Re: Windows vs Linux Performance

DA Morgan wrote:
> mccmx_at_hotmail.com wrote:
> >> The biggest issue I have seen with RAC on Windows they didn't even test.
> >> The point of RAC is not how much water can I push through the pipe. The
> >> point of RAC is FAILOVER.
> >>
> >> My experience with Windows and failover is that what happens with Linux,
> >> subsecond, can take up to 25 seconds with Windows which is not acceptable.
> >
> > Fair point but the whole point of the document was to compare
> > performance of Windows vs Linux, not failover.
>
> But that's not RAC and thus it is meaningless.
>
> > I would lke to have seen the comparison done without the added
> > complexity of RAC involved.
> >
> > Matt
>
> I would too. But again the test is meaningless except as marketing
> fodder for Microsoft. Which Linux packages were installed? I don't
> know. Which kernel parameters were set? I don't know. What were the
> Oracle initialization parameters? I don't know. Integers are not data.
> --
> Daniel A. Morgan
> University of Washington
> damorgan_at_x.washington.edu
> (replace x with u to respond)
> Puget Sound Oracle Users Group
> www.psoug.org

Performance is not linear. Permutations compound dissimilarities. What works in one environment may not work in another. Windows may perform better in one instance and not in another. Each environment is as unique as there are opinions on this board. The choice should boil down to which one adds the most benefit to the company.

Think of this, I could pay a Nascar team to keep a fancy Nascar running and track ready. If all I do is drive the car to work, I've wasted lots of money and have used none of this performance. Is the Nascar better than my car? For speed, yes, for practicality, no. Is Windows faster, better, easier to use than Linux. Sometimes yes and sometimes no. But this is the wrong question.

 Can Linux out perform Windows. Is Linux a better more efficient system? Probably yes. But what's the point if it costs more and the use is doesn't justify the means? The basis should should be cost and cost is relative to each situation.

Notice the "Report Query Test", figure 4 (p13-14). 32bit outperforms 64bit. Unexpected? It should not be. Performance is not necessarily linear. The paper ends with the assumption that the generic Windows driver is at fault and had a vendor provided 64 bit storage driver been available, it would have bested 32 bit. This may be the case, but they certainly did not prove that out. At a minimum, the xplans should be included if this assumption is to be made. If the plans are exactly the same, then it I'd buy the driver guess, but without that info, who knows? Was cpu costing on?

So factoring in drivers, users, data, statistics, memory settings, init settings, session settings, storage setup, network speed, admin skill-set and moon phase, we come up with an enormous variety of possible situations that all work against any kind of standardization that can be generalized across the board. There is danger behind the reasoning that one system is better than the other. It is based on the same reasoning that says a Nascar is faster than a Mini. It is certain that it is, but it makes for a very expensive commute and does not factor cost into the equation. Received on Tue Jan 23 2007 - 08:48:54 CST