[Asterisk-Users] Tuning the Linux kernel?

[Andrew Kohlsmith]

> prety much self-adaptive.  The why to gain performance (if
> you need it) is to throw hardware at the problem.  Using
> fast SCSI disks, loads of RAM and faster CPU(s) are the way to go
> but, big but, unless you've hit an identifiable bottle neck
> there is no point.

Amen.  A lot of people spend a lot of time optimizing what I call "corner 
cases" -- optimizing things that run so infrequently or have such little 
impact on the overall system that it's not worth optimizing at all.

Memory usage is one: I mean I don't think I can buy anything under 256MB 
DIMMs anymore; I don't install swap space either (2.4.x is absolutely 
braindead in swap/RAM usage) -- I'd be surprised if 90% of the standalone * 
systems out there will come close to using that single 256MB DIMM I 
install.

Same thing with compiling the kernel with optimizations.  I think that for 
the most part, *'s processor usage lies in any codec translations that may 
be occurring.  I have MMX enabled in my zapata drivers and turn the 
PROC=`uname -m` into PROC=athlon or pentium4 or whatever it is I have in 
the system.  I've seen measurable improvements in the codec translation 
times.

(Aside: I told the 2.4.22 Linux kernel that I was on an athlon chip and had 
all manner of instability, random segfaults and crashes...  Moving back to 
i486 solved the problem.  Compiler bug or kernel bug I'm not sure.)

(Aside 2: Does anyone know why the md driver does not select the fastest 
RAID5 checksumming method?  I don't use md RAID5 but it always seems to 
pick the second-fastest of all the methods it tests, as seen in dmesg.)

Regards,
Andrew