Re: Best File System for partitions over 600GB
Henrique de Moraes Holschuh wrote:
On Mon, 26 Mar 2007, Mike McCarty wrote:
This is untrue. If power fails during a write, and the drive
scribbles on the disc in a spiral pattern as the head moves
toward the parking area, that particular disc is hosed.
This is a device issue, no filesystem may fix it. Not that I expect even
the crap we buy today for desktops and servers to be THIS dumb.
Yes, a file system can fix that. But it has to be a file system
which understands redundant hardware.
Not true. Read what I wrote above. Even in the face of a complete
meltdown of a disc, the systems I'm talking about would not
lose data.
Easy to do with a RAID with enough redundancy, but then you may get a lot of
problems if something else than a disc meltdowns, and that is NOT something
that uncommon.
No, not true. The system I'm talking about can recover from any
single component failure without any data loss. Depending on what
fails, there may be some reduction in processing capacity.
The bottom line is: you need a filesystem that fully journals everything
that always need a rollback (data doesn't when you only write unused areas
of the disk), always orders everything that needs ordering, AND you need the
entire chain from that filesystem to the disc platter to behave. Otherwise,
you can lose data indeed. It is not easy even if you don't factor in
defective software, firmware or hardware.
What makes you think that the FS I am talking about doesn't
have those features (except journalling, which is not necessary)?
The system I'm referring to has:
redundant separate power supplies
redundant separate processors
redundant separate backplane connections
redundant separate disc controllers,
each of which is accessible from both processors
via both backplanes
redundant discs
each of which is accessible from both controllers
a file system which is aware of all the above, and
which negotiates control of said hardware
via a separate, redundant, communication path
especially made for that purpose
No journalling or rollback is supported[*]. All writes take
place first to one disc, verified, then to the other disc.
No corruption is possible unless a two-point failure occurs.
No one component failing can cause corruption that the file
system cannot recover from, period. The system requires no
down time to replace any one failed component. It just
continues to run, and gracefully recovers from the failure.
Eventually, the system is fully functional and fully
redundant again. System failures are automatically noted,
and failed components are not used. When components are
replaced, this is automatically noted, and the system
automatically begins recovery procedures.
[*] The database system on there does support journalling
and "commit", but not the file system per se. That's at a
higher level.
Mike
--
p="p=%c%s%c;main(){printf(p,34,p,34);}";main(){printf(p,34,p,34);}
Oppose globalization and One World Governments like the UN.
This message made from 100% recycled bits.
You have found the bank of Larn.
I can explain it for you, but I can't understand it for you.
I speak only for myself, and I am unanimous in that!
Reply to: