SUMMARY: Mirroring Questions Part 1 & 2

From: Chris Edwards <chris_at_decay.com>
Date: Fri Feb 03 2006 - 11:10:24 EST
LOTS of thanks to Darren Dunham, Ken Rossman, Michael Sullivan, Alexander, Alan Epps & Thomas Payerle for their replies.
  This is my first summary post so give me a bit of room to screw it up! :)
  I posted 2 questions on mirroring... and my summary is...
  If you use two diffrent size hard drives to mirror you can reclaim the un-used space on the larger of the two drives, this seems to not be recommended because you will take a performance hit.  
  If you run into the hard limit of 7 partitions/slices on your drive you can use soft partitons for non-critical partitons.  
  When you create your metadb on your mirrored drives you need atleast 3 per drive and they can be in the same partiton on each drive.  Metadb's are small but very important.  
  My nasty little error could be viewed by using the "svcs -x" command which then showed the errors, which was a cannot load console error of some sorts.  It has to do with Solaris 10 smf (Service Management Facility) that is replacing the good ole rc.X booting sequence. I have no knowledge of how this works.  Oh, the error stoped showing up on its own, which I don't like because I want to make sure it doesn't come back.
  I used these two links to do the orginal mirroring...
  http://www.sun.com/bigadmin/content/submitted/svm_mirroring.html
http://slacksite.com/solaris/disksuite/disksuite.html
  Feel free to send me any questions, comments or corrections.
  Chris Edwards
  ---
  Mirroring Part 1
  ---
  I have successfully for the first time mirrored the boot drive on my SunFire V100 (whoop whoop).  Now here is the real question...  My primary drive is 40gb and the mirror drive is 80gb (it was a drive I had laying around.).  Is there anyway for me to use the other 40gb's left on the 80gb drive?  I'm using all 7 slices/partitions on my drive... maybe I can add another slice?  I can't figure out how to take the next step.   Thanks for any info.
  ---
  Mirroring Part 2
  ---
   
Here is my next 2 questions and an error I am now getting.
 
Performance
 
1.  If I mirror with drives of different sizes, root disk being the smallest and the mirror being the largest, will I suffer performance problems?  Both disks have identical partions setup when looking thru the format command expect slice 2.
 
2. On the metadb I have set up 2 30meg slices, one on each drive to contain the metadb, is this enough or should I have more? I may be confused in this area.
 
My nasty little error...  When I reboot the system sometimes it comes up fine then other times it gives me this error... Which I have no clue as to why this is happening.  
 
---
 
LOM event: +36d+20h12m50s host reset
g ...
  Sun Fire V100 (UltraSPARC-IIe 548MHz), No Keyboard
OpenBoot 4.0, 1024 MB memory installed, Serial #57166297.
Ethernet address 0:3:ba:68:49:d9, Host ID: 836849d9.
 
Executing last command: boot
Boot device: rootdisk  File and args:
SunOS Release 5.10 Version Generic_118822-27 64-bit
Copyright 1983-2005 Sun Microsystems, Inc.  All rights reserved.
Use is subject to license terms.
Hostname: gandalf
Jan 31 14:55:54 svc.startd[7]: svc:/system/boot-archive:default: Method "/lib/sv
c/method/boot-archive" failed with exit status 1.
Jan 31 14:55:54 svc.startd[7]: svc:/system/boot-archive:default: Method "/lib/sv
c/method/boot-archive" failed with exit status 1.
Jan 31 14:55:54 svc.startd[7]: svc:/system/boot-archive:default: Method "/lib/sv
c/method/boot-archive" failed with exit status 1.
[ system/boot-archive:default failed (see 'svcs -x' for details) ]
Requesting System Maintenance Mode
(See /lib/svc/share/README for more information.)
Console login service(s) cannot run
Root password for system maintenance (control-d to bypass): Requesting System Ma
intenance Mode
(See /lib/svc/share/README for more information.)
Console login service(s) cannot run
Root password for system maintenance (control-d to bypass):
 
