From owner-freebsd-current  Tue Jun 13 04:54:07 1995
Return-Path: current-owner
Received: (from majordom@localhost)
          by freefall.cdrom.com (8.6.10/8.6.6) id EAA25532
          for current-outgoing; Tue, 13 Jun 1995 04:54:07 -0700
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.34])
          by freefall.cdrom.com (8.6.10/8.6.6) with ESMTP id EAA25479
          for <current@FreeBSD.ORG>; Tue, 13 Jun 1995 04:53:57 -0700
Received: (from bde@localhost) by godzilla.zeta.org.au (8.6.9/8.6.9) id VAA23001; Tue, 13 Jun 1995 21:49:06 +1000
Date: Tue, 13 Jun 1995 21:49:06 +1000
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199506131149.VAA23001@godzilla.zeta.org.au>
To: amurai@spec.co.jp, tom@uniserve.com
Subject: Re: GENERIC kernel & some basic UNIX pointers
Cc: current@FreeBSD.ORG
Sender: current-owner@FreeBSD.ORG
Precedence: bulk

>> But Predictor-1 will try compression whole date when talk to peer ppp.
>> This will give us two advantage as follows:
>> 
>>      1. Decrese system over head due to reducing a sio H/W interruption.
>>         (H/W interruption is required big cpu time especially latest cpu)

>  What about the extra CPU time and extra latency required for the actual 
>compression?

The CPU time is smaller than the saved (interrupt + i/o + sio protocol)
time if the cpu is sufficiently fast.  (A faster cpu reduces the compression
and sio protocol times but has less effect on the interrupt time and almost
no effect on the i/o time).  Serial i/o is fairly efficient in FreeBSD, so
"sufficiently fast" is about as fast as a P90 if compression is the same
speed as compress(1) and the compression ratio is 50%.

The latency should rarely be larger - if compression is worth doing, then
it must be much faster than serial i/o.

Of course, you lose both CPU time and latency for incompressible data.

Bruce