Veritas-bu

[Veritas-bu] 2 tape drives issue

2003-03-14 15:02:51
Subject: [Veritas-bu] 2 tape drives issue
From: dbase77 AT yahoo DOT com (Feroz F. Basir)
Date: Fri, 14 Mar 2003 20:02:51 +0000 (GMT)
Hi,

This is weird. I changed buffer size to default 32k
and no. of buffer to 8. And I run my restore again.
This time it only took about 30 minutes to restore 35
Gb data. That is half of the actual total size of 70
GB. Tell me something, how buffer size can affect the
restore? I mean increase it good for backup but very
bad for restore. Small value  in buffer size is very
good for restore. I am very confuse here. 

Here is my setting:

buffer size 256K, num of buffer 16. Give backup time
to 38 minutes for 74 GB oracle data. Restore half of
the size (35 GB) give me 3 hrs.

buffer size 32K, num of buffer 8. Give backup time 5
hrs. Restore half of the size (35 GB) give me 30
minutes only.

I hope somebody can explain the reason behind this. I
mean what is the best setting I can use in my
scenario.

Another thing is, bptm log. where about is this log
file? can you give me the actual log file name. i
tried to search whole system, no luck at all. am i
missing something here or what?

Thank you again.

regards,
feroz 

 --- "Donaldson, Mark"
<Mark.Donaldson AT experianems DOT com> wrote: > Restores
always take longer than backups; there's
> probably a good reason but
> I've never delved too deeply into it.  I plan
> approximately (1.5 *
> backuptime) to estimate restore time.  There's a
> "Performance & Tuning"
> guide you can download from Veritas if you wish. 
> Indexing the image
> database is a suggestion for enhancing restore
> speeds.
> 
> Multiplexed backups take longer to restore and/or
> duplicate, too.
> 
> I thought there was a separate buffer setting for
> restores but I'm having
> trouble finding a reference so I may be remembering
> wrong.
> 
> To verify your buffer settings, search the bptm log
> file for lines like
> this:
> [2851788] <2> io_init: setting send network buffer
> to 262144 bytes
> [2851788] <2> io_init: using 32 data buffers
> [2851788] <2> io_init: buffer size for read is
> 262144
> 
> -M
> 
> -----Original Message-----
> From: Feroz F. Basir [mailto:dbase77 AT yahoo DOT com]
> Sent: Friday, March 14, 2003 6:10 AM
> To: Donaldson, Mark
> Cc: Veritasbu (E-mail)
> Subject: RE: [Veritas-bu] 2 tape drives issue
> 
> 
> Hi,
> 
> OK, I increase my buffer size to 256K and no. of
> buffer is 16. Good news is, backup took 38 minutes
> for
> 74 GB oracle datafile. Very happy. But not yet as
> restore took longer though, in fact it still running
> after past 1 hr mark. What am I doing wrong here? Am
> I
> not allow to have win:win situation here? Or
> win:lose
> situation? Do you think I can have balance
> situation? 
> 
> Anybody want to share their experience regarding
> this
> issue?
> 
> Thank you in advance.
> 
> regards,
> feroz 
> 
> --- "Donaldson, Mark"
> <Mark.Donaldson AT experianems DOT com>
> wrote: > Drive compression, in Unix, is controlled
> by
> the
> > major & minor numbers
> > associated with the device files.  These major &
> > minor numbers, are
> > therefore referenced by the filename. 
> > 
> > For solaris, look for the /dev/ entry for your
> > drives ending in "cbn", ie:
> > /dev/rmt/0cbn.  The "c" in "cbn" stands for
> > "compressed".  If you use that
> > name in your storage unit setup (tpconfig -l), you
> > should get compression at
> > the drive.
> > 
> > Yes, I'd definitely increase the 32K to at least
> 128
> > & maybe 256K.  Don't
> > screw with the number of buffers unless you're not
