Re: Curiouser and curiouser

On 7 May 2007, in the Usenet newsgroup alt.os.linux.suse, in article
<1178570138.525892.154010@xxxxxxxxxxxxxxxxxxxxxxxxxxxx>, Vlad_Inhaler wrote:

(Quoting Ted T'so from 2004:)

I was actually told about this by an XFS engineer, who discovered this
about the hardware. Their solution was to add a power-fail interrupt and
bigger capacitors in the power supplies in SGI hardware; and, in Irix,
when the power-fail interrupt triggers, the first thing the OS does is
to run around frantically aborting I/O transfers to the disk.
Unfortunately, PC-class hardware doesn't have power-fail interrupts.

Thinking about that, running xfs / Reiser / jfs should be ok on a
decent laptop? Unless the laptop 'freezes' on you (my old one did
this a few times), xfs should be just fine.

I dunno - I've never looked at the schematics of a laptop to see how it
gets power, and how the /RESET signal is generated. My gut feeling is
that it's probably similar, except that there _MIGHT_ be something
monitoring the DC input (battery, or rectified mains) for a minimum
level, possibly in addition to the power supply outputs. I know that
some laptops provide a warning that battery voltage was low, and it was
time to quit while it was possible to do so, but I've never seen it use
similar to a UPS battery signal - to cause a clean shutdown.

I have used xfs for years on Linux desktops and have had one power
outage where I lost data. Two weeks ago :-( That partition is now
running under ext3.

In theory, if you have a UPS that can cleanly shut down in the event of
a power failure, the file system shouldn't matter much. If you lack the
UPS, then it gets interesting in the Chinese meaning. If there is no
disk activity, things might be OK. If there is something trying to write
to disk... "Gentleman, place your bets!"

Reiser has been a lot less reliable for me. Serious data loss several
times, and I thought it was me . . .

Both of the guys here who ran the evaluations are on leave, and the
only thing I've got at the moment is a note reporting data loss in thirty
of the fifty tests they ran (basically yanking the mains plug), and
fourteen of those thirty losses were unrecoverable. This note reflects
testing done last October, but I don't have the earlier test reports (we
started testing journalling filesystems in 2001 when Red Hat included
ext3 in their 7.2 release). Otherwise, I've seen _very_ frequent
anecdotal evidence (mainly on Usenet) for problems with the Reiser file
system, slightly less for xfs and jfs. Experience with ext3 (or e3fs)
has been reasonably good, and at least at home, I've not had any file
losses. How much this is due to how the file system is being used, I
can't say. My servers and two workstations are on UPS, but three
other workstations are not (but they're also not used that much). At
work, all servers have UPS (and backup generator), and all workstations
have /home mounted as a network drive (though the local drive contains
everything else). These run ext3, and while I'm not associated with the
helldesk, I haven't heard of any significant number of complaints
attributable to power problems.

Old guy
.

Relevant Pages

  • Re: Memtest stops at pass 78% and other problems....
    ... the UPS should have made brownouts ... This assumes your power supply contains essential ... missing functions result in hardware damage, ... brownout completely irrelevant. ...
    (alt.comp.hardware.pc-homebuilt)
  • Re: Power outage problem
    ... But a UPS is recommending where a system utilizes Write caching. ... Not as a hardware protection, but to allow data to be flushed to the ... Abrupt power loss in that situation can result in data loss. ... One of the best investments I ever made in computer hardware. ...
    (microsoft.public.windowsxp.general)
  • Re: Uninterruptable Power Supplies for Alphas
    ... >> We don't need UPS for uptime, since all these system control other ... >> stuff that goes down when the power fails. ... >> reduce our hardware repair and replacement costs, ... Don't know whether they still have that VMS agent available. ...
    (comp.os.vms)
  • Re: Uninterruptable Power Supplies for Alphas
    ... our management decided to drop hardware maintenance a while back. ... >>We don't need UPS for uptime, since all these system control other stuff ... >>Are there power conditioning solutions that are less expensive than UPSes ... I would need a way to do an orderly shutdown before the UPS shuts down. ...
    (comp.os.vms)
  • Re: Uninterruptable Power Supplies for Alphas
    ... our management decided to drop hardware maintenance a while back. ... As long as there is failover capability and redundancy built into the servers ... > We don't need UPS for uptime, since all these system control other stuff ... I have lost VMS servers through a series of events (loss of power, ...
    (comp.os.vms)