this started from the server becoming "adaptive" or "predicting")
PS Note, for one (automated, looped) rtmpdump version I did not close the stream, connection, session properly
(stopped, next one, when I found out, from the meta data, that I alrdy had the video).
My IP number got blacklisted after some "automated nights"! (the local "BBC").
That is, I started some 100 or more streams, but left most "hanging", non-properly closed after only some 2-4 seconds
</pre>
</body>
</html>