Re: Would a 2nd processor really be a waste of time???? help

From: Bill Vermillion (bv_at_wjv.comREMOVE)
Date: 07/17/03 

Date: Thu, 17 Jul 2003 02:24:41 GMT

In article <Ic5Ra.81020$JA5.1643994@news.xtra.co.nz>,
Lee and Nadine <lee.nadine@xtra.co.nz> wrote:

>We are looking at different aspects on what to do about our
>older unix sco 5.5 server that tends to run slow at times
>(333mhz with 196 meg ram). using cpqmon it shows the CPU "2 sec
>avg" the "10 sec avg" and the "20 sec avg" as maxxed out when
>things run slow.

Put 'maxed out' into real numbers.

>One idea was to buy another processor and more ram, however this
>is what one supplier has told us - see below. I have no idea if
>what he is saying is correct of if hes trying to drum up more
>sales. Some ideas on what he wrote would be greatly appreciated.

>As quickly discussed on the phone today the server is dual processor
>capable.

>Your server has only 1 processor.

>Memory is not an issue unless the system is swapping , if this
>was the case on your system you would not be able to login.

That's not true. You can still log in on a system that is swapping
- but the login may be slow. Only when you get to the point of
massive thrashing do you lose any ability to log in.

>Your system is not swapping and so is not the cause of the slow system.

Did he check the system when it was running the slowest?

>The server is only 333 MHz, what would a 5 year old 333 MHz PC
>with 75 users logged in run like?

Well I've had systems like that with 120 users logged in - but it
was only 300MHz - and it ran pretty well. It had a fast disk
system.

One thing I don't see mentioned is disk performance.

Output of sar with about 10 results from about 15 second intervals
when the system is slow would surely help. Don't run intervals
less than 10 seconds or sar itself will likely impact the results.

>The Unix system takes full advantage of Symmetric Multi
>Processing when the SMP license is installed.

But your programs nay not.

>The only advantage of this would be seen when the system as
>it stands gets processor bound and slows down , with dual
>processors the load is shared and this processor bound state
>takes a little longer to occur .

>A second processor does not speed up a server but shares load at
>current processor speed. i.e. 333Mhz

>It should be noted that even with a second processor the system
>may become processor bound very quickly.

Was anything done to determine if the machine was processor bound.
That was not stated.

>Estimated Pricing:

>SCO Unix SMP License
>$2,900

>Second 333 MHz Processor [ subject to availability]
>$1,000

>Installation approximately 2 days labour [ out of hours rate ]
>$ 2,400

>Plus travel and accommodation

It's been quite typical that upgrades to older systems don't bring
the perfomance of a newer system.

>As discussed this is not a very good idea due to the small
>possible advantage it may bring during peek usage and cost
>benefit. Even after installation you may find that there has
>been no improvement in performance .

>Again as mentioned on the phone we need to get the new server
>installed as soon as possible so users can benefit from superior
>technology enhancements of the past 5 years.

You need to do some more testing to find out exactly where the
problem lies. You didn't mention the program you used - some
database designs should be taken out and destroyed.

Your question is sort of like asking 'how big a truck do I need to
move from xx to yy?' - without saying what you have to move. 

-- 
Bill Vermillion - bv @ wjv . com