Hi Michael,
Thanks for your detail explanation. Really appreciate.
Cheers,
Chadwick
>From: "Michael Miller" <mmlist AT mjmm DOT org>
>To: "Chadwick Leung" <poleetea AT hotmail DOT com>,
>veritas-bu AT mailman.eng.auburn DOT edu
>Subject: RE: [Veritas-bu] LTO2 backup speed just up to 5 MB/sec under
>Netbackup 4.5
>Date: Mon, 15 Mar 2004 13:27:54 -0000 (GMT)
>
>Hi
>
>on solaris the system call is failing due to the large size of the shared
>memory NBU is trying to allocate. The buffers are allocated from shared
>memory. You need to have atleast the following amount of shared memory
>available:
>
>SIZE_DATA_BUFFERS * NUMBER_DATA_BUFFERS *
>CONCURRENT_ACTIVE_JOBS_ON_THIS_MEDIA_SERVER
>
>So, if the size is 64k and number is 16 and you have 2 drives with mpx of
>8 you would calculate shared memory as:
>
>64k * 16 * (2 * 8) = 16384k = 16MB
>
>To tune this you modify /etc/system shm values, we run for example:
>
>* Shared memory
>set shmsys:shminfo_shmmax=1073741824
>set shmsys:shminfo_shmmin=1
>set shmsys:shminfo_shmmni=220
>set shmsys:shminfo_shmseg=100
>
>
>You may need to change semsys or msgsys parameters too. you need to
>reboot for the /etc/system changes to take effect. As you can see, our
>largest shared memory segment is 1GB on the box I took the above config
>from.
>
>Good luck. I think with LTO drives,you may be able to push the
>data_buffer_size to 256k (262144 bytes). And crank the buffers to, say,
>32. That would require 8MB of shared memory PER ACTIVE backup. Notice
>that your drive could write that buffer to tape in less than a second, so
>if you multiplex 4 clients/streams, your buffer requirements would just to
>32MB.
>
>Good luck... Just keep cranking the buffers up till your performance tails
>off. If your server is purely a netbackup media server, you can safely aim
>to use half (or maybe more even...) the machines ram for buffers...
>
>The whole idea is to keep the drive streaming data, when the drive has to
>stop and backtrack and restart writing due to lack of data, your
>performance will suck.
>
>Enable logging for the bptm process on the media server and look for
>waited for empty / full buffers. Waiting for full buffers means your tape
>is writing faster than the media server can source data... if you are
>waiting for empty buffers you are sourcing more data than you can write to
>tape... try to avoid waiting for full buffers by incresasing the number of
>buffers available.. and maybe try multiplexing multple clients together.
>
>good luck!!
>
> > Hi Douglas and Mark,
> >
> > Thanks for your input.
> >
> > I have created the files SIZE_DATA_BUFFERS and NUMBER_DATA_BUFFERS with
> > value of 64512 and 16 respectively, the throughtput has a significant
> > improvement at about
> > 19 MB /sec .
> >
> > I have tried the following combinations but it seems the above values
>pair
> > is the best:
> >
> > SIZE_DATA_BUFFERS 32768 64512 64512 64512 131072 131072 262144
> > 262144
> > NUMBER_DATA_BUFFERS 16 8 16 24 4 8 2
> > 4
> > NET_BUFFER_SZ 32768 64512 64512 64512 131072 131072 262141
> > 262141
> > THROUGHPUT (KB/sec) 9662 9721 18758 failed 9887 failed 9580
> > failed
> >
> > I am still looking for an optimum value pairs.
> >
> > For those failed cases, I got the following errors:
> >
> > 11:27:13.070 [19805] <16> io_init: system call failed - Invalid argument
> > (at
> > tmcommon.c.1976)
> > 11:27:13.082 [19805] <16> io_init: could not allocate enough shared
>memory
> > for backup buffers
> >
> > How I can adjust the system shared memory in this case?
> >
> > Regards,
> > Chadwick
> >
> >
> >>From: Douglas Ritschel <dritschel62 AT yahoo DOT com>
> >>To: mark <mark AT steelfamily DOT org>, Chadwick Leung <poleetea AT hotmail
> >>DOT com>,
> >>veritas-bu AT mailman.eng.auburn DOT edu
> >>Subject: RE: [Veritas-bu] LTO2 backup speed just up to 5 MB/sec under
> >>Netbackup 4.5
> >>Date: Sun, 14 Mar 2004 07:03:50 -0800 (PST)
> >>
> >>You can increase the blocksize that netBackup uses, by
> >>creating a file:
> >>/usr/openv/netbackup/db/config/SIZE_DATA_BUFFERS. The
> >>file should contain one line with powers of 2 times
> >>1024.
