43 replies to this topic

[hypnotoad]

Hi,
i try to install the oe packages:

libhandle1_3.1.7-r1_mips32el.ipk
xfsprogs-fsck_3.1.7-r1_mips32el.ipk
xfsprogs-mkfs_3.1.7-r1_mips32el.ipk
xfsprogs-doc_3.1.7-r1_mips32el.ipk
xfsprogs_3.1.7-r1_mips32el.ipk

from another source (the pli repository doesn´t offer it)
But it seems that xfs support isn´t enabled in the kernel, could this be true, cause i can´t "see" my hdd after formating it with xfs (i formated it under debian)
Is there any particular reason why you guys don´t support xfs?
As you know, there are a lot of users out there with the freezer problems, guess this thread is well known, and as far as i know, dmm has also problems with freezers and plan to use xfs in future as the standard filesystem.
So if any pli dev can drop me some lines it would be nice

[MiLo]

Probably you'll need kernel-module-xfs or so. I've tried it (also tried btrfs, which was even worse) but it didn't perform as well as ext4 did, so I never followed up on it.

[MiLo]

Is there any particular reason why you guys don´t support xfs?

No one ever asked for it. You're the first.

[MiLo]

By the way, I can record 8 HD channels simultaneously, and run a "dd if=/dev/zero of=/media/hdd/blank" alongside it, without causing a glitch in the recording. I'm using an old 100GB 5400rpm laptop disk for those tests.

[hypnotoad]

I know, i also got no problems, but realy huge amount of users has problems, and i guess its not a "the problem sits in front of the monitor" bug.
So my guess was, just give xfs a try, ghost seems to be also testing with xfs and he has better IO results than with any ext fs.
btrfs is still fresh and kind of beta so this is not a solution, the history of xfs reaches back to the mid 90´s so i guess its fully developed.

Do you have the Kernel Module available? Cause DMM seems to build it directly into the Kernel, so no chance on dreamboxupdate ;-)

[SpaceRat]

By the way, I can record 8 HD channels simultaneously, and run a "dd if=/dev/zero of=/media/hdd/blank" alongside it, without causing a glitch in the recording. I'm using an old 100GB 5400rpm laptop disk for those tests.

It's not a problem of recording properly...
I can have as many recordings at the same time as my tuner configuration allows, they are all 100% ok, even according TS-Doctor.

But, like about everybody, I can not even play back a single recording or timeshift without freezers/glitches.
As mentioned above, the problem is not within the recording (As the TS verifies perfect with TS-Doctor and simple rewinding enables the user to watch the same part of the recording without freezers) but has to do with READ access to the file.

Also, ext4 is known to give the worst results, while ext3 worked for some and ext2 for some more people.

This said, I support ANY approach that can make our Enigma2-crap a usable Receiver.
So if xfs might help, give it a try.

[MiLo]

I've added the XFS module for all boxes, and put the xfsprogs package into the feed. Should be available in tomorrow's build.

[SpaceRat]

I've added the XFS module for all boxes, and put the xfsprogs package into the feed. Should be available in tomorrow's build.

Tnx

[Martin_B]

Thanks too !!

[hypnotoad]

great! thx milo ;-)

[SpaceRat]

Has anyone done any tests with xfs yet?
I'm currently emptying my HDD via network which takes quite a lot of time. I hope to be able to test xfs on monday, I was just wondering if anyone else has already tried ...

BTW: Maybe we can have IBM's JFS for a test ride too? And/or ReiserFS ...

[hypnotoad]

Guess anything goes wrong, i just make a update/upgrade and install the needed packages (added a new partition with sfdisk and than mkfs.xfs on it) but busybox mount don´t know xfs?
(testing the same drive on my debian box results in no problem, i can mount and write on it)
In the upgrade process i get amongst others the new kernel-3.4.3.

root@et9x00:~# mount -t xfs /dev/sda1 /tmp/test/
mount: unknown filesystem type 'xfs'

[MiLo]

You'll probably need a "big" mount program too, like the "util-linux-mount" package.

[MiLo]

Has anyone done any tests with xfs yet?
I'm currently emptying my HDD via network which takes quite a lot of time. I hope to be able to test xfs on monday, I was just wondering if anyone else has already tried ...

Well, I DID write:

... I've tried it (also tried btrfs, which was even worse) but it didn't perform as well as ext4 did, so I never followed up on it.

[MiLo]

BTW: Maybe we can have IBM's JFS for a test ride too? And/or ReiserFS ...

Exactly what are you hoping to accomplish?

With these big recording files, the best filesystem in terms of performance is probably FAT, because it has the lowest overhead and is simple to handle. It will beat the others. But you can't have files over 4GB in size so that makes it rather useless.

The performance requirements are really really low.
If there's something that doesn't seem to work right, it's most likely the fault of the application (that would usually be enigma2) and you shouldn't blame the filesystem for being just the messenger.

[SpaceRat]

... I've tried it (also tried btrfs, which was even worse) but it didn't perform as well as ext4 did, so I never followed up on it.

The part marked in bold is the problem ... so far, ext4 gives the worst results, worse than ext3 and far worse than ext2.
I don't care if 8 simultanous recordings would be possible (If Enigma2 was able to use the tuners wisely).
My wife is giving me a hell of a time because I traded a working Humax iCord HD for an Enigma2 box, which can not even play a recording/timeshift well when there is ZERO background activity.

[SpaceRat]

Exactly what are you hoping to accomplish?

My Vu+ Ultimo being able to play back recordings/timeshift without glitches/freezers that are not within the file (Rewinding helps, but is just a very bad workaround).

With these big recording files, the best filesystem in terms of performance is probably FAT, because it has the lowest overhead and is simple to handle. It will beat the others. But you can't have files over 4GB in size so that makes it rather useless.

That's where I agree.
I would use ext2, but it is not journalled, and endless runs of fsck on boot render ext2 rather useless too.

The performance requirements are really really low.

And that's why I still believe there must be a filesystem that performs better on reads than ext4 and still performs sufficiently on writes.

[hypnotoad]

i allready got "util-linux-mount" but no chance
i installed the busybox stuff from opendreambox but than i got

root@et9x00:/# busybox_dream mount -t xfs /dev/sda1/ /tmp/test/
mount: mounting /dev/sda1/ on /tmp/test/ failed: No such device

sure that the kernel module now added into the kernel directly?

[SpaceRat]

BTW: I made an observation which might be related to the poor performance of ext4!
While copying the whole harddisk to my PC across the network, the average speed is about 7 MByte/s. However, this average does not result from a slight hysteresis around these 7 MByte/s, but from near digital speed changes:
ON = 100 MBit/s
OFF = 0 MBit/s
You can monitor this using a network speed monitor on the PC ...

The iCord HD performed bad on his NIC and gave me an average speed of about 5-6 MByte/s, but it did so constantly.

Assuming it's not the networking code of the Enigma2 box, it would mean that the average read speed might be ok, but with those peaks down to zero playback just can't work without glitches.

[MiLo]

... I've tried it (also tried btrfs, which was even worse) but it didn't perform as well as ext4 did, so I never followed up on it.

The part marked in bold is the problem ... so far, ext4 gives the worst results, worse than ext3 and far worse than ext2.
I don't care if 8 simultanous recordings would be possible (If Enigma2 was able to use the tuners wisely).
My wife is giving me a hell of a time because I traded a working Humax iCord HD for an Enigma2 box, which can not even play a recording/timeshift well when there is ZERO background activity.

Sounds to me like your box is broken. If you still have warranty, send it back to the store.