Re: [Networker] Balancing load between storage nodes
2006-05-09 04:14:55
Hi Jim,
Before you give up on the "one path to the device down either HBA"
setup, have a look at my post (below) from January. On Solaris it's
simple to create locally named /dev/whatever devices, which you can then
soft link to whichever REAL device you want the data to go to. In your
case I'd recommend something like:
# Traffic down HBA0
ln -s /dev/rmt/0cbn /dev/JB1/Drive1
ln -s /dev/rmt/1cbn /dev/JB1/Drive2
....
ln -s /dev/rmt/4cbn /dev/JB1/Drive5
# Traffic down HBA1
ln -s /dev/rmt/15cbn /dev/JB1/Drive6
ln -s /dev/rmt/16cbn /dev/JB1/Drive7
.....
This setup also means you don't have to re-install (jbconfig) your
jukebox if your devices get re-ordered by a "reboot -- -r" : just point
the soft links to the correct devices, and keep going.
HTH.
Will
Jim Ruskowsky wrote:
Thank you Prasad and everyone else who responded.
I have taken half of my clients and switched the order for Storage Node
Affinity.
(I am grateful for "nsradmin -i" for such a task)
Having two paths was set up because the paths are there, so why not have
some
degree of failover, but seeing how much trouble it is causing, I may end
up
just assigning the drives to one or the other.
My "load balancing" wasn't meant to sent data to a single drive down two
paths
like a disk operation, but to balance the drives as a whole between the
two
available paths in a dynamic way. I may try to change the order I list
the drives
in the jukebox config before abandoning the whole idea. i.e. half the
drives
would get listed with their path down fibre-channel-A first, the other
half would
get their path down fibre-channel-B listed first. It is kind of a shame
to have a
data path available and not be able to take advantage of it in the case of
failure of the other path.
Prasad Chalikonda <prasadchalikonda AT yahoo DOT com> wrote on 05/08/2006
02:42:09 PM:
Hi Balki,
I was given a tip a couple of years ago which has proved fantastic for
dealing with situations like this.
You can call the Tape Devices at the Solaris level anything you like
(they're just links down into the /devices tree). So, our Networker
server is configured with devices called:
/dev/UOL/CMRdrive1
/dev/UOL/CMRdrive2
/dev/UOL/CMRdrive3
/dev/UOL/CMRdrive....
/dev/UOL/HMRdrive1
/dev/UOL/HMRdrive2
/dev/UOL/HMRdrive3
/dev/UOL/HMRdrive....
Where CMR and HMR designate 2 different tape libraries, and "drive1"
designates the drive number within the library.
The joy of configuring Networker this way is that WHEN your tape devices
change Solaris name (/dev/rmt/2cbn becomes /dev/rmt/20cbn), you can just
update some soft links, and not have to touch Networker!
server:ksh:# (11411) ls -l /dev/UOL
total 28
lrwxrwxrwx 1 root other 13 Sep 1 16:01 CMRdrive1 ->
/dev/rmt/0cbn
lrwxrwxrwx 1 root other 13 Sep 1 16:01 CMRdrive2 ->
/dev/rmt/1cbn
lrwxrwxrwx 1 root other 13 Sep 1 16:10 CMRdrive3 ->
/dev/rmt/2cbn
lrwxrwxrwx 1 root other 13 Sep 1 16:10 CMRdrive4 ->
/dev/rmt/3cbn
lrwxrwxrwx 1 root other 13 Sep 1 16:11 CMRdrive5 ->
/dev/rmt/4cbn
lrwxrwxrwx 1 root other 13 Sep 1 16:11 CMRdrive6 ->
/dev/rmt/5cbn
lrwxrwxrwx 1 root other 13 Sep 1 16:11 CMRdrive7 ->
/dev/rmt/6cbn
lrwxrwxrwx 1 root other 14 Sep 1 16:13 HMRdrive1 ->
/dev/rmt/21cbn
lrwxrwxrwx 1 root other 14 Sep 1 16:13 HMRdrive2 ->
/dev/rmt/22cbn
lrwxrwxrwx 1 root other 14 Sep 1 16:14 HMRdrive3 ->
/dev/rmt/23cbn
lrwxrwxrwx 1 root other 14 Sep 1 16:14 HMRdrive4 ->
/dev/rmt/24cbn
lrwxrwxrwx 1 root other 14 Sep 1 16:14 HMRdrive5 ->
/dev/rmt/25cbn
lrwxrwxrwx 1 root other 14 Sep 1 16:14 HMRdrive6 ->
/dev/rmt/26cbn
lrwxrwxrwx 1 root other 14 Sep 1 16:14 HMRdrive7 ->
/dev/rmt/27cbn
server:ksh:# (11412)
This is also useful if you've got access to your tape drives through
multiple HBA's, and wish to direct the IO for a specific tape drive over
a specific HBA. You just change the link to point to the underlying tape
drive on the second HBA.
