ings,
but I was running in single-user mode and couldn't regain control of
the console), but this should be fairly repeatable.
Fortunately, no data loss was caused by this.
--
Osma Ahvenlampi
tests were done over multiple controllers so SCSI's ability
to command multiple devices in parallel was not a factor in the tests.
--
Osma Ahvenlampi
ere's absolutely no
> advantage in having multiple md devices instead of a single device.
Wrong again. Even if md was single threaded, disks are much slower
than CPU, and DMA disks (especially with SCSI's tagged queuing) are
very easily parallelized.
--
Osma Ahvenlampi
size. However, at least if
you're using Samba 2.0 and Netatalk 1.4b2-asun, you don't have to do
anything about the 2G issue. Yes, clients think the volume only has 2G
available, but that's because the server is lying to them to keep them
happy. It'll work just fine.
--
Osma Ahvenlampi
ne by one. Not good for performance.)
--
Osma Ahvenlampi
ove 2.2.6 until an official patch exists and the
filesystem problems in the vanilla 2.2.x, x>6 problems have been
resolved.
--
Osma Ahvenlampi
with root and
other volumes pointing to the RAID partitions, and then raidhotadd the
partitions on the first disk. It sounds more complex than it really
is..
--
Osma Ahvenlampi
orts many platforms, and is surprisingly
inexpensive as commercial backup software goes. Check
www.arkeia.com. I have no other relation to them except as a satisfied
customer.
--
Osma Ahvenlampi
systems which are on an UPS still crash. One of those power events
> crashed a server, which was on a decent UPS, and the RAID-5 on it was
Off-topic, but a fully on-line UPS (not simply line-interactive)
should protect even against phase loss. An example of such would be
Powerware (former Exide) series 9.
--
Osma Ahvenlampi
y would be a
350MHz P-II, one U2 controller and 4+ disks. The CPU will do RAID-5
parity checksums at 850MB/s, ten times the speed of one Ultra2 SCSI
channel. A CPU upgrade will not make this system faster - more disks
and SCSI controllers will.
--
Osma Ahvenlampi
wo UltraSCSI controllers, the
SCSI bus bandwidth also exceeds the disk capacity. The resync is
happening at media speed. There's no way a hardware solution could be
faster than that, and low-end devices are very likely slower.
--
Osma Ahvenlampi
idually) Very perplexing!
I don't quite understand what exactly you were trying. I can tell you
that dump for sure does work on a raid array. Did you try to dump the
mount-point directory or the device itself? Try "dump /dev/md0".
--
Osma Ahvenlampi
s to log errors in the system log, and
-k is for you by the kernel daemon. Errno 2 is "No such file or
directory" (perl -e '$!=2; print "$!\n";' is one way to check).
--
Osma Ahvenlampi
.7 has shown problems with the raid patches (apparently related to
a scheduler update activated in 2.2.8).
--
Osma Ahvenlampi
1 2
/dev/hdc1 /boot.alt ext2 defaults,nodev,nosuid,noauto 1 2
--
Osma Ahvenlampi
w as high bandwidth for
a single file, but would allow more efficient access to many files at
once (such as in a large mail, news or web server).
--
Osma Ahvenlampi
locks
level 5, 32k chunk, algorithm 2 [5/5] [U]
md4 : active raid5 sdf8[5] sde8[4] sdd8[3] sdc8[2] sdb8[1] sda8[0] 30587136 blocks
level 5, 32k chunk, algorithm 2 [5/5] [U]
unused devices:
--
Osma Ahvenlampi
Dietmar Stein <[EMAIL PROTECTED]> writes:
> Readperformance will only increase by using raid0 (stripe), but it will
> not be twice times faster.
Does the Linux RAID-1 code still not stripe reads? I thought it did.
--
Osma Ahvenlampi
y ignored, the latest code handles
RAID start/stop automatically inside the kernel process.
--
Osma Ahvenlampi
able to sync to the other
devices?
(I'm aware of the failed-disk patches and have used them
successfully. Is there another way?)
If you ask me, mke2fs should be modified to reserve that 4k in the end
of the volume by default..
--
Osma Ahvenlampi
faster than
raid5.
--
Osma Ahvenlampi
single SCSI disk only as well as
both RAID configurations would be a good start. If you're really
interested, running some standard(ish) database benchmarks would be
nice, too.
--
Osma Ahvenlampi
Andy Poling <[EMAIL PROTECTED]> writes:
> If you'll be doing alot of writing, the controller can use it's memory to
> coalesce the writes and optimize the parity calculations for writing.
I believe the Linux buffer-cache will try to do the same.
--
Osma Ahvenlampi
one anything. "protocol
error, upgrade kernel raid support" would be much more informative.
--
Osma Ahvenlampi
dstart them (but since they're autostarted, they should be
running - see /proc/mdstat).
The partition ID is not "automount", it simply initializes the RAID
subsystem and makes the /dev/md? devices available to be mounted.
--
Osma Ahvenlampi
t does. The RAID superblock will be placed at the end of the
partition. So Grub would be able to read the kernel image off any one
partition of the RAID-1 set.
