1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

SDV FAQ

Discussion in 'TiVo Series3 HDTV DVRs' started by bdraw, Jul 3, 2007.

  1. Dec 6, 2007 #761 of 2401
    Luke M

    Luke M Member

    371
    0
    Nov 5, 2002
    Note that it isn't average demand that matters, but peak demand. A box which is tuned to a switched channel at 3AM isn't a problem, since the bandwidth would otherwise be wasted.
     
  2. Dec 6, 2007 #762 of 2401
    morac

    morac Cat God

    8,946
    22
    Mar 14, 2003
    NJ
    Actually back in the day when most people used dial up modems, this started to become a problem because people would tie up the phones lines for hours or even days at a time.

    That's the problem when a system isn't used the way it was designed to be used. SDV was designed to be used by humans who don't watch TV 24 hours a day. A TiVo basically does this and watches 2 channels at once to boot.
     
  3. Dec 6, 2007 #763 of 2401
    morac

    morac Cat God

    8,946
    22
    Mar 14, 2003
    NJ
    There's nothing stopping the box from sitting on a switched channel during a high peak demand time. If there are a bunch of TiVo's on a node and they are all tuned to a different SDV channel sometime during the day, the SDV channel pool could run out come prime-time even though no one is watching.

    I would like to think that the tunable SDV frequencies could be adjusted to take into account high demand. For example borrowing an extra "channel" normally used for VOD and temporarily reassigning it for SDV. I'm not sure if that's possible with the current implementation though.
     
  4. Dec 6, 2007 #764 of 2401
    mikeyts

    mikeyts Stream Warrior

    2,390
    2
    Jul 10, 2004
    San Diego,...
    Humans don't watch television 24 hours a day, so if you take steps that prevent TiVo from doing it it won't affect humans.

    Note that leased Cable DVRs are nearly as bad as TiVo; you can place them in a standby state that TiVo no longer seems to support, and some firmware will automatically stop buffering the non-viewed tuner if it's not used for a recording for a few hours to save wear and tear on the HDAs, but many people aren't going to bother to turn on standby mode.

    Not to worry--use of bandwidth for channels tuned only by unviewed tuners can and will be remedied by schemes like channel leasing as I described a couple of posts back. SDV will work.
     
  5. Dec 6, 2007 #765 of 2401
    morac

    morac Cat God

    8,946
    22
    Mar 14, 2003
    NJ
    The thing about leasing schemes for always connected/on devices is that there is normally one lease per device. For example, I'm not using my cable modem 24 hours a day, but when I go to use it I'm guaranteed to get an ip address from the DHCP server because there is enough ip addresses for all the modems in my area to be connected at the same time.

    What you are talking about is a scheme where if a device is determined to not be in use, it's lease is revoked. This works fine until someone tunes a SDV channel and someone else who may be watching a channel but hasn't interacted with his TiVo for a while finds his lease revoked and he's staring at a black screen.

    What it really comes down to, how much legroom the cable companies are willing to give when it comes to the amount of bandwidth they assign to SDV.

    I'll give a good example. I used to work at a company where we all used Rational products on a server. The company paid for 30 licenses, figuring that there wouldn't be more than 30 people working with Rational at one time. There were more than 30 people at the company. Well it wasn't long before we started getting "out of license" errors when trying to use the software, even with a 60 minute lease timeout. It was obvious to everyone that more licenses were needed. Everyone except management who suggested that if we ran into this issue to go find someone with a lease who wasn't using the software and ask them to release their license. Needless to say productivity wasn't especially high as a result of this.
     
  6. Dec 6, 2007 #766 of 2401
    dswallow

    dswallow Save the ModeratŠ¾r TCF Club

    50,975
    31
    Dec 3, 2000
    Long...
    The odds may still work out OK as suggestions are generally recorded based on viewing habits and there's still a reasonable chance someone else needs the same station... might even be several other TiVo users on the node needing it for a suggestion. ;)
     
  7. Dec 6, 2007 #767 of 2401
    morac

    morac Cat God

    8,946
    22
    Mar 14, 2003
    NJ
    Unless someone has some really weird viewing habits or his TiVo thinks he does. ;)
     
  8. Dec 6, 2007 #768 of 2401
    dswallow

    dswallow Save the ModeratŠ¾r TCF Club

    50,975
    31
    Dec 3, 2000
    Long...
    Wow... I can't believe that article is over 5 years old.
     
