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

Premiere Elite performance enhancements

Discussion in 'TiVo Premiere DVRs' started by brentil, Oct 10, 2011.

  1. aaronwt

    aaronwt UHD Addict

    19,173
    21
    Jan 31, 2002
    Northern...
    Just set the output to 1080i since most stations are 1080i. This is what I did with my Elite and it makes switching between tuners so much easier with sub 1 second changes. I ended up doing this with all my Premieres as well.
     
  2. brentil

    brentil New Member

    342
    0
    Sep 9, 2011
    Benchmarks!

    All tests were performed a minimum of 3 times, most were done 5. All runs are averaged while the outliers are ignored as is initial loads unless noted. Timing was done using a digital stopwatch and values were typically rounded to whole seconds unless I felt the 0.5s value was needed to indicate a feeling of being that bit faster.

    This image contains the data values in seconds comparing the previous 14.8c versus the new 14.9.2.2 along with the % difference.

    [​IMG]

    This image is a graph of the % increase showing the overall trend in increases.

    [​IMG]

    Aggregating this data ignoring the 0% values shows an average gain of 46% or nearly a 50% performance gain in the TiVo Premiere non-Elite with the new software that contains the new HDUI/Flash & 2nd core enabled. This is definitely a solid performance gain from primarily the 2nd core without a complete HDUI rewrite with a new multi-threaded Flash runtime.

    To me what this shows is that the main system was CPU limited for a majority of background tasks the UI relies on as the UI itself probably was as well. Providing true hardware threading the systems behind the UI (IO, database, network, etc) can now run more efficiently and finally concurrently.

    Oddly the moving of a Season Pass form bottom to top decreased in speed, where I had expected to see a solid improvement here. This task is possibly one of the most DB intensive in my opinion as it requires a significant amount of reorganization and validation. My guess would be the process needs to be optimized for more efficient memory/cpu use due to heavy thrashing.

    In terms of software improvements this is super solid gains. I can't wait to see what the v16 software and v2.0 UI bring to us!
     

    Attached Files:

  3. sbiller

    sbiller Active Member

    1,915
    0
    May 10, 2002
    Tampa, FL
    Brentil,

    Outstanding work! I thought the improvements would have been more modest but as you have surmised the gains are primarily the background tasks that were CPU/database intensive in the past. I always wondered why that connecting/preparing takes so long!

    This is great news. I'm really looking forward to seeing what kind of HDUI performance we see in Q1-2012 when 16.x is released.

    ~Sam
     
  4. BlackBetty

    BlackBetty I

    2,446
    0
    Nov 5, 2004
    feb-apr

    I won't hold my breath.
     
  5. crxssi

    crxssi Veteran TiVo User

    2,791
    0
    Apr 5, 2010
    Great work! Thanks so much!!!!! Have to go to work, but I will analyze it better later. One thing that would be nice would be to categorize things like I did in the change log- those tests related to the HDUI, those to the SDUI (which it looks like you didn't do) and those that are common to both or neither. (For example, things like launching Netflix/Youtube/Amazon, boot time, etc, are not attributable to the HDUI).
     
  6. socrplyr

    socrplyr Active Member

    1,076
    1
    Jul 19, 2006
    Brentil,
    Thanks for the benchmarks. I just wanted to verify that your column labeled % faster is actually some other measure. For example, when I compute the speed of the boot operation (to see how much faster it is), I get 1/450 and 1/240 according to your numbers. This leads to an increase in speed of 87.5%, not the 47% reported. This also makes sense with the second core being utilized well. I think the 47% must be more along the lines of the reduction in time.

    I think we all appreciate your hard work on this one.

    As for moving the season pass, I guess we should all be organizing them online. :)
     
  7. sbiller

    sbiller Active Member

    1,915
    0
    May 10, 2002
    Tampa, FL
    Interesting question. I had to think about this for a second and play with the numbers. I believe Brentil's calculations are correct.

    Brentil essentially used the formula, percent difference as ((q1-q2)/q2)*100. Since we were looking at a number that was going down (in most cases) the percent differences look smaller.

    If the boot time increased from 240 to 450 than we would have an 87.5% increase in boot time. Since the numbers went down from 450 to 240 we have a 46% decrease in boot time. It all depends on the denominator of the equation. When 450 is the denominator we have a smaller number for percentage difference.

    This would be the normal way to account for the difference.