> >>
> >>For example, 256 times 1024:
> >>
> >>262144
> >>
> >>You can also increase the number of data buffers by
> >>creating a file:
> >>
> >>/usr/openv/netbackup/db/config/NUMBER_DATA_BUFFERS
> >>
> >>It should contain one line which is a multiple of 2,
> >>the max is 256.
> >>
> >>Also can increase the fragment size of the storage
> >>unit to 2000.
> >>
> >>
> >>
> >>
> >>--- mark <mark AT steelfamily DOT org> wrote:
> >> > On 4.5 the limitation of a backup from local disk to
> >> > a tape is bpbkar which
> >> > has a block size of 32kb internally. this is fixed
> >> > and cannot be tuned,
> >> > whereas tar or cpio you can set a much larger
> >> > blocksize.
> >> >
> >> > in netbackup 5.0 this blocksize is 128k, but still
> >> > not tuneable. best thing
> >> > is to get multiple bpbkar processes running in
> >> > parallel by breaking the
> >> > filelist into streams of drives / mounts or files -
> >> >
> >> > ie rather than backing up
> >> >
> >> >
> >> > /usr
> >> >
> >> > run
> >> >
> >> > /usr/*
> >> >
> >> > or
> >> >
> >> > /usr/[a-m]* ( i think you can do this on unix
> >> > clients, but not windows )
> >> > /usr/[n-z]*
> >> >
> >> > and you will get a different backup stream for each
> >> > directory. This isn't a
> >> > good example on /usr - you have to make sure the
> >> > spec you pick doesnt
> >> > generate too many streams which are very small, and
> >> > also make sure the
> >> > wildcard spec has no 'gaps' otherwise some point
> >> > down the line you might
> >> > find you haven't been backing everything up.
> >> >
> >> > regards
> >> > mark
> >> >
> >> >
> >> > -----Original Message-----
> >> > From: veritas-bu-admin AT mailman.eng.auburn DOT edu
> >> > [mailto:veritas-bu-admin AT mailman.eng.auburn DOT edu]On
> >> > Behalf Of Chadwick
> >> > Leung
> >> > Sent: 14 March 2004 06:23
> >> > To: veritas-bu AT mailman.eng.auburn DOT edu
> >> > Subject: [Veritas-bu] LTO2 backup speed just up to 5
> >> > MB/sec under
> >> > Netbackup 4.5
> >> >
> >> >
> >> > Greeting!
> >> >
> >> > I have installed a Sun L8 autoloader which replaces
> >> > a Sun L9 model . The L8
> >> > has been successfully installed and can operate
> >> > without problem. However,
> >> > the average throughput is just up to 5 MB/sec under
> >> > Netbackup 4.5.
> >> >
> >> > I tried to TAR a 1.2 GB file direct to the LTO2
> >> > tape, it took about a
> >> > minute. Under Netbackup, backing up the same file
> >> > take about 4 mins. 30 sec.
> >> >
> >> > Any idea?
> >> >
> >> > Best Regards,
> >> > Chadwick Leung
> >> >
> >> >
> >>_________________________________________________________________
> >> > Get 10Mb extra storage for MSN Hotmail. Subscribe
> >> > Now!
> >> > http://join.msn.com/?pgmarket=en-hk
> >> >
> >> > _______________________________________________
> >> > 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
> >>
> >>
> >>__________________________________
> >>Do you Yahoo!?
> >>Yahoo! Mail - More reliable, more storage, less spam
> >>http://mail.yahoo.com
> >
> > _________________________________________________________________
> > Linguaphone : Learning English? Get Japanese lessons for FREE
> > http://go.msnserver.com/HK/30476.asp
> >
> > _______________________________________________
> > Veritas-bu maillist - Veritas-bu AT mailman.eng.auburn DOT edu
> > http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
> >
>
_________________________________________________________________
Linguaphone : Learning English? Get Japanese lessons for FREE
http://go.msnserver.com/HK/30476.asp
|