---
  REPLIES:
  --- 
You'd need to reclaim a slice.  Any way you could combine some?  Do you
split everything out (like /var, /opt, /usr, ....)?
  You can use softpartitions in some cases if they're not boot critical.
  Darren Dunham   
---
Are you saying your original boot drive has all 7 slices used?
  You *might* be able to pull off something with soft partitioning on a
few of the higher partitions, depending on what they are used for.
  If you have only a few partitions that could be considered actual boot
environment file system partitions (e.g. root, swap, /usr, /var), then
the rest of the slices could be coalesced into a single partition that
can be soft-partitioned.
  So say you take the top two or three partitions (well, not counting the
metadb partition of course), and combine them into one big partition.
Of course, this would involve downtime and dump/restore type 
operations.
Then make that bigger partition into a soft partition of as many slices
as you need.  This frees up one or more slice locations for additional
slices on the larger secondary disk.
  Sorry if I'm not being clear on this, but I think you'll be able to
figure it out.  I've found soft partitions to be exceptionally useful
in the past, though they can't be used for anything in the critical
boot environment (root, swap, /usr, maybe /var, depending).  Those
need to be hard partitioned for the boot sequence.
  More on this in later msgs, if needed.
  KR
  ---
  I wouldn't recommend it.  You'll likely run into disk performance  issues.
  If you mirror, the disks should be the same geometry, and used for  
nothing
else.
  You can use the remaining space, but again, it's not recommended.
  Mike
  ---
  Hi
  I can mistaken but as far as I know solaris allows only 7 slices so you
can't add one more. 
  Alex
  ---
  Chris,
  The short answer is yes, you can. The longer answer is you can, if you
are willing to rebuild the host. ;-)
  In a nutshell you can use disksuite/Solaris Volume manager to put a
75meg slice on both disks that will hold three copies of the disk
metadata (the three copies are VERY important), a slice for / and a
slice for swap. Then you can soft partition the disk for any additional
partitions you need. The mirroring would be done through disk suite.
  Alan
  ---
  > Performance
>
> 1.  If I mirror with drives of different sizes, root disk being the 
smallest and the mirror being the largest, will I suffer performance 
problems?  Both disks have identical partions setup when looking thru the 
format command expect slice 2.
Not sure, but I doubt anything will be really significant.  Mirroring 
in
general causes some loss of performance on write (must wait for 2 disks 
to
write the data, and even though in parallel (I think), the system 
probably
waits until the last disk is done (e.g. whichever disk had the longest 
way to
go to find the sector).  I think you gain a bit from mirroring on reads 
(either
drive can respond).  Having 2 different sized disks is likely to make 
the
writes even slower, and may lose some of the gain on reads, but I doubt 
iy
will really bother you.
>
> 2. On the metadb I have set up 2 30meg slices, one on each drive to 
contain the metadb, is this enough or should I have more? I may be 
confused in this area.
>
I have some of my recommendations for a cluster I manage available at
http://www.physics.umd.edu/rgroups/pnce/pcs-docs/Glue/disksuite/disksuite-dbmanagement.html
  30MB is fine.  I typically use 50MB, but both are overkill.  In Sol9, 
the
DB replica size jumped from 0.5MB to about 4MB.  You should put a 
couple copies
(I usually do 3) on each slice you are giving over to replication.  I 
presume
your 30MB does nothing but hold the metadb replicas; since your wasting 
a 
whole 30MB for that, you may as well put a couple on their.  I do not 
know
how much protection you gain from that redundancy, as is on the same 
slice,
but should gain some small protection from random cosmic rays, etc.
  There is, however, a possible suggestion for you.  Metadb replicas can 
coexist with data on a metadisk slice.  You can't use this for your 
standard
mirroring of the root/boot drives (since typically you install to non 
mirrored,
then mirror slices that are already mounted).  But you could, for 
instance,
make the metadb slice on the 80GB drive to be 40GB, put the replicas on 
that
slice, then make a metaslice on it, and mount the metaslice somewhere.  
The
metaslice is not mirrored or anything, but you don't have disk space to 
mirror
anyway, and can be used.  The metadbs will just silently take some 
space away
from the filesystem on the slice.
  (Alternatively, you could either use slice 2 to just be the extra 40GB.  
I asked
about using slice 2 like that once, and the response was varied.  MY 
weighted
opinion of the responses was that it is doable, but may run into some 
problems
with code that expects slice 2 to be the full disk (backup code may be 
a 
particular problem).  I decided that was probably safe on data disks, 
but not
worth it on root/system disks).
  > My nasty little error...  When I reboot the system sometimes it comes 
up fine then other times it gives me this error... Which I have no clue 
as to why this is happening.
>
Adding disksuite to a system you will typically get warning messages 
early in
bootup about failure of force load of some disksuite components (e.g. 
for
raid, hotspares, etc.).  This have been declared "ignorable" by Sun.
  I am not sure what is problem below, but sounds serious.  Do not ignore 
it.
  Tom Payerle
  ---
   
  
 


--- 
Chris Edwards
VP of Rumor Control & Misinformation
Living Reefs - http://www.livingreefs.com
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers
Received on Fri Feb 3 11:11:12 2006

This archive was generated by hypermail 2.1.8 : Thu Mar 03 2016 - 06:43:55 EST