apache-mpm-prefork debian perfomance tuning

Thankyou so much jose and Roberto for your Responses :-)

On 3/27/07, Roberto C. Sánchez wrote:
> On Tue, Mar 27, 2007 at 03:03:11PM +0530, Siju George wrote:
> > Hi,
> >
> > Just wondering for what amount of Load the default configuration of
> > apache2-mpm-prefork in Sarge is tuned?
> >
> > Will it server well about 1,000 users simultaneously?
> >
> It depends. There are so many variables that it is not even funny. You
> probably want to investigate thouroughly what it is that you are trying
> to do, figure out what packages you will be using and then look at the
> content that you will be serving. Only then will you even begin to have
> enough information to make something resembling a guess that is within
> an order of magnitude.
>

Yup

This one is going to be e-learning website.
Where people (anyone) can come register for free and upload their
content ( turtorials media at most 50 MB per user )

And others can come and download the and use them for free.

No Idea how many users will be there exactly :-(

What i am concerned is about the

http://en.wikipedia.org/wiki/Slashdot_effect

users are expected from

http://digg.com/
http://www.techcrunch.com/
http://www.solutionwatch.com/

i also went through

http://tech.cybernetnews.com/2006/07/14/cybernotes-the-analysis-of-a-digg-effect-4x-in-10-days/

http://blogs.tech-recipes.com/davak/2005/11/06/digg-effect-the-top-10-things-webmasters-should-know/

So i would like to prepare for that.

The server i have is

======================================================================
1x Dual-Core Opteron 2210 (1.8GHz) on a Dual CPU Motherboard

2x 2GB RAM/PC3200 (4GB Total)

2x Onboard GbE LANs

CD-ROM

1U 2 Hot-Swap Bays Chassis & Mounting Brackets

x Dual Channel SATA RAID Adapter

2x 500GB 7200 RPM SATA RE Hot-Swap HDD's
=========================================================================

So I hope I should be Installing ( right? )

kernel-image-2.6.8-12-amd64-k8-smp - Linux kernel image for version
2.6.8 on AMD64 SMP systems

to harness the full power of the CPU as currently it shows only

==================================================================
Linux websrv-1 2.6.8-11-amd64-generic #1 Thu Jan 11 21:39:50 PST 2007
x86_64 GNU/Linux
# cat /proc/cpuinfo
processor : 0
vendor_id : AuthenticAMD
cpu family : 15
model : 65
model name : Dual-Core AMD Opteron(tm) Processor 2210
stepping : 2
cpu MHz : 1795.549
cache size : 1024 KB
fpu : yes
fpu_exception : yes
cpuid level : 1
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge
mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext lm
3dnowext 3dnow pni cmpxchg16b
bogomips : 3547.13
TLB size : 1088 4K pages
clflush size : 64
cache_alignment : 64
address sizes : 40 bits physical, 48 bits virtual
power management: ts fid vid ttp [4] [5]

# mpstat -P ALL
Linux 2.6.8-11-amd64-generic (tutorom-dbsrv-1) 03/28/2007

03:37:08 AM CPU %user %nice %system %iowait %irq %soft
%idle intr/s
03:37:08 AM all 0.01 0.00 0.02 0.06 0.00 0.00
99.91 1003.62
03:37:08 AM 0 0.01 0.00 0.02 0.06 0.00 0.00
99.91 1003.62
==========================================================================

also it has

==========================================================================

3ware Storage Controller device driver for Linux v1.26.00.039.
scsi0 : Found a 3ware Storage Controller at 0xcc00, IRQ: 16, P-chip: 1.3
scsi0 : 3ware Storage Controller
Vendor: 3ware Model: Logical Disk 0 Rev: 1.2
==========================================================================

Just wondering mdadm can monitor this hardware Raid Controller? is it supported?

I hope to Install LVM and have seperate partitions for
"/var/log" and "/var/www" so I can adjust them if there is any need in future.

What do you suggest use up the whole 500 GB disks or leave some free
Space for later use?

Also

Should I use.

"apache2-mpm-worker" or "apache2-mpm-prefork"?

because

http://www.debianhelp.co.uk/apacheinstall.htm

says "apache-mpm-worker" could be a risk.

Also this will be on PHP5 and will be contacting a MySQL 5.0 Database
Server with the same Specs.

Hope these informations are enough :-)

Please let me know if you need more.

Thankyou so much once again :-)

Kind Regards

Siju

On Tuesday 27 March 2007 08:44, Roberto C. Sánchez wrote:
> On Tue, Mar 27, 2007 at 03:03:11PM +0530, Siju George wrote:
> > Hi,
> >
> > Just wondering for what amount of Load the default configuration of
> > apache2-mpm-prefork in Sarge is tuned?
> >
> > Will it server well about 1,000 users simultaneously?
>
> It depends. There are so many variables that it is not even funny.

so true.

> You
> probably want to investigate thouroughly what it is that you are trying
> to do, figure out what packages you will be using and then look at the
> content that you will be serving. Only then will you even begin to have
> enough information to make something resembling a guess that is within
> an order of magnitude.

even then, to answer a question like "how many simultanious users can the site
handle" is near impossible. you'd first need to define "simultanious". you
may begin to attempt to answer a slightly different question of "how many
visitors per hour" or some such. even then, the best guesstimate would be the
order of magnitude. simply because there are so many variables involved.

i do stress testing every year or so for a cluster of web servers. "httperf",
snmp, cactus, rrdb et al. are your friends. enable "response time" logging in
apache and keep an eye out for pages taking longer than 0 seconds. (afaik,
apache rounds down to the nearest second). enable slow query logging in
mysql. then use explain to find out what's causing each 'slow-query' to take
as long as it does. fill the database with data before doing your tests. the
database behaves differently with a lot of data vs. without (eg. mysql uses
table scans even when there are indexes when it figures it has to look at 75%
or the data in the table or some such). and if you do end up using a stress
generator (for the lack of a better term) like 'httperf', make sure you are
not measuring the performance of the machine generating the requests.

even the smallest delay you would barely notice when not under stress tends to
magnify and contribute to heavy load on the system when it gets busy. so the
only way to find out is to really put the system under heavy load using
httperf or the like..

hope that helps.

>
> To put it in perspective, you have something equivalent to "can a
> roadway handle 100 vehicles?" Without providing any other information
> to help, it is impossible to answer.
>
> Regards,
>
> -Roberto

--

anoop aryal