Page 2 of 2
Re: Normal thekraken Behavior? [Yes]
Posted: Mon Apr 01, 2013 8:46 pm
by Nathan_P
That would be the one, the guides at [H] have been updated though and all references there have been removed
Mod edit: Thanks Nathan, I've removed the link to [H].
Re: Normal thekraken Behavior? [Yes]
Posted: Mon Apr 01, 2013 11:12 pm
by Gehacktesmacher
I am using EXT4.
Nice, I will test this option. Thanks for help.
Re: Normal thekraken Behavior? [Yes]
Posted: Tue Apr 02, 2013 8:30 pm
by Nathan_P
Its still a valid fix but the [H] guides now use a ramdisk before writing anything to the drive.
Re: Normal thekraken Behavior? [Yes]
Posted: Sat Apr 13, 2013 1:51 am
by tear
And, for the record, Kraken doesn't have anything to do with Gehacktesmacher's issue.
It's multi-year-old bug that's present in all SMP FahCores and series 6 clients.
Re: Normal thekraken Behavior? [Yes]
Posted: Sun Apr 14, 2013 12:36 am
by Joe_H
tear wrote:And, for the record, Kraken doesn't have anything to do with Gehacktesmacher's issue.
It's multi-year-old bug that's present in all SMP FahCores and series 6 clients.
Not so much a bug, as file access programming methods were used that do not behave the same on EXT4 filesystems with barriers enabled. Those methods ensured that the data was reliably written to disc instead of possibly residing only in the filesystem cache. It is the reaction of the filesystem coding to these file writes that has changed, F@H is not the only code that has run into issues with the changes made in default Linux filesystems.
Re: Normal thekraken Behavior? [Yes]
Posted: Sun Apr 14, 2013 1:09 am
by tear
Tom-a-to, tom-ah-to.
Either way, it's a brain-dead method of ensuring data are secure on disk.
fsyncing every 4k is plain stupid because a short-file, no matter how securely stored on-disk, is unusable by FAH.
Imagine you have 100MB to write as part of WU completion process.
Imagine something happens 8 kB away from completion and your file is 8 kB short Cool, you have 100MB - 8kB of data secure on disk.
Are they usable? No. You've just thrashed your medium with log, metadata, and data updates for no good reason, congratulations.
And what about writing zeroes and syncing every 4k to "clear" the files? What purpose does that serve? Ever heard of unlink?
Or just adjusting file header to mark that data are stale?
I did list number of better paradigms in one of older posts. You may want to spend some time exploring those
BTW, I could code it better myself. And I do. I do it for a living.
Re: Normal thekraken Behavior? [Yes]
Posted: Fri Feb 28, 2014 4:23 pm
by bruce
WhitehawkEQ wrote:I don't know if anyone knows this already but here goes, I have found out that if I stop the client then re-start it, I will end up having 2 instances of thekraken running and they fight causing frame times of 1 hour or more. So re-boot before re-starting the client.
Re: Normal thekraken Behavior? [Yes]
Posted: Fri Feb 28, 2014 7:09 pm
by tear
Please create a new thread as your symptoms are very different from the OP.
Also, can you provide the following information:
- client version
- Kraken's version
- Linux distribution and version
Thank you!