  9. Dec 6, 2007 #769 of 2401
    mel.simmons

    mel.simmons New Member

    24
    0
    Mar 17, 2002
    Del Mar CA
    You are correct that the shape of tail is very important in how many channel requests are blocked. I used the values in my post above (169 channels, 334 tuners, 100 QAM streams) in a Monte Carlo simulation (with 1,000 trials). I used a Zipf law distribution for channel popularity with a power law slope of -1. The BigBand study (http://www.bigbandnet.com/index.php/tech_whitepaper_statswb.html) actually finds slopes of -0.64 to -0.74, which would make the tail longer, and make blockages more frequent. However I am not convinced by their fitting to the data, and I'm willing to assume a slope of -1 until better data shows up. The Motorola white paper you cited did not have any numbers on their graph of channel popularity, but it looked similar to a Zipf distribution.

    With these values, my simulation predicts that 22% of channel requests will be blocked. For the frequency of blocks to be less than 1%, 145 QAM streams would be required to deliver the 169 channels. Or, as noted above, if most of the tuners are off most of time, the situation is greatly improved.

    In this analysis, it is very important to know the shape the curve of channel popularity. It's tempting to guess that it follows a Zipf law, because so many phenomena seem to look that way (http://www.nslij-genetics.org/wli/zipf/). But I have not seen published data adequate to support that. Perhaps the cable companies have good data? Or perhaps TiVo has good data on their user's viewing habits?

    I would be very interested in knowing whether TiVo usage is like human usage. Or does the search capability of TiVo lead to more frequent recordings from "unusual" channels in the long tail?
     
  10. Dec 6, 2007 #770 of 2401
    mikeyts

    mikeyts Stream Warrior

    2,390
    2
    Jul 10, 2004
    San Diego,...
    And this is a problem...how??? TiVo can either be programmed to automatically revert to some known non-SDV channel or to display a "Please select a channel to watch" message. In the scheme that I described, this is only gonna happen for unused, unrecorded background tuners and for displayed foreground tuners where (1) an SDV channels was tuned, (2) the channel hasn't changed in a very long time and (3) there was a failure to respond to a pop-up "are you still watching this" query. In other words the lease was dropped because no one was watching or recording a tuned SDV service, which is a perfectly reasonable thing to happen. Hell, if someone touches the remote, it can automatically re-request the last channel tuned; the message can read "Standing by, please press INFO to view the last channel." When the user returns, it's quite possible that the lease count never reached zero and there are still others using it so it's still present in the same transport stream as the same program and can be brought back up instantly; if not and there's a couple of seconds wait then BFD.

    No one requires that SDV be completely transparent. Access to another 100 HD channels should be worth a slight delay here and there.
     
  11. Dec 6, 2007 #771 of 2401
    ah30k

    ah30k Active Member

    2,211
    0
    Jan 8, 2006
    You tossed out a lot of numbers and theorems but something must be wrong in your simulation because 22% blockage seems WAY off base. I'd have to study your model a bit to determine whether or not I put any faith into it.
     
  12. Dec 6, 2007 #772 of 2401
    ah30k

    ah30k Active Member

    2,211
    0
    Jan 8, 2006
    Mel,
    The BigBand simulation you hyperlinked found that
     
  13. Dec 6, 2007 #773 of 2401
    MichaelK

    MichaelK New Member

    7,308
    0
    Jan 10, 2002
    NJ

    the neilson ratings make it pretty clear too for anyone looking for specifics- people can just look up the primetime ratings and it quickly becomes clear that beyond the top 20-30 channels the odds of anyone in a 500 home node wanting them drops off dramatically.

