Skip to contentSkip to conversation

Compiling options?

DISCUSSION Robert L. Harris 4 participants 4 messages Mar 25, 2003

lore
Share & tools
lore
#0 lore raw

Ok, I currently manage a site with a couple hundred machines ranging from P2 through AMD-Durons. The prevaling theory was to make a single kernel compiled for a "pentium classic" and then load in drivers for about everything under the sun.

Is there anything good written up on if this is the best way for our 4 person admin team to keep managing this or should the boxes be custom tuned for specific groups of machines within reason? "What does it buy us" is one of the big questions as swapping out that many kernels and testing 5-8 different varriants is a big buyin on time.

Robert

:wq!
---------------------------------------------------------------------------
Robert L. Harris | PGP Key ID: E344DA3B

                                         @ x-hkp://pgp.mit.edu 

DISCLAIMER:
These are MY OPINIONS ALONE. I speak for no-one else.

Diagnosis: witzelsucht

IPv6 = [email protected] http://ipv6.rdlg.net
IPv4 = [email protected] http://www.rdlg.net

#1 lore raw
Quoted 5 lines · Robert L. Harris

Is there anything good written up on if this is the best way for our 4 person admin team to keep managing this or should the boxes be custom tuned for specific groups of machines within reason? "What does it buy us" is one of the big questions as swapping out that many kernels and testing 5-8 different varriants is a big buyin on time.

Depends largely on the workload. Compiling a kernel targetted at Athlon/Duron will factor in the 3dnow memory copying code which speeds up bulk copies quite a lot. If your workload doesn't involve that much memory copying, you'll likely not notice that much of a difference though.

Dave

#2 lore raw
In reply to Robert L. Harris
Quoted 4 lines · Robert L. Harris

Ok, I currently manage a site with a couple hundred machines ranging from P2 through AMD-Durons. The prevaling theory was to make a single kernel compiled for a "pentium classic" and then load in drivers for about everything under the sun.

If your boxes range from PII through to AMD duron build for 686, but the basic theory is the same.

A 386 kernel really hurts later CPUs
A 486 kernel is generally fine
A 686 kernel speeds stuff up a little more

The only CPU that is really helped by a custom kernel is the rather rare IDT winchip which is 10-30% faster with the right kernels

#3 lore raw
Quoted 6 lines · Alan Cox

If your boxes range from PII through to AMD duron build for 686, but the basic theory is the same.

A 386 kernel really hurts later CPUs
A 486 kernel is generally fine
A 686 kernel speeds stuff up a little more

Should add that a 586 kernel is horrid on 686 and Athlon machines - the scheduling is worlds apart. Use it only on true 586 machines.

If you need a common denominator, go with 486 as Alan pointed out. In your case 686 seems safe and sane, though.

Robert Love