[jmarshall]

Thumbs won't be stored in databases, correct. However, there's no reason why a database couldn't be used to handle whether there's a cached image, where it is, when it was last cached etc.

[xexe]

Hate to be the one to break this to you

http://wiki.xbmc.org/?title=The_XBMC_Dat...e_Overview

$ file MyVideos34.db
MyVideos34.db: SQLite 3.x database

Yes XBMC uses sqlite dbases but not for thumbnail data storage.s.

But YET AGAIN please stop going OT this thread is to do with fs optimisation for userdata.

thanks

Thumbs won't be stored in databases, correct. However, there's no reason why a database couldn't be used to handle whether there's a cached image, where it is, when it was last cached etc.

jmarshall are these the actions that make the disk the bottleneck rather than say the reading of the data files themselves?

[dc2447]

Thumbs won't be stored in databases, correct. However, there's no reason why a database couldn't be used to handle whether there's a cached image, where it is, when it was last cached etc.

Pretty clumsy though, stating a file on the filesystem in a known location is going to be a lot faster than first going to the database to get the cache information then grabbing the file, imvho

[jmarshall]

Apologies xexe for taking this offtopic. I should think that reading the individual files are the bottleneck, yes. Most of the db would be in memory I should think.

@dc2447 - perhaps we could take the discussion to this thread: http://forum.xbmc.org/showthread.php?tid=61943

Cheers,
Jonathan

[xexe]

jmarshall np that snippet of info gives me enough to go researching benchmarks.

It may come down to some experimentation in the end as I have just read a good few posts on the kernel mailing lists on internal politics of fs devs engineering benchmarking tests to make their fs look the best.

The tricky part with experimentation will be getting XBMC to give me quantative results I can compare against but i suppose that is half the fun.

[TeknoJnky]

The tricky part with experimentation will be getting XBMC to give me quantative results I can compare against but i suppose that is half the fun.

Well, perhaps the easiest way is to take an existing library/thumbnail (large) cache and install it with xbmc on different partitions formatted with different file systems and then simply time the load times with a stop watch.

You would have to be aware of and/or take precautions against differences relating to filesystem fragmentation tho.

[xexe]

Yeah I think i will try to find a way to not use a stop watch. The error margin in human timing will likely be the majority of improvement we can expect to see.

the debug log will likely show me something if i look.

Is there any reason I cannot do this with a virtual machine. If so this testing becomes much easier to achieve.

[TeknoJnky]

Yeah I think i will try to find a way to not use a stop watch. The error margin in human timing will likely be the majority of improvement we can expect to see.

Uh. yeah thats the point.. if you can't measure the improvement with a stopwatch, then maybe its not even worth the difference?

If there is only a few mili-seconds difference between ntfs, ext3, and ext4, in opening your videos, then it doesn't really matter which you pick.

But if there is more than a couple real seconds (which should be plenty measurable with a handheld stop watch), then yeah it will matter alot, and probably matter more in the future as the library continues to grow.

[xexe]

I dont disagree with your sentiment but I suggest that a stopwatch is no way to conduct a systematic comparison. Perhaps I am being anal but I dont like the inacuracy.

[TeknoJnky]

Your right, you are being anal you may want it, but you don't need that kind of accuracy, you just need to know which filesystem is or is not faster in something more signficant than a few milliseconds.

In any case, you could also refer to xbmc log file and compare timestamps from the same operation(s) on the different FS's if you really need to fill that want.

Anyway, unless someone else has already done or is motivated enough to these kinds of tests and posts results, your probably gonna have to do this yourself.

[xexe]

Im pretty sure i dont even need to do timestamps alot of the actions are timed and post info inline. But I digress the main thing is which is consistantly fastest for all common operation, how we come to that conclusion doesnt really matter as long as its accurate enough.

I think i will get help, theres been one person already on IRC shown interest in this.