Veritas-bu

[Veritas-bu] Backing up a large amount of small files

2002-11-02 17:05:36
Subject: [Veritas-bu] Backing up a large amount of small files
From: scott.kendall AT abbott DOT com (scott.kendall AT abbott DOT com)
Date: Sat, 2 Nov 2002 16:05:36 -0600
This is a multipart message in MIME format.
--=_alternative 0079916B86256C65_=
Content-Type: text/plain; charset="us-ascii"

i can't say that i have tried this, i just know that it's possible.  it's 
just done by adding the right thing to the file list.  let everyone know 
what the difference in performance is.

if you're looking at a windows netbackup system admin guide, it's page 83, 
"windows nt/2000 disk-image (raw) backup".

it looks like this /\\.\c:

this note from the manual may be something to be aware of... "Before starting a 
disk-image backup, NetBackup locks the logical drive to 
ensure that no changes occur during the backup. If there are open files on 
the logical drive, a disk-image backup is not performed."


- Scott





Johnny Oestergaard <joe AT joe DOT net>
Sent by: veritas-bu-admin AT mailman.eng.auburn DOT edu
11/02/2002 02:23 PM

 
        To:     scott.kendall AT abbott DOT com
        cc:     Bruno.Bossier AT comparex DOT be, veritas-bu AT 
mailman.eng.auburn DOT edu
        Subject:        Re: [Veritas-bu] Backing up a large amount of small 
files


I could live with only a daily backup for DR on the volume with all our 
scanned documents (lots and lots of small files) and then do a normal full 

backup every weekend and new files every day.

I haven't found out how to do the "track by track" backup, but I know that 

flash backup is not supported at the moment on StorageTek SVA9500 and also 

not on Windows.

If I get a real nasty problem I could also just restore the volume to a 
different LUN and then move the file I need back to the production LUN. 
The 
SVA9500 (and V960 and V2X) is a great platform for doing this kind of 
stuff.

Since we only need a volume of less the 100 GB for our image files a total 

restore wouldn't take that long. Compression on that kind of volume with a 

lot of free space should be good so restoring 30-60 MB/s should be 
something we should see (I would hope)

Do you remember where in the policy you tell that you want a raw backup 
and 
not a normal file backup?
We run 4.5

/johnny

At 13:48 02-11-2002 -0600, scott.kendall AT abbott DOT com wrote:

>netbackup by itself can do an image, or block by block, backup.  you just 

>have to use the right thing in the file list. can't remember off the top 
>of my head what it is but it is documented in one of the netbackup 
>docs.  the problem with this is it's really only for full DR.  you can't 
>restore a single file, only the entire volume.
>
>to take a backup block by block, but be able to restore a single file, 
you 
>have to also look at and keep the file to block mapping so you can 
>reference it to pull a single file out of the image.  that's exactly what 

>flashbackup from veritas does.  unfortunately, it's only available on 
>certain platforms and windows isn't one of them yet... but it's supposed 
>to be in the works!!!  I think a lot of us will be looking at this when 
it 
>finally comes out.
>
>
>- Scott
>
>
>
>Johnny Oestergaard <joe AT joe DOT net>
>Sent by: veritas-bu-admin AT mailman.eng.auburn DOT edu
>
>11/02/2002 04:43 AM
>
>         To:        "David A. Chapa" <david AT datastaff DOT com>, 
> Bruno.Bossier AT comparex DOT be
>         cc:        veritas-bu AT mailman.eng.auburn DOT edu
>         Subject:        Re: [Veritas-bu] Backing up a large amount of 
> small files
>
>
>We have the same type of problem and except that we have turned OTM off 
and
>have less files the only way I see that we can solve the problem is by
>dividing the directories into different streems and streem the backup to
>disk (so that it doesn't hold a tapedrive for days)
>Later we will move the backup from disk to tape but that should happen at 
a
>speed that we can handle.
>
>It's great to have so fast tapedrives as we have today (if NT/Win2K could
>pump the data at that speed)
>
>Our files are on a StorageTek SVA9500 and it's still so slow that we cry.
>It's not the disks that are the problem but the time it takes for the
>operatingsystem to open and close the files.
>It would also be a problem on any other operatingsystem (but could be it
>would be a little smaller)
>
>I have been thinking if NBU could take a physical backup of the drive
>(track by track) This should be fast but would also backup empty tracks 
but
>at 30MB/s or more who cares.
>
>/johnny
>
>At 12:03 30-10-2002 -0500, David A. Chapa wrote:
> >Well, I did see something like this at a client and I always found the 
best
> >solution to something like this is to just format the dang thing :-)
> >
> >But Seriously now...
> >
> >What RAID level, 5?
> >
> >RAID5 is great for writes but not for reads, and with as many reads as 
you
> >need
> >done for this backup you would definitely hit timeouts.  This is 
exactly the
> >problem one of my current clients is facing.
> >
> >What about the last time SCANDISK was run?  We found a heavily 
fragmented
> >disk
> >caused these types of timeouts as well.
> >
> >Some of the other things we tried was to break up the server backup at 
the
> >file
> >list.  This was a bit more labor intensive, but it did work to some 
degree.
> >
> >HTH
> >
> >David
> >
> >Quoting Bruno.Bossier AT comparex DOT be:
> >
> > > We have a server with a directory which contains over 3 million 
files 
> for a
> > > total amount of nearly 150 Gb. We are trying to back this up in a
> > > reasonable amount of time (12 to 14 hours maximum). OTM is enabled. 
> We have
> > > not succeeded so far. A lot of time is lost at the beginning of the 
> backup
> > > when the backup is going through the directory structure to check 
all
> > > files. This takes several hours. The first we already did until now 
is to
> > > set the maximum cache size to 0 (unlimited). This stopped the errors 

