OT: Unix can't win

[Bill Vermillion]

On or about Wed, Jan 05 12:31 , while attempting a Zarathustra 
emulation Bill Campbell thus spake: 

> On Wed, Jan 05, 2005, Bill Vermillion wrote:
> >As Bill Campbell was scratching "For a good prime call 
> >391581 * 2^216193 -1" on the wall, he suddenly said:

> >> On Tue, Jan 04, 2005, Ted B Dodd wrote:
> >> >http://www.eweek.com/article2/0,1759,1746236,00.asp

> >> That's news? Why do you think all the proprietary Unix vendors
> >> other than SCO have been moving into the Linux space (or at
> >> least trying to :-)?

> >And of course that information came from a Microsoft spokesman too.

> >> I did find the comment about the company switching from AS400s
> >> to Windows 2003 server interesting, particularly since I've
> >> been told by local people who Should Know(tm) that Microsoft's
> >> main accounting applications still run on AS400s, and that
> >> their attempts to run them under Windows have failed miserably.
> >
> >> ...

> >The article also mentioned that Enterprise systems were moving from
> >proprietary Unix to MS and/or Linux.   But what you would call
> >an enterprise system 10 years ago is not the same as you see to day
> >- when more companies are computerized.

> >The proprietary systems are getting larger and more powerful - but
> >the commodity/pc/Intel architecture systems are far more powerful
> >today so some of the old systems may not be needed.

> Don't underestimate the Apple X-Servers and their X-Raid
> boxen as well. The X-Raid systems are more than competitive
> with NetApp and other network storage, and the X-Servers have
> excellent Apache, java, and support for all the standard *nix
> server software. I'm not ready to stop using Linux and FreeBSD
> servers in favor of Apple yet, but they are making inroads on
> the Microsoft web server space.

And as of this week Apple will be selling to regular users the
same X-server configuration with dual Power G5 chips running at
2.4Ghz that is powering the supercomputer cluster that put that
cluster at number 7 in the top 500 list of supercomputers.

That's a pretty mean achievement for the first time on the list.
It's installed at Virginia Tech.

The new machines have dual 1Gbit Ethernet and an 800Mhz Firewire
connection.  Can have up to 8GB ram on board with 1.2 Terabytes
of internal storage in the 1RU rack.  Shipping with 1.5 GB RAM
and an 80GB drive it comes in a $3999.  That's a lot of power
for a 1RU machine.

It also has PCI-X in it.  I wa a bit surprised that with all the
other advances that they went that way and not with PCI-Express.
But since the latter will probably eventually replace the Firewire
and Apple uses that heavily that may have had something to do with
it.  Thankfully all these new serial technologies, SATA and PCI-Express,
are going to mean an end to the cable clutter internal with
all those wide parallel wires. And those 80 connector ATA 66 and up
are awfully stiff and hard to route when you have good quality
cables.

> >Now - here is a question.  On the meltdown of the airline computer
> >system during the Christmas holiday does anyone know the operating
> >system and/or hardware environment.

> >The old systems such as Sabre just never failed.   My hunch [not
> >proven] is that someone was trying to do something cheaper, and
> >those AS-400s as you mention were awfully rugged, and the Z-series
> >from IBM running the IBM OS or multiple instance of Linux just
> >don't fall over.

> As a Burroughs MCP user in my main frame days, I looked on the
> IBM systems as horribly inefficient and expensive (much as I
> view Microsoft today). One has to remember that IBM never wrote
> a line of code that wasn't designed to sell more hardware.

But the did run and run.  And the Burroughs - now Unisys - were and
are real workhorses.   The PC style architecture isn't as robust or
fast.  That's why the proprietary machines can still have a market
- at least as I see it.  

> I have heard that the economics of running a bunch of apache
> servers on virtual Linux systems on the IBM main frames is
> significantly less than doing the same thing on Sun hardware
> running Solaris. While clusters of Linux boxes running under
> Beowulf are great for parallel computational jobs, they
> aren't generally applicable for Enterprise level database and
> transaction processing.

You do have to take into consideration the footprint of
a Z-server compared to the clusters of other machines.  Not have to
connect everything helps.  The IBM implementation where the virtual
servers can communicate directly instead of over a typical TCP/IP
connection is also a big plus.

The Z-server looks like it could be a good machine for an ISP
installation - as the space consumed for the same number of virtual
machines is far less.

But once you get to this level what you need to do is going to
dictate what machine you use.

Can you imaging one of the Z-serves with 2500 virtual Linux servers
each running filePro.   That would be one serious revenue stream
for fpTech.

Bill

-- 
Bill Vermillion - bv @ wjv . com