roku stalls and gets stuck in MB3

[dcauger 3]

MB3 V3.0.5441.2 (I same problem with V5441.1)

I have2x Roku 3  (both 4200s V5.6 build 60) with MB 1.72. Rokus are hard wired. They are on the same subnet as MB server (Win 2012 Essentials dual xeon proc with 8 cores, 24GB ram, 4x1Gb NIC teamed... plenty of horse power here). W2012 is clean, nothing in the event logs for system, hardware, or applications. MB does not crash. Even when doing lots of transcoding to multiple places, server never tops 20% or uses more than 12GB of ram. no hyper-v or VMware, its a physical deployment.

 

To the problem... and I've been struggling with this one for a while. Never posted on it. I've Tried several things to resolve.

 

On my Roku 3, when I sort my movie collection by genre and enter a genre type, then use my roku remote to move from one object to the next, roku will eventually get stuck. By stuck, I mean I am in the genre type and able to move from object to the next, left and/or right. I can not exit out of that genre type, enter into a specific movie folder, or even press the home button and get out of MB all together when . All I can do is scroll from left to right within the selected genre. If I put the remote down, go away for a while, then when I return I will be able to exit that genre, enter the movie, goto roku home, etc.

 

I've found that If I enable the "info box" option, the info box seems to display the item I was stuck on. If I am patient and press nothing, the info box will catch up to the object that I am on. Or If I have movie A, Movie B, Movie C, etc.. I could have movie C highlighted and the info box will show me content for movie A. If i put the remote down and do nothing but watch the screen, eventually the info box will catch up to movie C. When the info box and the highlighted movie are the same, I can exit out of the genre with the back button, enter the movie, press the home button and exit MB, etc. everything will work as expected. Once back in sync, If I move left or right, I will eventually get the info box and the highlighted movie out of sync, and then have to wait for the info box to catch up. It seems to help if I navigate back to the object the info box describes, however, I can not say if that is real or perceived.

 

My belief is that the info box is displaying the effect, I have no idea what the cause is.

 

If I try the exact same thing with roku on a TV series, it works fine. I can enter genres, navigate, enter folders, no issues. Info box and highlighted series are always equal. Sometimes there is a slight lag between the two, but that is seconds, not dozens of seconds or minutes.

 

I have just about as many movie objects as I do TV episode objects.

 

I've gone over my folder structure and naming conventions. renamed them to ensured both exactly matched the recommended MB structure (as do my TV series structures). I've cleared out all my meta data from my movie collection and re-fetched (I use Media Center Master). I allow MB to write to the folders. I don't have any other 3rd party metadata in the folders. Just what MCM and MB has created.

 

Strange enough, If I open a web client, internet explorer, chrome, or safari, I've tried all three (combination of wired and wireless) it works great. I can click on anything and search or filter however I wish. I can enter and exit genres, movies, etc. No problems. Same goes for my droid with a MB app (wireless). I am not using connect for any of this. 

 

I have even done a test side by side.. same content from two sources. test was movies->Genre->move to movie G->open. Roku gets suck and while the roku was sorting itself out, I was able to get to and start the movie on my droid. Actually, I was about 8 minutes into the movie before I was able to start playing it on the roku. I tried a similar test on the other Roku with a laptop using the web client. Very similar results.

 

Only my two roku 3 boxes have this problem, and it is identical on both which makes me believe its something to do with the channel, not the MB server.

 

I do have a roku 2 wireless sitting on a shelf. I bought that on Woot as an experiment to see how that worked compared to my WMC boxes. We liked the Roku but the Roku 2 was a bit laggy during playing (wireless). Roku 3s replaced my WMC boxes. I was debating hooking up the Roku 2 with the MB 1.72 release to see how that functioned. I have not done this yet.

 

Thoughts or direction?

 

Dean

 

 

 

[Luke 36992]

That's a tough one. I've never seen that or heard of it being reported before. Guess we'll have to keep an eye out for it.

