Veritas-bu

Re: [Veritas-bu] NUMBER data buffers

2011-10-18 13:43:35
Subject: Re: [Veritas-bu] NUMBER data buffers
From: "David McMullin" <David.McMullin AT CBC-Companies DOT com>
To: "veritas-bu AT mailman.eng.auburn DOT edu" <veritas-bu AT mailman.eng.auburn DOT edu>
Date: Tue, 18 Oct 2011 13:43:23 -0400
If you do windows backups, you need to also multiply # of drives with "all 
local drives" - essentially how many 'child' processes can be started by your 
parent job as well. Figure - how many jobs are running at once? Each job will 
get the memory allocated to it.

Buffer settings?

It will depend on your drives and configuration - the speed of your drives, 
your hba and TAN infrastructure.


I had LTO2 drives, my buffer size was limited. I moved to LTO5 - I increased my 
buffer values.
I have done some testing - backing up the same file with different numbers of 
buffers, and be aware it can be counter intuitive - sometimes a lower number of 
buffers gets you more throughput.

64K - larger number seems better.
128K - sweet spot at 64 buffers
256K - sweet spot at 96 buffers
512K - sweet spot at 16 and 64 buffers!

Best throughput at 64 X 128K buffers!

However - I write to an VTL and duplicate to tape - and the duplication process 
is 'stuck' with the original buffer size - so duplicating backups written at 
64K buffers is painfully slow in comparison to 256K buffers.

I am still seeking a "professional opinion" on optimal buffer size for LTO5 
drives.

I have one media server I had to set to 8 X 64K buffers, to get it to slow 
below 20M/Second per channel, due to slow storage - I was crushing the 
application.


Here is a spreadsheet I built, my speed is mostly limited by the source storage 
disks and the fiber/switch TAN 
                        

Number Size             Total   KB per  waited/delayed KB written
Buffer Buffer   Second Second   buffer
8       65536           605     20,904  0       0       12,246,528 - cancelled 
- too slow
16      65536           604     46,801  0       0       27,494,656 - cancelled 
- too slow
32      65536           605     74,268  0       0       54,272,032
64      65536           757     73,498  25K     25K     54,272,032
96      65536           360     160,233 106     243     54,272,032
128     65536           463     120,826 48      154     54,272,032
                                                                
8       131072  953     45,135  0       0       42,022,400
16      131072  683     81,455  26K     26K     54,272,032
32      131072  422     135,666 7K      7K      54,272,032
64      131072  280     205,184 55      128     54,272,032
96      131072  296     193,573 28      110     54,272,032
128     131072  358     160,427 12      41      54,272,032
                                                                
8       262144  695     80,014  26K     26K     54,272,032
16      262144  336     170,204 1K      1K      54,272,032
32      262144  293     196,601 62      206     54,272,032
64      262144  298     194,335 14      38      54,272,032
96      262144  295     199,338 10      33      54,272,032
128     262144  312     182,813 1       43      54,272,032
                                                                
16      524288  286     204,707 43      93      54,272,032
32      524288  293     196,913 25      86      54,272,032
64      524288  287     204,287 0       0       54,272,032
96      524288  294     194,663 1       6       54,272,032

-----Original Message-----

Message: 1
Date: Tue, 18 Oct 2011 08:49:19 -0500
From: Heathe Yeakley <hkyeakley AT gmail DOT com>
Subject: [Veritas-bu] NUMBER data buffers
To: NetBackup Mailing List <veritas-bu AT mailman.eng.auburn DOT edu>
Message-ID:
        <CAAWsBU5Qdsi-Kew8fWE=K3ye+6rqn7-kXKZHbwVmqGAr-eobWw AT mail.gmail DOT 
com>
Content-Type: text/plain; charset="iso-8859-1"

So I've read the tuning guide, I've played around with different options for
SIZE and NUMBER of buffers and I understand the formula of SIZE * NUMBER *
drives *MPX as it relates to shared memory.

Here's my question. Of the four parameters:

MPX level

# of drives (I have 12 drives)

NUMBER of buffers

SIZE of buffers (must be multiple of 1024 and can't exceed the block size
supported by your tape or HBA)

The NUMBER of buffers and MPX level seem to be the two variables here. I
have MPX set pretty low (2 or 3) and NUMBER of buffers set to either 16 or
32. When I multiply it all out, I get a hit on my shared memory of less than
a GB. My media servers are dedicated linux hosts that only function as media
servers and that's it. Furthermore, they each have somewhere around 35 - 50
GB of memory a piece.

With my current configuration, I'm not even scratching the surface of the
amount of shared memory that's sitting idle in my system while my backups
run at night. Is there any reason I *shouldn't*** jack the NUMBER of data
buffers up to... say... 500? 1000? I've seen some people mention that they
have the number of buffers set to 64, but can we go higher?

I've searched around to see if there's a technote on the upper limit of the
NUMBER buffers parameter. If there is such a tech note, I can't find it.

Any ideas?

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

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