    For more info -->

    http://mathforum.org/library/drmath/view/58083.html

    http://www.percent-change.com/
     
  8. socrplyr

    socrplyr Active Member

    1,076
    1
    Jul 19, 2006
    I agree that his computations are correct (just question the labeling). For me, if you refer to something as being some % faster, then you are computing the change in speed/rate at which it does something. If we are both in cars and I am driving 50% faster than you are, then my speed is 1.5x yours. If I say I will get there in 33% less time, then my speed is again 1.5x yours. They both represent the same reality. Maybe I was just being too picky about a subjective definition of faster. I just think speed/rate when the word faster is said, not time. Meh, oh well.
    Sorry for the unnecessary distraction.
     
  9. brentil

    brentil New Member

    342
    0
    Sep 9, 2011
    I did the math as sbiller mentioned. My label term might be not exactly correct for the math I used but I think most people will understand what it means.


    Yeah, I'm kicking myself for forgetting to go test the SDUI before I got the update...

    I'll reorganize my values so they're more intuitive to read, I finished this about 1am last night and was just too giddy to get data out. :D

    The problem with that is I've tried organizing them online but then when my TiVo hits in it erases all the changes to organization I made online and replaces it with what was on the TiVo. I haven't tested this since the 14.9 update though.
     
  10. crxssi

    crxssi Veteran TiVo User

    2,791
    0
    Apr 5, 2010
    OK, I spent more time studying the numbers you supplied. Again, thanks for your work, I know what it is like to spend a lot of time on this stuff ;) I did update the changelog thread: http://www.tivocommunity.com/tivo-vb...00#post8858900 a little bit based on what I saw.

    The things you tested are mostly unrelated to the user interface, but more on underlying processes. And on those, results you are showing are MOSTLY inline with what I expected with another core going on line. Linux will nicely use the extra processing power to increase the speed of things it more directly deals with, like: searching, season pass setup, shifting to live TV, deleting programs, possibly network transfer, booting, and some aspects of playback. So they have face validity. But I also can notice the performance changes on many of those tasks from just observation, which is further positive verification/confirmation.

    At first, I was a little surprised that the launching speed of the sub-modules, such as Netflix, Youtube, Amazon, etc showed no change. But after thinking about it a while, those are also Flash based, and the loading/switching itself is probably constrained by being a single thread.

    I would NOT, however, based on this data, and from what I have experienced/observed after the update, make a generalization that the HDUI has significant performance improvements. Of course, that is a qualitative analysis (*). It is unfortunate that the data does not have such things as "time to completely finish drawing" screen X, Y, or Z. Those are actually the type of actions users are likely to notice the most as lagging.

    Other UI tests would be perhaps too difficult to have even tested (like lag between keystrokes in menus). You do have TWO metrics in that genre, I think, the "Menu up/down" transitions. Possibly more, depending on your definition of "Load Details from My Shows" and "Load Folder" and "Info". I just can't seem to verify those through observation.

    (*) I would use the following analogy: Speed limit on the road is 55 MPH. You are in a lot of traffic and crawling at 10 MPH. Something changes and then you are finally going 15 MPH. That is a whopping 50% speed improvement! Taken alone, that would certainly be very significant. But will you really notice that much difference? Especially if you want to be going 55? This is how user expectations can very much skew perceived performance.

    As an example of specific HDUI testing: It takes my updated Premiere non-Elite in HDUI, 5 seconds just to finish drawing everything when I select "My Shows" from "TiVo Central" the first time, and 3.5 seconds the second time. It takes 2.5 seconds for the screen to finish drawing after highlighting a program from "My shows". It takes 5 seconds to finish drawing the screen if I then select that program. It takes 7 seconds to finish drawing the screen if I select to "explore" the program. I repeated each just to make sure. In each case, the right hand side of the screen was updated in about half to 3/4 the time, and the change of the discovery bar was last.
     
  11. bradleys

    bradleys It'll be fine....

    3,687
    4
    Oct 31, 2007
    Such a Scrooge...

    [media]http://crooksandliars.com/files/vfs/2011/09/scrooge.jpg[/media]
     
  12. crxssi

    crxssi Veteran TiVo User

    2,791
    0
    Apr 5, 2010
    Bah.... humbug....
     
  13. lessd

    lessd Active Member

    7,695
    5
    Jan 23, 2005
    CT
    My PC display in HD is very fast BUT the display board cost 2.5 times the total cost of the TiVo or to put it another way it cost 1/2 of a TP with lifetime, so how could TiVo HDUI compete with what some of us are use to, and i don't have the fastest display board I could get, my Windows 7 display rating is 7.4 out of a possible 7.9. It takes a lot of power to display HD graphics quickly.
     
