dump(8) && -b option

[Warren Block]

On Mon, 3 Feb 2014, Matthias Apitz wrote:

>
> Hello,
>
> Last weekend I was playing around with the dump options to improve the
> dump speed to my external USB drive. I tried using the -b option to
> increase the blocksize of the written junks:
>
> # dump -0auL -b 100 -f - /usr
> ...
>  DUMP: finished in 40 seconds, throughput 7485 KBytes/sec
>
> # dump -0auL -b 10 -f - /usr
> ...
>  DUMP: finished in 70 seconds, throughput 4274 KBytes/sec
>
> with -b 1000 the dump hangs forever:
>
> # dump -0auL -b 1000 -f - /usr
> ...
>  DUMP: dumping (Pass III) [directories]
> (hangs)
>
> on of its child is doing this where fd 3 is the device to be backed up:
>
> pread(0x3,0x28809000,0x1000,0x89909000,0xa5,0xbfbfcc88) ERR#5 'Input/output error'
> pread(0x3,0x28809000,0x1000,0x89909000,0xa5,0xbfbfcc88) ERR#5 'Input/output error'
> pread(0x3,0x28809000,0x1000,0x89909000,0xa5,0xbfbfcc88) ERR#5 'Input/output error'
> pread(0x3,0x28809000,0x1000,0x89909000,0xa5,0xbfbfcc88) ERR#5 'Input/output error'
> pread(0x3,0x28809000,0x1000,0x89909000,0xa5,0xbfbfcc88) ERR#5 'Input/output error'
>
> Why is this?

Don't know, but I remember searching and finding reports of problems 
with -b values larger than 64.  Those reports are years old, and 
unfortunately I can't find them now.

> Maybe the best is doing the write to the disk with dd(1), for example
> with
>
> # dump -0auL -f - /usr  | dd bs=8m > /dev/da0s1a/usr.dmp

Could make for interesting benchmarks.  My guess is that the benefits 
would taper off after a smaller buffer size than that.  Using -b64 at 
the same time would probably help reduce overhead.