Re: for linux, which CPUs are almost always faster than which other CPUs?



At Sat, 14 Feb 2009 10:01:39 -0800 (PST) Bennett Haselton <bennett@xxxxxxxxxxxxx> wrote:


Is there any way to find out how to compare two different CPUs for the
purposes of running a Web server (other than doing benchmarks which
are not practical in this situation), just based on the model of the
CPU?

For a *web server*, the main bottleneck is you disk drives / mass storage
system. The *second* bottleneck will be memory. Generally, for web
servers, processor speed is not really a very important issue. Assuming
your disks are fast enough, you have enough memory, a faster
and/ormulti-processor/core will be a help. *Most* of what a *web
server* is doing is moving bits from the disk to the network.


Everywhere I've asked this before, people have said things like "There
are no absolute ways to compare CPUs because there are too many
factors. The only reliable way to do a comparison is to run your app
on both, and make careful observations. It takes time and work and
there are no shortcuts."

This is generally going to be true. Really.


And then I'll ask someone about comparing CPU A and CPU B, and they
say, "Well of COURSE cpu A is almost always faster, EVERYBODY knows
THAT!"

These answers are not likely to be universally meaningful. Anyone
making a statement like that probably does not really know if
his/her answer is neccessarily apllicable to *your* situation
anyway.


OK, so *that* was the kind of answer I was looking for the first time
around -- the comparisons between CPUs where "everybody knows" that
one is faster. Like, everybody knows that a Xeon is faster than a
Pentium (with similar clock rates), but that seems like the kind of
information people pick up ad hoc over the years. To leap-frog the
process of spending years accumulating that sort of information in
bits and pieces, what is the closest thing to a sorted list of CPUs,
where 90% of the time, the higher CPU in the list is the faster one?

It depends... Xeons have more cache than Pentiums -- this makes them
'faster' if/when having a larger memory cache reduces real memory hits.
This is application dependent. Celerons have *less* cache than Pentiums
-- this makes them 'slower' if/when having a smaller memory cache
increases real memory hits. If you have an application that has a
smaller need for memory cache, a Celeron will be just as fast as a
Pentium or a Xeon. If your application has a 'middle' need for memory
cache a Pentium will be faster than a Celeron and just as fast as a
Xeon. If your application has a 'large' need for memory cache a Xeon
will be faster than a Pentium, which in turn will be faster than a
Celeron. In the case of a web server, having more cache helps some, but
if you have lots of web pages, you are going to be moving large slabs of
memory and/or disk blocks around. Even Xeons have a limited sized
memory cache. The difference between memory access speed and cache
memory speed is small compared to disk access speeds (and network speeds
as well). If you site is not a huge traffic site (or has huge bandwidth
needs, eg a video or audio site), you just need a processing element
that can move blocks as fast as they can be read off disk. A Celeron
should be fast enough for that.


Could it be as simple as looking at the price list for CPUs all from
the same seller, and assuming that the more expensive ones are faster
for most purposes?

No. Consider: a Ferrari is great if you want to go really, really fast.
It is also a very expensive car. But is it any good for hauling 500
gallons of water up a steep mountain road? A ratty old Ford pickup will
do that better and cost 1/10 the price...


The only rules of thumb that I have to use are (a) if everything else
in the processor name is the same, then the one with the higher MHz is
faster, and (b) if two dedicated servers offered by the same hosting
company are the same in every respect except for the processor, then
the more expensive one is probably faster. (But, that doesn't work
for comparing servers across different hosting companies since their
markup amount varies, so $59 one at one company may actually be
faster than a $69 one somewhere else.)

No.


For example, from 3 different hosting companies right now I could get
a server with a Celeron 420, a 2.4 GHz Pentium or a "Celeron 2.6" for
about the same price. Are there any two of these where I can say,
almost always, that one is faster than the other? For example,
according to
http://en.wikipedia.org/wiki/Celeron
the Celeron 420 has a 1.6 GHz clock rate. I assume that what the
other hosting company is calling a "Celeron 2.6" is a Celeron that
runs at 2.6 GHz, even though the Wikipedia article doesn't list any
Celeron model that runs at 2.6 GHz. So is the "Celeron 2.6" almost
certainly the better deal than the Celeron 420? And how would I
compare either of those to the Pentium?

Get one of each and try it out.


Bennett


--
Robert Heller -- 978-544-6933
Deepwoods Software -- Download the Model Railroad System
http://www.deepsoft.com/ -- Binaries for Linux and MS-Windows
heller@xxxxxxxxxxxx -- http://www.deepsoft.com/ModelRailroadSystem/

.



Relevant Pages

  • Re: 64-bit memory consumption
    ... CPU cores and memory keep pace with each other, ... There is time spent in GC that won't be affected only to the GC cycles, you get indirect hits in the form of poor cache coherency and cache pollution in the code that uses GC'ed memory. ... Desktop CPUs with significantly larger caches aren't there yet, and on a server, your cache is shared by multiple applications, and you can't appropriate it all for your application's purpose without affecting the rest. ...
    (borland.public.delphi.non-technical)
  • Re: Slowly Terminal Server response
    ... physical memory is an indicator of available memory. ... If Task Manager only shows a single CPU graph, then only one CPU is installed. ... A good general guide to keep in mind is that most applications require the similar amount of memory as they would on a desktop for each user. ... I normally start with a general guideline of 25-30 concurrent users for a server with 4 GB of RAM and 2 CPUs and then do testing for a specific set of required applications. ...
    (microsoft.public.windows.terminal_services)
  • Re: CTRL-ALT-DEL SCREEN VANISHES
    ... for memory and CPU can give you an idea of what is going on. ... The server runs the following apps/services ... force the problem to occur making it more difficult to troubleshoot. ...
    (microsoft.public.windows.server.security)
  • Re: Slowly Terminal Server response
    ... You can use threadmater to cap the cpu usage on your access app: ... How much memory is left when the 20 sessions are running? ... MVP - Terminal Server ...
    (microsoft.public.windows.terminal_services)
  • Re: Benchmarking multiplications and additions
    ... function in my programs in order to find out how long it took the CPU ... compare to the amount of time you've already spent reading ... performance measuring problem. ... faster than memory can supply operands or consume results. ...
    (comp.lang.c)