[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: ext3 patch for 2.4.6 (plain or ac1)



The machine I was doing this on has 2G of memory -- when you had said
"memory boundary conditions" I thought you meant in software -- not
hardware.  This system has been rock solid banging the heck out of the disks
for months on ext2.
(Gbyte datasets flying around -- both read and write).
Seeing as how a raid resync is really hitting the write buffers hard and an
fsck is hitting the read buffers hard that might be where it hits  a
problem -- after I waited for the raid resync to finish the fsck worked
fine.  But then it barfed on tiobench.
I've got a little leeway right now with doing some testing on this machine
(most people are on vacation for the next week or two).

Just say that you think it's "stable" and I'll give it another whack -- I'd
just like a little warm fuzzy to know that I'm not wasting my time.

----- Original Message -----
From: "Andrew Morton" <andrewm uow edu au>
To: <ext3-users redhat com>; "Mike Black" <mblack csihq com>
Sent: Wednesday, July 04, 2001 12:06 PM
Subject: Re: ext3 patch for 2.4.6 (plain or ac1)


> Mike Black wrote:
> >
> > When you say "looking good" do you mean bug-free?
> > When I was testing the latest it barfed on my system while doing a raid1
> > resync and fsck at the same time (previous emails described the
problems).
> >
> > I can test it again if you think it's supposed to be better able to
handle
> > the boundary memory conditions.
>
> We never got to the bottom of that (I think).  It wasn't clear
> whether it was a hardware problem or not.
>
> But yes, 0.0.8 will do bad things if the system seriously
> runs out of memory.  The current development tree works
> around out-of-memory problems in a brute-force manner (ie:
> it simply retries) and that appears to work very much better.
>
> We still have problems with memory shortage in data=journal
> when doing nutty things to machines with 64 megs of memory;
> I'd recommend you stick with the default mode, data=ordered.





[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]