On Fri, 16 Dec 2005, Neil Brown wrote:
> > - Does RAID6 have disadvantages wrt write speed?
>
> Probably. I haven't done any measurements myself, but from a
> theoretical standpoint, you would expect raid6 to impose more CPU load
> (though that may not be noticeable) and as raid6 need to see the whole
> stripe to update the P and Q blocks (it cannot do a subtract old, add
> new update) there could well be more IO happening, particularly on
> large arrays (>5 devices).
>
> Of course, whether this caused noticeable reduction in throughput and
> latency would be very workload-dependant.
I now have several servers using RAID-6 in a production environment - so
far so good. For the most part, they are all connected to 100Mb networks,
and they are file-servers (rather than serving to a local application) and
throughput from the servers to the clients hasn't been a problem. I have
no problems saturating the Ethernet with file traffic in experiments I've
performed (mainly running Bonnie via NFS!)
Running Bonnie locally, I can get over 100MB/sec on some of these boses,
depending on the disks and processor, so it might just be able to cope
with Gb Ethernet, but if I were building something with those speeds
in-mind, then I probably wouldn't be using comodity hardware...
There is something that says that if you have a batch of disks which are
likely to fail then theres a certian probability that they'll all fail
round about the same time, so having an array with a cold spare might help
that, however that might be more hassle than it's worth, who knows!
Bonnie (and hdparm) are fairly crude when it comes to benchmarks, but they
give a fair idea of what to expect when streaming data - here is output of
a Bonnie run over a 6-disk RAID-6 array (SCSI) on a Dull 2850 dual Xeon
jobbie:
Version 1.02b ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
black 2G 74811 34 35960 16 79723 17 461.7 1
black,2G,,,74811,34,35960,16,,,79723,17,461.7,1,,,,,,,,,,,,,
I haven't tried to do anything clever (not sure I could with this!) All
drives are on the same SCSI bus (I did ask for the split backplane option
when I ordered it, but Dull said it couldn't be done, even though they
advertise it)
md6 : active raid6 sdf6[5] sde6[4] sdd6[3] sdc6[2] sdb6[1] sda6[0]
394523648 blocks level 6, 64k chunk, algorithm 2 [6/6] [UUUUUU]
I was geting over 200MB/sec in & out of it when I configured it as a
RAID-0 when testing it.
And to compare, on a crappy old twin xeon/500MHz box with 8 external SCSI
drives:
Version 1.03 ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
bertha 1G 38659 69 18568 50 44925 73 562.2 8
bertha,1G,,,38659,69,18568,50,,,44925,73,562.2,8,,,,,,,,,,,,,
md0 : active raid6 sdl1[11] sdk1[8] sdj1[5] sdi1[2] sdh1[10] sdg1[7]
sdf1[4] sde1[1] sdd1[9] sdc1[6] sdb1[3] sda1[0]
179203840 blocks level 6, 64k chunk, algorithm 2 [12/12] [UUUUUUUUUUUU]
There are split 4+4 over 2 SCSI buses, but the server only has one PCI
bus, so it all ends up going down the same path anyway.
So I'd say that with RAID-6 you can easilly produce something that will be
fully capable of saturating a 100Mb network, with capacity left over, but
you'll have to work a bit harder to saturate a Gb network.
Gordon
-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