--
Osma Ahvenlampi
ysical partition for the
boot files (MBR, kernel image and LILO map) since LILO does not know
how to deal with software-RAID devices, however everything else can be
on RAID.
--
Osma Ahvenlampi
might want to upgrade (but it's a big
effort), and another that I just set up with the most recent snapshot
that apparently I could have configured for RAID-swap had I known
about this.
--
Osma Ahvenlampi
t install the RAID kernel patches that
go along with the raidtools.
--
Osma Ahvenlampi
Ingo Molnar <[EMAIL PROTECTED]> writes:
> it does work for me (i do not actually use it as such, but i've done some
> stresstesting under heavy load). Let me know if you find any problems.
Hmm? Since when does swapping work on raid-1? How about raid-5?
--
Osma Ahvenlampi
to manually edit /etc/raidtab to keep in sync". Ta.
How about a program that would scan the running raid arrays a create an
/etc/raidtab file that matches the running configuration?
--
Osma Ahvenlampi
here? The partitions certainly aren't in use or
anything.
--
Osma Ahvenlampi
is a great FS with journaling,
metadata attributes, attribute indexing (lighting-fast file searches
based on name, size, and whatever else you want) and speed, but of
course using it under Linux currently would be somewhat strange.
--
Osma Ahvenlampi
ing with the
information. What matters is that the files pointed to by the map and
image options are not on RAID volumes.
--
Osma Ahvenlampi
Benno Senoner <[EMAIL PROTECTED]> writes:
> I don't know how much slower it is to swap do a regular file instead
> using the raw partition,
You can't swap to a file on a software RAID-1/5 filesystem either.
--
Osma Ahvenlampi
the bootup sequences which would
> be on /dev/sda, /dev/sdb, or /dev/sdc etc...
And what about the system BIOS looking for the master boot record (ie,
LILO) on the first disk? If that disk is completely dead, fine, it'll
try the next one, but if it's just returning corrupt data? (Never mind
IDE hardware that would even handle the completely-dead case).
--
Osma Ahvenlampi
rting to floppy
Software RAID can not solve it because the software isn't running at
the time the boot sector needs to be accessed. Hardware RAID does, as
long as it has a BIOS driver.
--
Osma Ahvenlampi
RAID, you can see it from the /proc/mdstat output,
hardware controllers report the information in their drivers'
respective information. Simplest solution is to poll this information
periodically and trigger the hotel's fire alarm if something goes
wrong. ;)
--
Osma Ahvenlampi
is running background
resync. 80MB/s busmaster-DMA SCSI sure is nice. ;) (23MB/s streaming
speed in bonnie benchmarks).
When I tried PIO-mode on my workstation box, the system became "chunky"
even with the one drive doing heavy I/O. PIO puts a huge strain on the
CPU. DMA w
re apparently is something seriously wrong
there. I can vouch for the stability of the earlier patches, though.
--
osma ahvenlampi || | || | | | || | r a z o r f i s h
|| | | h e l s i n k i
Hmm. I use SCSI on high-performance systems, but if IDE is so bad, why
does NASA use IDE? ;)
http://beowulf.gsfc.nasa.gov/bds/bds.html>
--
I didn't know it was impossible when I did it.
Osma Ahvenlampi <[EMAIL PROTECTED]>
MOLNAR Ingo <[EMAIL PROTECTED]> writes:
> > We should remove the raidstop commands from our reboot/halt scripts - it's
> > no longer needed? What about halt, especially?
> yep, it should all work magically. halt too.
Sorry to be a dumbass, but I want to make this 100% clear to myself
before I do
have not followed it too closely.
--
Blessed are the inept for they will inherit the skies.
Osma Ahvenlampi <[EMAIL PROTECTED]>
re 5400RPM
> drives with slower internal transfer rates and latency than my ST15150W's.
The last Bonnie number is not K/sec. It's seeks per second, and
entirely incomparable to the other numbers.
--
Do not take life too seriously; you will never get out if it alive.
Osma Ahvenlampi <[EMAIL PROTECTED]>
[1] sdb2[0] 17510656 blocks level 5, 32k chunk,
algorithm 2 [3/3] [UUU]
--
Left to themselves, things tend to go from bad to worse.
Osma Ahvenlampi <[EMAIL PROTECTED]>
n them, fsck should parallelize for
both.. But how can it figure out which volumes are on the same
devices?
--
The less you bother me, the sooner you'll get results.
Osma Ahvenlampi <[EMAIL PROTECTED]>
volume? Is there a solution besides
repartitioning to smaller volumes (and how should I proceed with that,
given that I can't dump the filesystem..?)
--
The only way to amuse some people is to slip and fall on an icy pavement.
Osma Ahvenlampi <[EMAIL PROTECTED]>
ional, as when I reconnect the drive, it will start to
reconstruct itself in the background. However, I'd like to know that I
could replace a broken drive without a system restart, as that's what
the hot-plug backplane is there for.. :/
--
Murphy's Law only fails when you try to
48 matches
Mail list logo