home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.sys.sun.admin:9419 comp.sys.sun.hardware:6231 comp.unix.solaris:416
- Newsgroups: comp.sys.sun.admin,comp.sys.sun.hardware,comp.unix.solaris
- Path: sparky!uunet!mcsun!sun4nl!fwi.uva.nl!casper
- From: casper@fwi.uva.nl (Casper H.S. Dik)
- Subject: Re: SPARCclassic and SunOS 4.1.x
- Message-ID: <1992Dec14.175428.26161@fwi.uva.nl>
- Sender: news@fwi.uva.nl
- Nntp-Posting-Host: adam.fwi.uva.nl
- Organization: FWI, University of Amsterdam
- References: <Bz4D02.DD7@news.fai.com> <1992Dec14.071012.10498@zip.eecs.umich.edu> <1992Dec14.084639.18266@fwi.uva.nl> <Bz9446.AB6@cs.columbia.edu>
- Date: Mon, 14 Dec 1992 17:54:28 GMT
- Lines: 66
-
- carson@cs.columbia.edu (Carson A. Gaspar) writes:
-
- >I, personally, have nothing against the SVR4 API. It's the
- >_implementation_ that is horrible. The code as shipped from USL is
- >horrible (unless SVR4.2 fixed things) and many vendors actually ship
- >it without re-writing so it works properly and performs decently.
-
- We have the SVR4.2 code and yes, it doesn't look nice. We never ran
- it on any hardware, so I can't judge the performance.
-
- >Now, as far as I know, Sun has fixed the known bugs in USL's code, and
- >has greatly improved performance. _HOWEVER_, the numbers still show
- >that 4.1.3 is _much_ faster than Solaris 2.1. Now, if you have a sun
- >multi-processor machine, Solaris 2.1 may be faster as it _finally_
- >implements an SMP kernal. Of course, solbourne has an SMP kernal
- >based on 4.1.3 that screams, but they don't sell their OS for
- >non-solbourne hardware (too bad).
-
- I don't know where you get the ``_much_ faster than''. The
- press releases tell us there is a difference in the areas were
- performance was affected most of about -10%. Some performance was actually
- improvement over 4.x. I've just started to play with Solaris 2.1
- (much faster than 2.0) on an SLC, and I don't feel that it is
- slower than 4.x. I'm also impressed by the ease of installation and
- the new features that make things easier to handle. I'm just not
- sure whether all these new features will scale well in an large
- installation.
-
- >The real question is - when is Sun going to stop slowing down the O/S
- >with each major release (I know several people who have upgraded to
- >SunOS 3.x for performance improvements - they find it unfortunate that
- >they can't run it on all their Sun hardware).
-
- My feeling is that they just did. If we disregard 2.0 (which was
- a test release, nothing more), we can say that 2.x isn't much slower
- than 4.x, especially considering that they are still tuning Solaris 2.x.
- What bothers me most is the always increasing memory use. You could
- run 3.x in 4MB, you needed 8MB to get 4.x going, now you need 12MB to
- get Solaris 2.x going. (The memory sizes are the absolute minimum,
- to be fair to Solaris 2.x, it should be noted that that memory requirement
- is because the installation is done swapless. Need feature, being able
- to remove all disk-based swapspace)
-
- > Sun (and, to be fair,
- >most O/S vendors) keep shipping systems that are more and more
- >bloated. There are several alternative kernal architechtures that
- >have been implemented, tested, and proven to outpreform monolithic
- >kernals (Henry Massalin's Synthesis spawns threads about an order of
- >magnitude faster than SunOS calls functions). It's well past time
- >that the industry took advantage of the new technology. Mach is a
- >nice first round, and proof that you can maintain the familiar API(or
- >APIs!) while greatly improving performance. While NeXTStep has not
- >yet moved to the Mach microkernal (due to stability considerations),
- >it will be interesting to pit NeXTStep 486 against other vendors' SVR4
- >(including Solaris) on the same hardware and see what the numbers have
- >to say.
-
- This would definitely interest me too. In a few months time we will be able
- to benchmark NeXTStep 486 vs Solaris 2.x/368 (or whatever they will call it)
- vs linux vs 368BSD vs BSDI vs the different SVR4s (Destiny/Dell etc) on
- exactly the same hardware. The numbers will tell us.
-
- (I've changed the newsgroups headers a bit, this doesn't seem to belong in
- a misc.forsale.* group)
-
- Casper
-