[ale] sorta [OT] IO vs clock cycles

[Jeff Hubbs]

Whoa, that's a good question that I'll take a handwaving stab at.

There are so many intrinsic differences between the systems in question
and there has been so much acceleration in the x86-and-compatible world
just in the last three years that I'd be hard pressed to say without
actually trying to run some crude tests.  

However, what I remember from my NT heyday is that NT really, really
responded well to L1 and L2 cache.  My intuition says that a doubling of
cache could make one perceive a HUGE performance benefit under NT even
with no other changes.  

I have a machine running WinME downstairs (currently it's just padding
my SETI at Home stats, but its main purpose is high-end audio I/O).  It's a
Socket 7 Azza mobo, and it has an AMD K6-2+/500 in it.  By all
indications, it should be pretty fast, but I'm here to tell you that it
takes 91 hours for it to do a work unit - almost 1/8th as fast as the
K7/750 I'm typing on now.

It seems to me as though the mere differences in clock speed and K6-2
vs.K7 architecture don't account for all of that.  Cache differences
(this K7 has 512KB on-board) and I/O are going to make up for the rest. 
It has seemed to me that that K6-2+/500 is very much tied down by the
mobo.  But, just in the last year or two, mem I/O has increased in big
leaps, especially in the dual Athlon boards (I don't know if Intel
boards has had a similar breakthrough).  Gaming appears to be driving a
lot of that growth - you can't say that for the RS/6000!

So, having said that, there's the NT-vs.-Linux question.   In NT/2K-land
- your ability to influence the JVM's performance is fairly limited to
swapping out hardware and, if the app in question is disk-intense, you
can work for gains in that realm.  However, in Linux-land, you have all
those options and more, relating to the kernel, the libraries, and maybe
even the JVM software itself.  Think how often the kernel developers
make some big breakthrough, like the new VM system or the preemptive
kernel patch.  And, let's not even get into the issue of having a
supposed server OS that must run a GUI at the same time.

- Jeff  



On Thu, 2002-11-07 at 14:36, Cade Thacker wrote:
> Need a little bit of knowledge from you guys/gals. I got in a discussion
> about running Java in an enterprise system, and one of the gentlemen I
> repect a great deal said that Java would be much happier running on an NT
> box with a very high clock cycle, then a Unix work horse from IBM. His
> reasoning was that the Unix box is built(and build well) for IO intensive
> tasks, DB type stuff, where NT is build to take advantage of clock speed.
> Therefore the JVM would be happier on NT.
> 
> On the same topic, we always hear that the mainframe downstairs has less
> horse power then our PCs, but yet the mainframe handles Millions of
> transactions a second, and my PC dogs out running 5 browsers at once.
> Please explain.
> 
> Where do you guys think linux would fall in this debate?
> 
> Thanks guys, and as always looking forward to your answers.
> 
> 
> --cade
> 
> On Linux vs Windows
> ==================
> Remember, amateurs built the Ark, Professionals built the Titanic!
> ==================
> 
> 
> 
> 
> ---
> This message has been sent through the ALE general discussion list.
> See http://www.ale.org/mailing-lists.shtml for more info. Problems should be 
> sent to listmaster at ale dot org.
> 



---
This message has been sent through the ALE general discussion list.
See http://www.ale.org/mailing-lists.shtml for more info. Problems should be 
sent to listmaster at ale dot org.