Hi Geoff,
I don’t think you told us whether the physical server is Windows or AIX, I’m
writing here from my experience with Windows.
When you install the journal engine, it does create a separate journal DB for
each drive that you want journaled.
I have never installed multiple instances of the journal.
The “journal engine” doesn’t actually participate in the backup. Yep, it isn’t
part of the backup at all.
What the journal engine does is invoke a Windows function that monitors file
system activity and makes a list of the files that have changed. (In the
journal DB). So when the backup runs, it is essentially doing a “dsmc
selective –filelist=filea,fileb,filec, etc”, getting that list of files from
the journal DB.
So the journal engine doesn’t play into the speed of the backup.
On a big fileserver (70+ million files, say), the change rate is usually very,
very low. So getting the backup done is usually pretty trivial, when all you
are doing is backing up the new/changed stuff and you don’t have to traverse
the filetree(s).
So backing up “quicker” doesn’t’ have much to do with how many journal engines
you have, AFAIK.
With the journal engine operating, it’s just about how much data you have to
move during the backup, no extra time traversing the file tree.
That being said, there ARE serious reasons to make lots of LUNS.
· The biggest one: when you use the journal engine, IBM still
recommends that you do a periodic backup with –nojournal, to pick up things
that the Windows monitoring function the journal engine uses has missed. So
periodically, you still have to do a dsmc incr that traverses the filetree. My
experience with Win2K3 was that over 70,000,000 files and it becomes nearly
impossible to traverse the filetree in less than 24 hours. You want to use
something like Windows dfs or mounted drives to make the directory trees look
“nice” and rational for the users, but actually have the directories spread
across smaller separate LUNS which can’t have more than a cazillion files each.
· If you have multiple LUNS, you can set resourceutilization=10 and have
up to 4 pairs of backup sessions running at once
· If something invalidates the journal (and something will), you’ve only
invalidated the journal on part of your backups, so you don’t have to scan the
whole filetree to revalidate it (and you will have to revalidate the journal
for a LUN, at some time)
· Think about restores. Journaling helps you back up, it doesn’t help
you restore. The bigger the LUN, the harder/longer it takes to put it back.
You can get two 1 TB luns restored in the almost the same time as 1, with
multiple restore streams.
I can’t think of any reason you’d need to install multiple journal services,
unless the change rate on the filesystem is too high for 1 journal service to
keep up with it. And if you look at the parms for the journal service, there
are buffering values and other misc stuff to tune the engine so that it can
better keep up with the change rate, before resorting to multiple services.
I’m sure other folks will have different experiences to share.
W
From: avalnche96@ [mailto:yahoo.com avalnche96 AT yahoo DOT com]
Sent: Tuesday, July 31, 2012 5:03 PM
To: Prather, Wanda
Subject: Re: RE: [ADSM-L] Multiple journal engines on a single server
I understand Wanda. The Customer does not want us to use ndmp so they are
moving some data to a physical server so we can use journal.
Sent from my LG Thrill™ 4G smartphone with glasses-free 3D on AT&T
------ Original Message ------
From : Prather, Wanda
To : Geoff Gill;
Sent : 7/31/2012 14:29
Subject : RE: [ADSM-L] Multiple journal engines on a single server
Easy answer, you can't use journaling on a NAS, as the client can't be
installed there.
If it's a Netapp, use snapdiff, solves the problem easily.
If it's a non-Netapp NAS, you either suffer through NDMP, or you set up proxy
relationships and let clients back up the shares via CIFS.
-----Original Message-----
From: ADSM: Dist Stor Manager [mailto:ADSM-L AT VM.MARIST DOT
EDU]<mailto:[mailto:ADSM-L AT VM.MARIST DOT EDU]> On Behalf Of Geoff Gill
Sent: Tuesday, July 31, 2012 10:36 AM
To: ADSM-L AT VM.MARIST DOT EDU<mailto:ADSM-L AT VM.MARIST DOT EDU>
Subject: [ADSM-L] Multiple journal engines on a single server
Hi All,
Sometimes when I read things that make sense it causes me to question if it
really works so I thought I'd throw this out there to see if anyone is doing it.
I've read that you can put up multiple journal engines on a single server and
I'm wondering if anyone has tested it and I'm also curious if you have decided
if it has any advantages or disadvantages. I was thinking, because NDMP has
been squashed for a specific customer, I need to find out the best of what's
left. Hearing statements like, "we want to move all the data curretly x number
of servers currently access on the NAS to a single server", it makes me
question how we're going to handle this single backup.
I currently can't tell you how many millions of files we're talking about nor
can I say how much data we're talking about, but it seems to me that it would
make sense to create multiple drives to spread it out to be able to use
multiple journal engines to track everything, and I was hoping it might make it
quicker. One other question is if it would be better to schedule seperate
backup windows for those different drives to help spread things out and get it
backed up "at least somewhat timely".
Any suggestions would be welcome.
Thank You
Geoff Gill
|