> > getting sufficient
> > throughput - both changes can increase memory
> usage
> > so tweak them up slowly.
> > The LTO drive wants big blocks of data for better
> > throughput & compression.
> > Feeding it 32K teaspoonfuls will starve it.
> > 
> > Using both drives will speed backup times but you
> > have to plan for it in
> > policy creation, storage units, pre- &
> post-schedule
> > jobs, etc.  I backup
> > nearly 350G of oracle data nightly (direct scsi,
> no
> > network) using four
> > streams to two DLT drives (7-8 MB/sec max).  It
> > takes about 6.5 hours.  I
> > use a NUMBER_DATA_BUFFERS of 64 and
> > SIZE_DATA_BUFFERS & NET_BUFFER_SIZE of
> > 262144 (256K).
> > 
> > HTH - Mark
> > 
> > -----Original Message-----
> > From: Feroz F. Basir [mailto:dbase77 AT yahoo DOT com]
> > Sent: Thursday, March 13, 2003 10:27 AM
> > To: Donaldson, Mark;
> > veritas-bu AT mailman.eng.auburn DOT edu
> > Subject: RE: [Veritas-bu] 2 tape drives issue
> > 
> > 
> > Hi,
> > 
> > Thank you for replying. How do I go about turn on
> > the
> > LTO drive compression, as these drives are attach
> to
> > L25 tape library at the back. Any command I can
> use?
> > I
> > also going to uncheck the netbackup compression
> see
> > how it goes tonite. Regarding the buffer setting,
> > I'm
> > using the default which I think 32K. Do you think
> > this
> > buffer setting cause the extra time? I attached
> the
> > tape library to our SUN box and ONLY backup data
> > from
> > the SUN. No network backup or so ever. I'm very
> > suprise about the time that I have here.
> > 
> > What negative impact can I get if I use bigger
> > buffer
> > size? And if I use both drives at the same how's
> the
> > writing work? Does it write at the same time or
> one
> > after another?
> > 
> > I'll let you know the outcome tomorrow morning.
> > 
> > Thank you.
> > 
> > regards,
> > feroz
> > 
> >  --- "Donaldson, Mark"
> > <Mark.Donaldson AT experianems DOT com> wrote: > Don't
> use
> > the Netbackup client compression.  It's
> > > software based and causes
> > > a high client load, slowing the backups.  Your
> LTO
> > > drive is capable of
> > > hardware compression, use that.
> > > 
> > > LTO is good for about 15 MB/sec native, perhaps
> > > 25-30MB streaming rates if
> > > you can get the data to it fast enough & the
> data
> > is
> > > decently compressible.
> > > So 74G at 15 MB/sec is only 84 minutes so you're
> > > really underperforming. 
> > > 
> > > Check your buffer sizes.  For my LTO drives, I'm
> > > using 256K buffers & it
> > > makes a big performance difference.  Check this
> > > whitepaper for a description
> > > buffer sizes & counts.  Make sure to set the
> > buffers
> > > for both client & media
> > > server. 
> > > http://seer.support.veritas.com/docs/183702.htm 
> > If
> > > your system is
> > > Windows, check out this instead:
> > > http://seer.support.veritas.com/docs/244652.htm
> > > 
> > > Also consider the path from client to media
> > server. 
> > > If it's standard
> > > ethernet, it's only capable of about 10MB/sec,
> and
> > > your drive will be
> > > underutilized.  Consider Gigabit ethernet if
> > > possible.
> > > 
> > > That should be enough to start with - post back
> > any
> > > results you get.
> > > -M
> > > 
> > > -----Original Message-----
> > > From: Feroz F. Basir [mailto:dbase77 AT yahoo DOT com]
> 
=== message truncated === 

__________________________________________________
Do You Yahoo!?
Everything you'll ever need on one web page
from News and Sport to Email and Music Charts
http://uk.my.yahoo.com

<Prev in Thread] Current Thread [Next in Thread>