Veritas-bu

[Veritas-bu] iostat during duplication - strange

2002-04-03 14:38:06
Subject: [Veritas-bu] iostat during duplication - strange
From: jonw AT n2h2 DOT com (Jon Walton)
Date: Wed, 3 Apr 2002 11:38:06 -0800 (PST)
I do not have multiplexed originals, and reads match writes exactly.

Is it possible your originals are multiplexed?   If so, and you are
demultiplexing during the duplication, you will most definitely read more
that you write.

Regards,

Jon

On Wed, 3 Apr 2002, Donaldson, Mark wrote:

> So here I am, testing the I/O characteristics of my new buffer settings for
> duplications.  I'm doing an "iostat" and I see the following for the two
> tape drives that are source and destination for this dup.
> 
> $ sudo iostat -xtnc 5 | egrep "rmt.[01]|wsvc"
>   r/s  w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>  32.0  0.0 8192.1    0.0  0.0  1.0    0.0   31.1   0 100 rmt/0
>   0.0 14.0    0.0 3584.0  0.0  0.6    0.0   45.4   0  63 rmt/1
>   r/s  w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>  31.2  0.0 7986.9    0.0  0.0  1.0    0.0   31.9   0  99 rmt/0
>   0.0 16.8    0.0 4300.6  0.0  0.7    0.0   38.7   0  65 rmt/1
>   r/s  w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>  31.0  0.0 7935.8    0.0  0.0  1.0    0.0   32.1   0 100 rmt/0
>   0.0 16.4    0.0 4198.3  0.0  0.5    0.0   32.8   0  54 rmt/1
> 
> It's easy to see which is source and which is destination.  The strange
> thing, consistent over about 10 minutes of this, is the nearly 2-to-1
> difference between read volume (kr/s) & write volume (kw/s).  
> 
> My first thought was blocking factors for read and write were different but
> the iostat man page says these columes are KBytes/sec not blocks/sec.  
> 
> Next I speculated that it was read-ahead buffering but eventually the
> buffers would be full and I'd see the volume turn around - which I haven't.
> 
> 
> I've verified that this is the only source and destination tapes in use -
> the other tapes in other drives are doing other work.
> 
> tpconfig has each of these devices with the same settings, /dev/rmt/[01]cbn.
> I would expect input volumes to match output volumes since the compression
> is taking on the device.
> 
> So the new quandry - why doesn't the read volume match the write volume?
> 
> BTW: Sol 2.6 master/media, VxNB 3.4.1.
> 
> -----------------------------------------------------------------------
>    Mark Donaldson - Sr. Systems Engineer
>    Experian EMS - Denver Colorado
> -----------------------------------------------------------------------
>