[dcauger 3]

I've put a little time into this. It looks like the roku devices exhaust all of their TCP receive buffers often during folder navigation. At this time, I believe this is due to how the roku channel requests objects from the MB3 server and then how the MB3 server responds. There seems to be an issue with the MB3 server in how it interacts with the OS.

 

When the roku runs out of TCP receive buffers, objects in the MB3 log say they are delivered to the target IP address, but are not actually transmitted on the wire. When this happens, the Roku hangs until the session is tore down and rebuilt. Once the TCP 3-way handshake is complete, the process starts all over again and the Roku "unlocks" or picks back up where it left off. Visually, it's easy to see as the roku's media info window is out of sync with the movie highlighted. Once they get back in sync the session has been reset. All the while, the MB server never notices (or at least logs) there is a problem. It continues to try and deliver content.

 

I've only seen this with my roku devices (regardless of device generation). It happens regularly when I am navigating in a collection or genre. When I navigate the default folders, everything is fine (i.e. movies).. I can walk the movie folder without seeing this issue. when playing content, transcoding works exactly like I would expect to see a tcp session behave and is very stable.

 

I have to SPAN an interface on my switch to analyze from Roku's perspective to ensure what I am saying here is correct. Right now I only have traces from the server side. I've posted this a bit prematurely to ask a couple of questions..

 

Has anyone else mention anything like this? anyone notice or complain that they have an issue on roku navigating collections or genres?

 

When the Roku MB channel is in testing, how large is the dataset you are testing against? dozens of objects, 100s, 1000s? 

 

When in testing and navigating, and the roku requests content, what is the time to deliver the first image? How long from the delivery of the first image does it take to the delivery of the last?

 

DCA

[dcauger 3]

Done some more digging here. I don't think this is specifically a MB problem. I am going to turn to Roku. From the looks of it, there is nothing that they will be able to do. 

 

Don't read into this that there is some deficiency with the Roku product either. I would state the contrary. For what this little fella does at its price-point... Brilliant.

 

I believe I am pushing it too hard.

 

As I continue on, if I find something of interest, Ill continue to post on this thread.

 

DCA

[Luke 36992]

you can also try the dev server because some advances have been made to trim api output. just make sure it's for testing only so that you're not impacted if we need to temporarily break something during the dev cycle (see the guidelines in the alpha area). in addition the server has new features to trim the output even more, so i'll be adding updates to the app to take advantage of them.

[dcauger 3]

Update (solved):

 

Pulled some more traces. What I saw lots of TCP receive window zero on the roku, lots of dropped packets, then congestion notification/mitigation steps within TCP. Roku's and server are in the same subnet, so no routers, NATs, etc. between. Rokus are wired to a managed or layer 3 switch. Server is attached to the same switch, vlan, as the rokus. nothing magical here. 

 

Originally, I thought I had hit the limits of the roku 3 device - after all, for $70 it does a decent job..

 

someone posted elsewhere in the forum that reducing image size helped their "lag" issues with their roku. I tried this, and noticed an improvement. The network traces were much cleaner. Much of the TCP receive window zero issues were gone. I guess it makes sense. I then used Media Center Master to re-size all my images (max poster/cover width = 320 pixels, max backdrop = 1280 pixels). navigating between objects became much more fluid. genre's and collections were still unresponsive, Still had tcp session timeouts..

 

More digging pointed me to a library.db on my MB3 server. I shut down the server, renamed library.db to library.db.old (and two other library* files), and started the server backup. Took a while for the server to finish a re-scan of my library. Once done, the new library.db was about 1/3 the size of the old library.db (I saved the old library db if any developer at MB wants to look at it).

 

navigating genre's works great. I had to uninstall and re-install auto-box sets to get collections to function. Upon the reinstall of auto box sets and completion of the re-scan, everything worked as expected.

 

Traces are much cleaner now. seldomly see issues on the wire. I would recommend folks consider the above if they have issues deemed as roku stalling, lagging, MB channel crashing, etc. I used to have all of those issues. The above solved it for me.

 

DCA