dump faster from remote than from local

[Don Lewis]

On  5 Oct, Randy Bush wrote:
> 4.9rc of yesterday
> 
> dump from B->A i.e. from a system on local ether
> 
>   DUMP: DUMP: 2082890 tape blocks on 1 volume
>   DUMP: finished in 478 seconds, throughput 4357 KBytes/sec
>                                             ^^^^^^^^^^^^^^^
> and a local dump on A
> 
>   DUMP: DUMP: 3560987 tape blocks on 1 volume
>   DUMP: finished in 3694 seconds, throughput 963 KBytes/sec
>                                              ^^^^^^^^^^^^^^
> why is local slower than remote?

What are you dumping to on A?  If you are dumping to a file on the same
spindle, you spend a lot more time doing long seeks.  Is there other I/O
occuring on the same spindle as the filesystem that you are dumping on
A?

Could the file system on B have a small number of large files while A
has a large number of small files?  Has the file system on A been run in
a near-full condition for a long period of time so that there are a lot
of disk blocks that are poorly placed?  What results do you get if you
dump each file system to /dev/null on the local machine?