We've got 3 Solaris servers sharing 14 devices, and this naming scheme
is wonderful for un-confusing things.
HTH
Will
Jim,
First off, I'd recommend a bit of redesign. Any reason
you have two
paths to each drive? Tape drives don't act like disks.
Having two paths
won't load balance I/O. Additionally, it will cause
issues.
After you have configured just 1 path to each tape
drive, you control
which Storage Node a client is backed up to, among
other things, with
what is called - Storage Node Affinity. It is a
resource in the Client
set up. If left at default, that particular client is
backed up by the
NSR Server (Master). To force a client to go to a
Storage Node, put that
name up first, and in case SN fails, it will go to the
NSR Server.
HTH
Prasad.
Jim Ruskowsky wrote:
Hello list
Maybe somebody can offer some advice on the
following situation.
The setup....
We have 10 tape drives in a single library attached
to a fibre switch.
We have two networker servers (master and storage
node) attached to that
same switch - each server has two paths to that same
switch.
The fabric is configured so that each tape drive can
be seen on both paths
on each networker server. We have dynamic drive
sharing licensed for all
10 drives.
So for example, drive #0 has four distinct paths
according to networker
/dev/rmt/0cbn (master server, first fibre
channel)
/dev/rmt/10cbn (master server, second fibre
channel)
rd=jcnetworker2:/dev/rmt/0cbn (storage
node, first fibre
channel)
rd=jcnetworker2:/dev/rmt/10cbn (storage
node, second fibre
channel)
The tape pool "DAILY" is set up as default with no
specific devices
checked off (so all drives should be used)
The problem....
When a savegroup runs, the server only uses the
drives attached to the
master server - ignoring the existence of the
storage node. I ended up
trying to write to 10 LTO3 drives down a single
fibre channel. What is
the best way to load balance between all my paths.
I've tried
specifically checking off specific paths to devices
in the tape pool
setup, but then it just picks a path to the master
server and ignores the
rest.
Thanks for any help.
Jim
Jefferies archives and reviews outgoing and incoming
e-mail. It may be produced at the request of
regulators or in connection with civil litigation.
Jefferies accepts no liability for any errors or
omissions arising as a result of transmission. Use by
other than intended recipients is prohibited.
To sign off this list, send email to
listserv AT listserv.temple DOT edu and type "signoff
networker" in the
body of the email. Please write to
networker-request AT listserv.temple DOT edu if you have any
problems
wit this list. You can access the archives at
http://listserv.temple.edu/archives/networker.html or
via RSS at
http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER
__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
Jefferies archives and reviews outgoing and incoming e-mail. It may be produced at the request of regulators or in connection with civil litigation.
Jefferies accepts no liability for any errors or omissions arising as a result of transmission. Use by other than intended recipients is prohibited.
To sign off this list, send email to listserv AT listserv.temple DOT edu and type
"signoff networker" in the
body of the email. Please write to networker-request AT listserv.temple DOT edu
if you have any problems
wit this list. You can access the archives at
http://listserv.temple.edu/archives/networker.html or
via RSS at http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER
--
w.parsons AT leeds.ac DOT uk
UNIX Support
Information Systems Services
The University of Leeds
+44 113 343 5670
To sign off this list, send email to listserv AT listserv.temple DOT edu and type
"signoff networker" in the
body of the email. Please write to networker-request AT listserv.temple DOT edu
if you have any problems
wit this list. You can access the archives at
http://listserv.temple.edu/archives/networker.html or
via RSS at http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER
|
<Prev in Thread] |
Current Thread |
[Next in Thread>
|
- Re: [Networker] Balancing load between storage nodes, Brian Narkinsky
- Re: [Networker] Balancing load between storage nodes, Jim Ruskowsky
- Re: [Networker] Balancing load between storage nodes,
Will Parsons <=
- Re: [Networker] Balancing load between storage nodes, Fazil Saiyed
- [Networker] Balancing load between storage nodes, Fazil Saiyed
- Re: [Networker] Balancing load between storage nodes, Fazil Saiyed
- Re: [Networker] Balancing load between storage nodes, Gatti
- Re: [Networker] Balancing load between storage nodes, Faidherbe, Thierry
- Re: [Networker] Balancing load between storage nodes, Terry Lemons
- Re: [Networker] Balancing load between storage nodes, Fazil Saiyed
|
|
|