> we saw
> > > in the Windows eventlog when OTM is started and then aborted after 
40
> > > minutes, then restarted again ....
> > >
> > > Can someone give some suggestions to speed up this backup ?
> > >
> > > Thanks,
> > > Bruno
> > >
> > >
> > > _______________________________________________
> > > Veritas-bu maillist  -  Veritas-bu AT mailman.eng.auburn DOT edu
> > > http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
> > >
> >
> >
> >
> >_______________________________________________
> >Veritas-bu maillist  -  Veritas-bu AT mailman.eng.auburn DOT edu
> >http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
>
>_______________________________________________
>Veritas-bu maillist  -  Veritas-bu AT mailman.eng.auburn DOT edu
>http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
>

_______________________________________________
Veritas-bu maillist  -  Veritas-bu AT mailman.eng.auburn DOT edu
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu



--=_alternative 0079916B86256C65_=
Content-Type: text/html; charset="us-ascii"


<br><font size=2 face="Arial">i can't say that i have tried this, i just know 
that it's possible. &nbsp;it's just done by adding the right thing to the file 
list. &nbsp;let everyone know what the difference in performance is.</font>
<br>
<br><font size=2 face="Arial">if you're looking at a windows netbackup system 
admin guide, it's page 83, &quot;windows nt/2000 disk-image (raw) 
backup&quot;.</font>
<br>
<br><font size=2 face="Arial">it looks like this </font><font 
size=2><tt>/\\.\c:</tt></font>
<br>
<br><font size=2 face="Arial">this note from the manual may be something to be 
aware of... &quot;</font><font size=2><tt>Before starting a disk-image backup, 
NetBackup locks the logical drive to ensure that no changes occur during the 
backup. If there are open files on the logical drive, a disk-image backup is 
not performed.&quot;</tt></font>
<br>
<br>
<br><font size=2 face="Arial">- Scott</font>
<br>
<br>
<br>
<br>
<table width=100%>
<tr valign=top>
<td>
<td><font size=1 face="sans-serif"><b>Johnny Oestergaard &lt;joe AT joe DOT 
net&gt;</b></font>
<br><font size=1 face="sans-serif">Sent by: veritas-bu-admin AT 
mailman.eng.auburn DOT edu</font>
<p><font size=1 face="sans-serif">11/02/2002 02:23 PM</font>
<br>
<td><font size=1 face="Arial">&nbsp; &nbsp; &nbsp; &nbsp; </font>
<br><font size=1 face="sans-serif">&nbsp; &nbsp; &nbsp; &nbsp; To: &nbsp; 
&nbsp; &nbsp; &nbsp;scott.kendall AT abbott DOT com</font>
<br><font size=1 face="sans-serif">&nbsp; &nbsp; &nbsp; &nbsp; cc: &nbsp; 
&nbsp; &nbsp; &nbsp;Bruno.Bossier AT comparex DOT be, veritas-bu AT 
mailman.eng.auburn DOT edu</font>
<br><font size=1 face="sans-serif">&nbsp; &nbsp; &nbsp; &nbsp; Subject: &nbsp; 
&nbsp; &nbsp; &nbsp;Re: [Veritas-bu] Backing up a large amount of small 
files</font></table>
<br>
<br>
<br><font size=2 face="Courier New">I could live with only a daily backup for 
DR on the volume with all our <br>
scanned documents (lots and lots of small files) and then do a normal full <br>
backup every weekend and new files every day.<br>
<br>
I haven't found out how to do the &quot;track by track&quot; backup, but I know 
that <br>
flash backup is not supported at the moment on StorageTek SVA9500 and also <br>
not on Windows.<br>
<br>
