Re: Doesn amanda no longer do estimates for a given host in parallel?
2009-08-27 18:06:31
Jean-Louis Martineau wrote:
man disklist, look for spindle, set them to different value or '-1' if
you want parallel estimate and dump.
Jean-Louis
stan wrote:
In doing some testing, it appears that Amanda 2.6.1 no longer does
estimates for multiple DLE's on a client in parallel.
Is this true? If so, can I configure Amanda to the old behavior? This
will
probably substantially increase the amount of time my run takes.
Just a quick note on the spindle item, since it might prove useful to
someone.
Sun traditionally uses a designation cxtxdxsx for a partition (or
slice), where the x's are numbers. The cx is the scsi bus instance or
controller number, the tx is the SCSI ID or target, and the dx is the
LUN. So, typically the cxtxdx indicates a drive, and the sx is then the
slice or partition on that drive. For all the cases I have had to deal
with, the cxtx uniquely defines the drive. So, I pull out the xx's, put
them in 2 digits per x form, and trim leading 0's to come up with
spindle numbers for my Amanda disklist.
I start by doing a `df -k | grep '/dev/dsk' | sort +5` to get a complete
list of mounted partitions in order by mount point. That makes it easy
to walk through the disklist and add spindle numbers.
My internal drives are typically something like c0t0d0, c0t8d0, c0t9d0,
c0t10d0, c0t11d0, and c0t12d0. Those would translate to 0, 8, 9, 10, 11,
and 12. The external drives might be c4t1d0, c4t2d0, etc. in one
cabinet, and c5t1d0, c5t2d0, etc. in another cabinet. Those would
translate to 401, 402, 501, and 502. There might be anywhere from 1 to 7
partitions with the same spindle number.
This gives me an easy, direct way of mapping partitions to unique
spindle numbers (which, from Amanda's point of view are just unique
numbers), and I don't have to create some extra lookup table or record
to remember them. The spindle numbers in my disklist should always
correspond with my `df -k` and be easily determined without hurting my
head trying to double check where my numbering has gotten to and whether
I've duplicated something or already used a number.
That said, I'm just getting into SAS multipathing with a dual connected
J4200 array. Multipathing substitutes a long Guid
(http://www.jmcp.homeunix.com/~jmcp/WhatIsAGuid.pdf) for the tx, so I
end up with something like c5t5000C50004B49973d0 for the disk
specification. Then I'm throwing those into a zpool. I haven't quite
decided how I'm going to deal with all this. On the other hand, this
array (with SAS multipathing and 15K rpm drives) is so fast that the
internal drives can't keep up with it. I did a copy of 80G from an
internal pool to the J4200 and then did a verbose zpool iostat at 10
second intervals. About every other iostat snapshot showed the J4200
idle. It seemed to be emptying buffers faster than the internal pool
could fill them. The internal pool was cranking on every iostat
snapshot. So, maybe I can just tell Amanda that all the DLE's on the
J4200 are on separate spindles. Let the inparallel regulate things.
--
---------------
Chris Hoogendyk
-
O__ ---- Systems Administrator
c/ /'_ --- Biology & Geology Departments
(*) \(*) -- 140 Morrill Science Center
~~~~~~~~~~ - University of Massachusetts, Amherst
<hoogendyk AT bio.umass DOT edu>
---------------
Erdös 4
|
|
|