Clement wrote:
>
> This thread is quite interesting. I think it comes to the following
> questions:
>
> - Is the reiserfs stable enough?
> - Is the overhead of reiserfs acceptable?
> - Is the time to 'fsck' valuable enough to justify the cost?
I think you're right - these are the big three. Also, one more though:
many people mention that if you have to worry about your systems
crashing/unmounting the FS properly, you're doing something wrong/have a
single point of failure.
In any given system or system of systems, there is some probability that
something will fail. Most of us know the math behind that; I won't go
into it. But rather than look at the probability that the power will
fail, how about looking at the probability that a file(system) will
become corrupt? Looking at it from this perspective goes beyond the UPS
failing or being non-existant. In fact, there're many other issues
involved.
For instance, a file can become corrupt because the power supply on the
mission critical machine fails and you don't have one of those servers
with three of them inside. Or the machine crashes for some reason.
Let's face it -- no OS is crash-proof. Perhaps the memory inside is
bad, or goes bad, causing the machine to crash. It takes me 2 minutes
to replace a DIMM, and 30 minutes to get the server back up while it
checks the 200GB attached. Motherboards fry. UPS's fail. Crackers
screw-up OS's. Basically, **** HAPPENS.
It's only a matter of time before some failure causes file system
corruption given that no person or machine or piece of code is perfect.
Using a journaling file system is just _yet_another_tool_ to fight
consequences when any of the above or other problems occur.
Yet another thing --- sooner or later, mounting/unmounting (due to safe
rebooting) is going to cause ext2fs to reach its maximum mount count and
it'll go through the whole fsck'ing at startup. Yes, you can up the
mount count, but the fact the designers of the FS made that feature
tells you that it's an important one for that FS. Perhaps for a planned
reboot of a server, the 5 minutes it's down is an acceptable outage...
but if to your surprise you're waiting 2 hours for it to just check the
FS's, that may not be so acceptable.
As for the question of is journaling too expensive in terms of
performance... well generally that answer is "no". In the case of
ReiserFS? I've never read or heard or spoken to anyone within the past
couple of months using it in production that thought it slowed-down
their systems to any measureable degree. Our tests concurr.
My original argument with "RaghuNath L" was that mounting a ResierFS FS
after a crash faster than an ext2fs FS after a crash was not the only
benefit. Also I argued that it is recommended for production
environments; of course, this is relative to your needs. No given
product is recommended for _every_ production environment, but the
original implication was that ReiserFS shouldn't be used in _any_
production environment, with which I disagreed.
We are presently expoloring changing a 100GB file server to ReiserFS.
This server has _never_ crashed, _never_ been hacked, _never_ had a
hardware failure, and it's UPS has _never_ failed either. But one day,
something will go wrong. Clients are much more understanding of a 10
minute outage than a 60 minute one :-)
-Fred
_______________________________________________
Redhat-list mailing list
[EMAIL PROTECTED]
https://listman.redhat.com/mailman/listinfo/redhat-list