    Granted that's national but I'd suspect that in any one neighborhood of 500 homes it would be similar- people tend to leave in clusters of similar cultures, socioeconomonics, lanuage, sports markets, etc ,etc

    also once you go all digital or OCAP and allow SDV for tivo's and other DVR's (media center pc's). The only people left out would be early cablecard TV buyers and those numbers are tiny and the fraction of them that use cards is even tinier. You could in theory get to that point of 300 tuners per node and then go all SDV and allow unlimited. (statistically I'm sure you can be somehwat bigger then 300 homes)
     
  14. Dec 6, 2007 #774 of 2401
    MichaelK

    MichaelK New Member

    7,308
    0
    Jan 10, 2002
    NJ
    I could be mistaken but I beleive on of the officail documents- either an FCC filing, or something like it specifically talkes about the leasing thing- or at least the fact that the dongle would need to 'give back' a channel after a certain period of time. I think it said tivo was to notify when 'done' or something and there was a discussion of if tivo would count suggestions as being done or not.
     
  15. Dec 6, 2007 #775 of 2401
    MichaelK

    MichaelK New Member

    7,308
    0
    Jan 10, 2002
    NJ

    and that peak time is called primetime and that's why the neilsen primetime ratings are a good place to look for the usage patterns that exist.
     
  16. Dec 6, 2007 #776 of 2401
    MichaelK

    MichaelK New Member

    7,308
    0
    Jan 10, 2002
    NJ
    SDV isn't really about the fact that everyone isn't watching all the time- it's that even during peak demand the majority of channels are hardly being watched- so you can leverage that.

    Even if Tivo's run suggestionms 24/7 on 2 tuners the fact that suggestions record some statiistic version of what people most like to watch means the tivo tuners as a whole will still likely be tuned to the same 30-50 channels that get the majority of viewing.
     
  17. Dec 6, 2007 #777 of 2401
    ah30k

    ah30k Active Member

    2,211
    0
    Jan 8, 2006
    Depends on how you define peak. Primetime would likely have the most common viewing pattern and thus the fewest number of unique channels. Alternatively, off hours would likely have the most likelyhood of having strange (?) channels being watched.

    I'd love to get the raw data and plot the usage curves dynamically over time (like a moving weather radar map).
     
  18. Dec 6, 2007 #778 of 2401
    MichaelK

    MichaelK New Member

    7,308
    0
    Jan 10, 2002
    NJ

    wow you seem to be really well versed in the math- can you try to find some ratings data and plug that in to your model- I'd love to see real world data plugged it.
     
  19. Dec 6, 2007 #779 of 2401
    MichaelK

    MichaelK New Member

    7,308
    0
    Jan 10, 2002
    NJ
    obviously i think the statistics work for the vast majority of cases.

    BUT- someone pointed something out in another thread- a close to real world situation that could completely upend the apple cart.

    what happens when windows media center pc's get upgraded with the patch for the dongle? And 2 guys on the block decide to have a testosterone battle over who has the best windows DVR- one makes an 6 tuner dvr and the other makes an 8. Now what if those 2 guys are oddballs and watch lots of stuff out in the tail? They rob 14 slots from everyone else or like 10% of the needed slots. THat can get ugly.

    if the tivo nero pc software thing turns out to be cool- I just might be one of the oddballs on the block building the silly sized pc dvr...
     
  20. Dec 7, 2007 #780 of 2401
    mel.simmons

    mel.simmons New Member

    24
    0
    Mar 17, 2002
    Del Mar CA
    Yes, I would agree the 22% blockage seems very high. This blockage results from the assumption that all 334 set-top boxes are turned on at the same time. The BigBand study found that only about 1/4 of the set-top boxes on a node were on at any moment (in trial A, this is 150 boxes out of 603 installed). If I rerun my simulation with only 150 tuners turned on, the blockage drops to less than 1%. This takes us back to the point in my previous posting: if TiVo never turns off (or releases its channels), it could have a severe impact on SDV performance.

    I apologize to other readers for all the mathematics in these postings. Perhaps I can summarize my key points:

    - SDV is a complicated statistical game that trades off cost (to the cable company) and quality of service (to the consumer).

    - Designing and managing SDV systems will be complex.

    - If they are not careful, cable companies could get this very wrong.

    - SDV depends upon patterns of consumer behavior that are uncertain and subject to change.

    - The performance of SDV will be altered by new consumer-side technologies, including the TiVo dongle.
     

Share This Page