  14. crxssi

    crxssi Veteran TiVo User

    2,791
    0
    Apr 5, 2010
    What makes the present HDUI inherently slow are design flaws in the way the interface is presented and updated and how it relies on live Internet-supplied data instead of locally cached information. It is presumably further crippled by lack of threading capability in the HDUI to take advantage of both cores. I do not think it is a limitation of raw computing power.

    Graphics card/chip speed not a useful comparison to a machine such as the TiVo... updating a 2D HD screen is not terribly difficult nor slow even with a puny GPU (or CPU). Where the GPU on a typical computer comes into play is 3D, texturing, massive vectoring/blitting and decoding video streams. The TiVo does no 3D, no texture mapping, no massive vectoring or blitting and video decoding is handled speedily by the chipset with almost no CPU load and with no performance issues at all.
     
  15. brentil

    brentil New Member

    342
    0
    Sep 9, 2011
    I agree with crxssi on this point for the most part actually, I know surprise! :D

    I think there is the possibility that the new UI & Flash code does contain some performance fixes in the HDUI but that would be mainly preventing the little hiccups we usually see. The HDUI speedup comes from 2 places.

    1 - Now that the 2nd core is enabled you can have 2 processes running up to 400 MHz each where before that 400 MHz was shared across ALL threads. So what you get now is the ability for the HDUI to solely reside on one core using up to 100% of that CPU before where before there was contention for resources which results in thrashing.

    2 - Now with the 2nd core enabled the tasks that were fighting the HDUI for CPU can now run at full speed themselves allowing them to feed the HDUI at a much faster rate. The HDUI relies heavily on disk IO, network IO, & database access which all had to fight with the HDUI itself before to run where now they have their own core to play.

    However now that the 2nd core is fully enabled this allows development of more advanced HDUIs especially if they can make the HDUI itself threaded. The main wait now when the HDUI is running is that all the data being loaded needs to realistically be loaded when you access it. If you have a heavily threaded UI you can pre-cache data based on histograms and expected behavior. If you know a user is paging down pre-cache the next 5 items before they get to it. If they 95% of the time go to Watch Videos and load Netflix pre-cache it before they load it.

    I really think the v16 code & v2.0 HDUI is where we'll see these types of improvements. As you can see from the v16 thread there are a significant number of changes revolving around the graphics engine as well as improved threaded processes. If they move to a newer Flash runtime that's more efficient as well then we can expect some solid gains.
     
  16. crxssi

    crxssi Veteran TiVo User

    2,791
    0
    Apr 5, 2010
    I am cautiously optimistic/excited about the v16 code. From what I saw of the videos, it did look like a huge improvement... Of course it is one thing to see a video of someone else using it, and another using it one's-self on one's own unit. I hope it doesn't take forever to roll out!
     
  17. brentil

    brentil New Member

    342
    0
    Sep 9, 2011
    I've dug in the kernel code and I know there's a lot of low level system stuff revision wise that will make the base subsystem a lot more modern with the possibility of some good gains. The biggest IF will be what they do about the HDUI in terms of what it runs on (either newer Flash runtime or moving to something like Java which I believe is what large portions of the Virgin box run) to actually translate those benefits to the user end.
     
  18. brentil

    brentil New Member

    342
    0
    Sep 9, 2011
    I updated the images in my post above to be more readable and also added them to the first post in the thread.
     
  19. crxssi

    crxssi Veteran TiVo User

    2,791
    0
    Apr 5, 2010
    I like it :)

    Although one could argue a case that "TiVo Button from live TV", "TiVo Button to My Shows", and "Live TV from TiVo Central" are also more like "Apps" (perhaps "modules"). I those cases, the user is not really navigating what we typically think of as the UI, but leaving the UI and performing the video playback function (or the reverse).

    Of course, one could make all kinds of arguments about just about anything ;)
     
  20. sbiller

    sbiller Active Member

    1,915
    0
    May 10, 2002
    Tampa, FL
    Brentil,

    Do you want to correct the first post to delete the errors that were discovered later in the thread? For example, we know that the memory is no 1GB in the Elite.

    Not a big deal but if someone reads just the first post they would have some misinformation.

    Thanks,
    ~Sam
     

Share This Page