home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!usc!cs.utexas.edu!wotan.compaq.com!moxie!texsun!cronkite.Central.Sun.COM!news2me.ebay.sun.com!exodus.Eng.Sun.COM!exodus!kessler
- From: kessler@sluggo.Eng.Sun.COM (Tom Kessler, Internet Spokesmodel)
- Newsgroups: comp.sys.sun.misc
- Subject: Re: Solaris 2.0 slower than 1.0
- Message-ID: <KESSLER.92Jul27190226@sluggo.Eng.Sun.COM>
- Date: 28 Jul 92 00:02:26 GMT
- References: <32280001@hpcss01.cup.hp.com> <32280003@hpcss01.cup.hp.com>
- <1992Jul15.005808.2470@ultra.com> <unruh.711171327@unixg.ubc.ca>
- <1992Jul15.194938.1305@cirrus.com>
- Organization: Rocket Scientists Anonymous
- Lines: 43
- NNTP-Posting-Host: sluggo
- In-reply-to: dhesi@cirrus.com's message of Wed, 15 Jul 1992 19:49:38 GMT
-
-
- >
- >In the System V world, TCP/IP networking goes through some extra layers
- >of processing.
-
- Yes, completely different architecture. More layers? It's hard to say.
- Does a Airplane have more layers than a grapefruit? (i.e. the question
- isn't particularly relevant).
-
- >
- >First, there is the emulation overhead: there is a native TLI layer, and
- >on top of that is emulation code to give you a BSD sockets interface.
- >This has overhead.
-
- Turns out it doesn't make much difference.
-
- >
- >Second, SVR4 does TCP/IP using "streams" modules, and that has overhead
- >too.
-
- And the socket structure doesn't/wouldn't if were multi-threaded as streams
- were? Like I said you're comparing airplanes & grapefruits.
-
- >
- >I'm sure Solaris is affected by these factors.
- >
- >The good news is that all this modularity should allow you to configure a
- >system without any TCP/IP or sockets, since both are now obsolete. :-)
- >--
-
- TCP/IP was the only thing we benchmarked that ran noticeably faster under
- Solaris 2.0 than it did under 4.1.2 (about 7% faster).
-
- NFS (which didn't change as much TCP/IP) is a bit slower under 2.0 than
- 4.1. Go figure.
-
- Armchair performance analysis is destined to be either wrong or lucky :-)
-
- Multi-threading, multi-processors, and working with SVR4 (the kitchen
- sink of unixes) all had a lot more to do with this than anything else.
-
- You really can't know how this OS will perform (or why it's slower) until
- you take specific benchmarks and measure why they perform differently.
-