Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 21 Feb 1997 16:43:08 +1100
From:      Bruce Evans <bde@zeta.org.au>
To:        bde@zeta.org.au, dkelly@hiwaay.net
Cc:        hardware@FreeBSD.org
Subject:   Re: _big_ IDE disks?
Message-ID:  <199702210543.QAA29486@godzilla.zeta.org.au>

next in thread | raw e-mail | index | archive | help
>> Run `dd if=/dev/zero of=/dev/null bs=1m count=1000' and complain if
>> the throughtput is much lower than 120MB/sec.
>
>nexgen: {366} uname -a
>FreeBSD nexgen 2.2-GAMMA FreeBSD 2.2-GAMMA #0: Sat Feb  8 07:30:46 CST 1997 
>    dkelly@nexgen.ampr.org:/usr/src/sys/compile/NEXGEN  i386
>nexgen: {367} dd if=/dev/zero of=/dev/null bs=1m count=1000
>1000+0 records in
>1000+0 records out
>1048576000 bytes transferred in 30.300733 secs (34605631 bytes/sec)
>
>System is a "P90" NexGen Nx586, PCI. CPU ID's as a 386, and thats the way 
>my kernel is compiled. Who do I complain to? :-)

My comment only applies to P5's with a 66MHz memory bus.  The speed should
be much the same for CPU speeds of 100MHz and larger multiples of 33[.3]MHz.
The above dd command essentially copies the same 4K kernel buffer to
sequential memory 256 times. .  The kernel buffer stays in the L1 cache
so reading it is almost free and the speed approaches the maximum main
memory write speed which is 133M * 4/3 on my P5/133 (I think it is for
a 6-2-2-2 burst cycle).

Non-Intel x86's aren't detected very well.  I think most of them are super
486's, so they can't possibly sustain 170MB/sec writes.

>OTOH, it does "make world" in just over 7 hours with its 64 bit wide bus. 
>About the same as my super cheap AMD 5x86/133-P75 on a 32 bit wide bus.

The 64 bit bus might allow it to do 170MB/sec, but even a P5 writes at
half speed unless writes go through the FPU (this is for cache misses,
which is the usual case for large writes).  On P6's, writes normally go at
full speed because they are from the cache, but cache lines are normally
automatically preallocated, so there are twice as many memory accesses
and the write bandwidth is halved.  Preallocation is an optimization
for normal writes and a pessimization for large writes.

>Maybe I'm in the market for a new CPU/MB. What gives 120MB/sec performance? 

P5's with a 66MHz memory bus.

Bruce



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199702210543.QAA29486>