[Soekris] Harddisk Slow

[Igor Sobrado]

In message <20070131105543.GB1800 at mathom.us>, Michael Stone writes:
> On Wed, Jan 31, 2007 at 10:21:04AM +0100, Igor Sobrado wrote:
> >bs is the block size, not the buffer size.  It is, however, an useless
> >parameter if the filesystem is not tuned to use that block size on all
> >transactions by means of newfs.  
> 
> It's not useless, as the numbers provided by the person you're replying 
> to show. A meg is overkill, but but it obviously was better than 512b. 

Hi Mike.

A block size of 1MB means that the even small scripts (e.g., a 600 bytes
shell script) will require 128KB to be stored.  Indeed, larger block
sizes show that the IDE controller is not the bottleneck on these
appliances.  I am just curious: if these drives are working in advanced
DMA modes (in most cases, UltraDMA 2) why is the interrupt load so high?

I am using a PowerEdge 350 here.  It can easily provide 20MB/s with
an 800MHz PIII procesor.  The net4801 has a relatively fast Geode
processor (266 MHz is not too slow).  Even my 400MHz PII laptop
is faster than the net4801 in both writing and reading:

$ dd if=/dev/zero of=test count=65536 
65536+0 records in
65536+0 records out
33554432 bytes transferred in 2.833 secs (11844134 bytes/sec)

$ dd if=test of=/dev/null count=65536
65536+0 records in
65536+0 records out
33554432 bytes transferred in 1.242 secs (27016450 bytes/sec)

Perhaps an interrupt routing problem on the net4801?  If there is
an interrupt routing problem, there is a hope this problem being
fixed in a future firmware upgrade, concurrently to the integer
overflow problem described in a previous post.

Best wishes,
Igor.