Re: no improvement using buffer with tar?

[servis@purdue.edu]

*- On  3 Feb, Jason Gunthorpe wrote about "Re: no improvement using buffer with tar?"
> 
> On Wed, 3 Feb 1999 servis@purdue.edu wrote:
> 
>> I guess I was a little disappointed with respect to its claims and its
>> actual performance.  I would like to speed up my backup, as it is now a
>> backup of my system containing about 5.5Gigs takes almost 5.5hrs. That
>> is only about 275k/s transfer rate which is way below the capabilities
>> of the drive. Which leads me to believe there can be improvements on the
>> software side of my backup.
> 
> Listen to the tape while it is running, it should basically emit a nice
> constant sound. If it doesn't and you hear it 'back up' then that is why

Yep, it does that.

> things are so slow for you. It may well be that you bought a tape drive
> that requires a higher data rate than your PCee can sustain, this is what
> buffer is supposed to help with. Make a -really- big buffer (like 10-50
> meg)  and set the low/high marks at the far end so that it generates a big
> wack meg of data and then starts feeding the tape.. Now if your disks and

Buffer would only accept a 20 meg maximum buffer.  How can I increase
the shared memory so that it will use more, I have 96 meg of ram?  The
backups are done at night so nothing is active so it should be able to
swap other things out.  I know nothing about shared memory. 

> your CPU (compression) can mostly keep up with the tape then you will be
> fine with this, but if the tape vastly outmatches your systems performance
> (ie it can stream 2meg/sec or something crazy like that) then you will
> see little or no speed gain and possibly even a loss. 
> 

I have a P233 and the specs on the scsi-2 tape drive are:

Transfer Rate (Kbytes/sec)		600/450/300
 Tape Speed
   Read/Write (ips)			75/51/33
   Search/Rewind (ips)			90
 Backup Rate, Sustained
   Native (Mbytes/min, max)		30
   Compressed (Mbytes/min, max)		60


Thanks for your input.  I have done some trials today and with what I
could use for the buffer size it did a little better.  It looks like
the bottleneck is when tar gets to large files, aka CPU.

Thanks,
 
-- 
Brian