If I get a real nasty problem I could also just restore the volume to a <br>
different LUN and then move the file I need back to the production LUN. The <br>
SVA9500 (and V960 and V2X) is a great platform for doing this kind of stuff.<br>
<br>
Since we only need a volume of less the 100 GB for our image files a total <br>
restore wouldn't take that long. Compression on that kind of volume with a <br>
lot of free space should be good so restoring 30-60 MB/s should be <br>
something we should see (I would hope)<br>
<br>
Do you remember where in the policy you tell that you want a raw backup and <br>
not a normal file backup?<br>
We run 4.5<br>
<br>
/johnny<br>
<br>
At 13:48 02-11-2002 -0600, scott.kendall AT abbott DOT com wrote:<br>
<br>
&gt;netbackup by itself can do an image, or block by block, backup. &nbsp;you 
just <br>
&gt;have to use the right thing in the file list. can't remember off the top 
<br>
&gt;of my head what it is but it is documented in one of the netbackup <br>
&gt;docs. &nbsp;the problem with this is it's really only for full DR. 
&nbsp;you can't <br>
&gt;restore a single file, only the entire volume.<br>
&gt;<br>
&gt;to take a backup block by block, but be able to restore a single file, you 
<br>
&gt;have to also look at and keep the file to block mapping so you can <br>
&gt;reference it to pull a single file out of the image. &nbsp;that's exactly 
what <br>
&gt;flashbackup from veritas does. &nbsp;unfortunately, it's only available on 
<br>
&gt;certain platforms and windows isn't one of them yet... but it's supposed 
<br>
&gt;to be in the works!!! &nbsp;I think a lot of us will be looking at this 
when it <br>
&gt;finally comes out.<br>
&gt;<br>
&gt;<br>
&gt;- Scott<br>
&gt;<br>
&gt;<br>
&gt;<br>
&gt;Johnny Oestergaard &lt;joe AT joe DOT net&gt;<br>
&gt;Sent by: veritas-bu-admin AT mailman.eng.auburn DOT edu<br>
&gt;<br>
&gt;11/02/2002 04:43 AM<br>
&gt;<br>
&gt; &nbsp; &nbsp; &nbsp; &nbsp; To: &nbsp; &nbsp; &nbsp; &nbsp;&quot;David A. 
Chapa&quot; &lt;david AT datastaff DOT com&gt;, <br>
&gt; Bruno.Bossier AT comparex DOT be<br>
&gt; &nbsp; &nbsp; &nbsp; &nbsp; cc: &nbsp; &nbsp; &nbsp; &nbsp;veritas-bu AT 
mailman.eng.auburn DOT edu<br>
&gt; &nbsp; &nbsp; &nbsp; &nbsp; Subject: &nbsp; &nbsp; &nbsp; &nbsp;Re: 
[Veritas-bu] Backing up a large amount of <br>
&gt; small files<br>
&gt;<br>
&gt;<br>
&gt;We have the same type of problem and except that we have turned OTM off 
and<br>
&gt;have less files the only way I see that we can solve the problem is by<br>
&gt;dividing the directories into different streems and streem the backup to<br>
&gt;disk (so that it doesn't hold a tapedrive for days)<br>
&gt;Later we will move the backup from disk to tape but that should happen at 
a<br>
&gt;speed that we can handle.<br>
&gt;<br>
&gt;It's great to have so fast tapedrives as we have today (if NT/Win2K 
could<br>
&gt;pump the data at that speed)<br>
&gt;<br>
&gt;Our files are on a StorageTek SVA9500 and it's still so slow that we 
cry.<br>
&gt;It's not the disks that are the problem but the time it takes for the<br>
&gt;operatingsystem to open and close the files.<br>
&gt;It would also be a problem on any other operatingsystem (but could be it<br>
&gt;would be a little smaller)<br>
&gt;<br>
&gt;I have been thinking if NBU could take a physical backup of the drive<br>
