On Mon, 5 Dec 2005, Neil Brown wrote:
On Friday December 2, [EMAIL PROTECTED] wrote:
On Thu, 1 Dec 2005, Neil Brown wrote:
What I would really like is a cheap (Well, not too expensive) board
that had at least 100Meg of NVRAM which was addressable on the PCI
buss, and an XOR and RAID-6 engine connected to the DMA engine.
there's the mythical giga-byte i-ram ... i say mythical because i've seen
lots of reviews but haven't been able to find it for sale:
http://www.giga-byte.com/Peripherals/Products/Products_GC-RAMDISK%20(Rev%201.1).htm
the only problem with the i-ram is the lack of ecc (it could be
implemented in a software layer though).
The lack of ECC is a real killer for any sort of raid/journalling
application. You want your journal to be more-resilient to failure
than your main array, not less.
And the fact that the board presents like an IDE controller (if I read
that right) means it is meant as fast-disk. Not as a generally useful
board to do interesting things with :-(
The idea of just providing memory slots that you plug your own memory
into is certainly interesting. It would mean making just one board
and the custom could choose how much memory they wanted to buy. (umem
below may several different sizes). I wonder if it affects
reliability at all...
umem.com have what look like excellent boards but they seem unwilling to
sell in small quantities...
http://umem.com/Umem_NVRAM_Cards.html
I also tried to contact them, so thats why I did not get any reply.
I have two of these in two different servers... I must have got them
at a time when small quantities weren't so unpalatable. They have
full ECC, and even have a driver in kernel.org kernels (I had to
convince them that GPL wasn't bad, but that wasn't too hard).
The boards I got only presented a 64M (I think) window to the PCI
buss, though I believe they have versions that present to whole of
memory instead of just one window.
Do you have any performance numbers that show how these cards increase
performance? I have tried doing it with a ramdisk and the performance
was not much better, using ext3 and journal on ramdisk. ext2 is still
much faster, here some quick tests with dd:
2.6.15-rc1-git2
time dd if=/dev/full of=/home/aaaa bs=4M count=4883
SW Raid 1+0
ext2
real 1m11.856s 1m16.050s 1m18.449s
user 0m0.020s 0m0.000s 0m0.010s
sys 0m44.630s 0m44.820s 0m45.870s
ext3 ordered
real 2m17.314s 1m51.732s 1m51.258s
user 0m0.030s 0m0.020s 0m0.010s
sys 1m41.370s 1m38.300s 1m36.820s
ext3 writeback
real 2m1.511s 1m40.958s 1m43.889s
user 0m0.070s 0m0.000s 0m0.000s
sys 1m33.390s 1m21.470s 1m23.430s
ext3 writeback+external journal in ram disk
real 2m5.357s 1m41.349s 1m36.228s
user 0m0.010s 0m0.020s 0m0.010s
sys 1m37.500s 1m24.110s 1m21.960s
Wonder how one can increase the write performance but still having a journal.
Holger
-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
