> For curiosity: do they defrag it while the system is idle? They should
> try in all ways to avoid the log to wrap. Otherwise they are forced to
> wait for I/O completation of the wrap-fault-handler.
They don't need to defrag frequently:
* only the metadata changes of NTFS are stored in the logfile.
* since the complete list of all unfinished transactions is also
stored in memory, they can write a checkpoint record of all unfinished
transactions, and this makes all older entries basically superflous.
A finished transaction is stored in the NTFS metadata, they simply
discard it from the log. If the system is idle, then they to that
every few seconds (5?)
* If they really overflow, then they wait (and this must be a big
problem, given the fact that they set the file size to 64 MB)
Note: most of my knowledge about NTFS is based on Helen Custers book
about NTFS, and that book contains more marketing slogans than
technical details.
One of our harddisks got fataly corrupted due to a stupid
implementation bug in NTFS (shutdown, restart, harddisk gone).
Their fsck was unable to correct the problem.
Unfortunately, I'm not able to reproduce the problem.
> the other performance problem of a real logfs is that we are
> not going to always read _data_ in the same order [...]
As I pointed out, you don't need to read that file except during
fsck. Infact, LFS probably doesn't qualify as a complete logging
fs, it more a cyclic buffer for transactions.
Regards,
Manfred
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/