&gt;(track by track) This should be fast but would also backup empty tracks 
but<br>
&gt;at 30MB/s or more who cares.<br>
&gt;<br>
&gt;/johnny<br>
&gt;<br>
&gt;At 12:03 30-10-2002 -0500, David A. Chapa wrote:<br>
&gt; &gt;Well, I did see something like this at a client and I always found the 
best<br>
&gt; &gt;solution to something like this is to just format the dang thing 
:-)<br>
&gt; &gt;<br>
&gt; &gt;But Seriously now...<br>
&gt; &gt;<br>
&gt; &gt;What RAID level, 5?<br>
&gt; &gt;<br>
&gt; &gt;RAID5 is great for writes but not for reads, and with as many reads as 
you</font>
<br><font size=2 face="Courier New">&gt; &gt;need<br>
&gt; &gt;done for this backup you would definitely hit timeouts. &nbsp;This is 
exactly the<br>
&gt; &gt;problem one of my current clients is facing.<br>
&gt; &gt;<br>
&gt; &gt;What about the last time SCANDISK was run? &nbsp;We found a heavily 
fragmented<br>
&gt; &gt;disk<br>
&gt; &gt;caused these types of timeouts as well.<br>
&gt; &gt;<br>
&gt; &gt;Some of the other things we tried was to break up the server backup at 
the<br>
&gt; &gt;file<br>
&gt; &gt;list. &nbsp;This was a bit more labor intensive, but it did work to 
some degree.<br>
&gt; &gt;<br>
&gt; &gt;HTH<br>
&gt; &gt;<br>
&gt; &gt;David<br>
&gt; &gt;<br>
&gt; &gt;Quoting Bruno.Bossier AT comparex DOT be:<br>
&gt; &gt;<br>
&gt; &gt; &gt; We have a server with a directory which contains over 3 million 
files <br>
&gt; for a<br>
&gt; &gt; &gt; total amount of nearly 150 Gb. We are trying to back this up in 
a<br>
&gt; &gt; &gt; reasonable amount of time (12 to 14 hours maximum). OTM is 
enabled. <br>
&gt; We have<br>
&gt; &gt; &gt; not succeeded so far. A lot of time is lost at the beginning of 
the <br>
&gt; backup<br>
&gt; &gt; &gt; when the backup is going through the directory structure to 
check all<br>
&gt; &gt; &gt; files. This takes several hours. The first we already did until 
now is to<br>
&gt; &gt; &gt; set the maximum cache size to 0 (unlimited). This stopped the 
errors <br>
&gt; we saw<br>
&gt; &gt; &gt; in the Windows eventlog when OTM is started and then aborted 
after 40<br>
&gt; &gt; &gt; minutes, then restarted again ....<br>
&gt; &gt; &gt;<br>
&gt; &gt; &gt; Can someone give some suggestions to speed up this backup ?<br>
&gt; &gt; &gt;<br>
&gt; &gt; &gt; Thanks,<br>
&gt; &gt; &gt; Bruno<br>
&gt; &gt; &gt;<br>
&gt; &gt; &gt;<br>
&gt; &gt; &gt; _______________________________________________<br>
&gt; &gt; &gt; Veritas-bu maillist &nbsp;- &nbsp;Veritas-bu AT 
mailman.eng.auburn DOT edu<br>
&gt; &gt; &gt; http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu<br>
&gt; &gt; &gt;<br>
&gt; &gt;<br>
&gt; &gt;<br>
&gt; &gt;<br>
&gt; &gt;_______________________________________________<br>
&gt; &gt;Veritas-bu maillist &nbsp;- &nbsp;Veritas-bu AT mailman.eng.auburn DOT 
edu<br>
&gt; &gt;http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu<br>
&gt;<br>
&gt;_______________________________________________<br>
&gt;Veritas-bu maillist &nbsp;- &nbsp;Veritas-bu AT mailman.eng.auburn DOT 
edu<br>
&gt;http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu<br>
&gt;<br>
<br>
_______________________________________________<br>
Veritas-bu maillist &nbsp;- &nbsp;Veritas-bu AT mailman.eng.auburn DOT edu<br>
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu<br>
</font>
<br>
<br>
--=_alternative 0079916B86256C65_=--