Performance enhancements

[Net Llama!]

On 12/30/02 15:40, Bruce Marshall wrote:
> 
>> On 12/30/02 15:33, Marvin P. Dickens wrote:
>> > Also, just so you know Lonnie: This is rediculous.
>> 
>> Agreed, telling people to turn off swap to improve performance, is 
>> ridiculous.
>> 
> 
> Just to stick my $.02 in here and stir up the pot.....  let's talk
> mainframes for a minute.
> 
> On IBM mainframes, if you can turn off the Dynamic Address Translation
> (required for virt memory and swapping) then you can achieve some real
> gains in machine performance.  All jobs run in 'real' memory and the
> hardware doesn't need to screw around with address translations.
> 
> Can this same thing be done on the i386 architecture?  I doubt it and I
> would doubt that Linux would be able to pull it off.  But Linux never
> ceases to amaze me.
> 
> In any event, there *is* the possibility of performance gain if address
> translation can be turned off.

I wouldn't doubt it if there was an adequate amount of physical memory. 
  But somewhere along the line, Peck side-stepped the entire detabe from 
his original claim of "improving performance by removing swap" to "if 
your box dies, its cause you don't have enough physical memory". 
There's a world of difference between memory considerations on a desktop 
box, and a server, and Peck flip-flopped between the two whenever it 
served to support his argument.  I'd say unlesss you have upwards of 1GB 
of memory, its outright foolish to disable/kill swap on a desktop 
system, since they're way too multi-purpose to be able to accurate spec 
out memory requirements in advance.  On a server, where the purpose is 
usually planned well in advance, you should be able to adequately gage 
memory requirements and avoid the creation of a swap partition.

-- 
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
L. Friedman                       	       netllama at linux-sxs.org
Linux Step-by-step & TyGeMo: 		    http://netllama.ipfox.com

   4:40pm  up 15 days, 23:49,  2 users,  load average: 0.22, 0.23, 0.35