On Mon, 7 Oct 2002 19:48:20 -0500, Willeat, Todd <TWilleat AT MHP.SMHS DOT COM>
wrote:
>I have a question about setting the block size and didn't see this when I
>searched the archives...
>
>We have one server and one storage node (both 6.1.1 running on Win2K).
>Currently, both machine are configured to allow 96K blocks. Each machine
has
>an Exabyte library with Mammoth-2 drives connected to Adaptec SCSI cards.
>
>My question is: Would it beneficial to change to a larger block size?
>According to Exabyte, the M2 can handle up to a 240K block. Since NetWorker
>requires the block size to be a multiple of 32K, this would leave me with a
>maximum possible block size of 224K.
>
>So, should I leave well enough alone, or should I make a change?
There are only two reasons for changing the block size:
(1) For compatibility between platforms, particularly since NetWorker's
default block size for Windows is often 64KB as opposed to more optimum
settings on Unix platforms. If you are already using 96KB block sizes you
have presumably already sorted out the issues with scatter/gather buffers
for your SCSI cards. If your environment is all W2K you don't have a
compatibility problem anyway.
(2) For performance improvements. You can only know whether you will
improve performance by running tests in your own environment using various
block sizes. I have on occasions done this, and found that higher is not
always better. I ran some tests with LTO drives and found no improvement
with higher block sizes over the default of 64KB. The exception was in a
SAN environment where I found that a higher block size partly overcame a
latency problem.
I would suggest that you have little to gain by changing your block size,
but by all means run some tests with bigasm and post your results if you
have the time.
--
Note: To sign off this list, send a "signoff" command via email
to listserv AT listmail.temple DOT edu or visit the list's Web site at
http://listmail.temple.edu/archives/networker.html where you can
also view and post messages to the list.
=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=
|
