Performance Problems.. Server hardware smoked by $500 box?

[John Straiton]

> Thus spake John Straiton (jsmailing at clickcom.com) [11/09/03 12:45]:
> > Thanks for your suggestions, unfortunately- yes, both scenerios in 
> > production (webserver + db server or db server acting as both) are 
> > slower than the development box serving off of the db server. There 
> > are no local services other than apache on the development machine.
> 
> What about the server machine?
I'm sorry I didn't make this clear in the beginning..it's funny what
stuff you forget actually is important to tell people.

The production webserver and the development webserver (when this
exercise started) were *exactly* the same in apache configuration
(except IP's) and general setup (except hardware differences and the
production running 4.8-S while the development ran 5.0-R). 

The important hardware differences were shown in my original email. 

> - filesystem usage
Since they were configured the same, and both out of "Live" production
when we tested, these would be the same

> - disk size *and* speed (i.e. 5400 vs 7200 RPM, ATA100 vs. ATA33...)
The production used a UW SCSI Seagate Cheetah (10K RPM), while the
development used a Maxtor UDMA133 (7200 RPM)

> - memory speed (PC100 vs. PC133, DDR200 vs. DDR400...)
Production  512MB PC133 ECC Registered  vs Development  256MB 266Mhz DDR

> - network troubleshooting (10Mb vs. 100Mb, distance between machines,
100Mbit, Full Duplex, less than a meter difference between all 3
machines and the switch they all share (Cisco 3524XL) Together they are
below 10 meters total cabling.

>   distance between servers, different cards being used...)
The servers use Intel integrated cards, vs we have a $7 no-name realtek
card in the development machine.

> - Software tuning / configuration
At one point (when we noticed the problem), the only difference between
the machines was that the development server had less apache servers
(Min/Max/Start) configured because even with just 5 starting, it would
begin to swap (cause it ran out of physical ram) after booting.

> If you're doing heavy disk I/O, then watch IO on the system (iostat).

When out of production, there's no reason either machine would have any
disk IO other than normal system logs.

> Look at network errors (netstat -i), duplex mismatches (link 
> lights on switch vs. what ifconfig tells you), and network 
> load (i.e. production is on a 10Mb hub, development is on a 
> 100Mb switch).  Etc., etc., etc.

They are all connected to the same switch, and the Cisco IOS reports
zero errors on any of the 3 interfaces. Even when the production machine
was live, it would have less than 3Mb/sec load on a 100Mbit/FullDuplex
network.


I know this kinda sounds like I'm being contradictory, but believe
me...I'm not. This is why I'm so stumped, because I've looked at
everything mentioned here but as of yet, haven't been able to justify to
myself that just the difference in DDR vs SDRAM and 500Mhz is what I am
to blame for it being visibly faster while serving up web pages.. I'd
buy that if I were comparing a difference of 50fps in Quake3 but we're
talking about *apache* here.. Heh..

I'd hate to think I need to buy a server with 500 more MHz just to test
that theory. Perhaps there's something like slo-mo for FreeBSD that I
can use to knock that AMD down to around 1Ghz without having to do it
via dipswitches on the motherboard... Or well..maybe I'll just crack the
case and turn it down (hopefully it'll run if I do that)

Thanks for the suggestions,
John Straiton
jks@ clickcom.com
Clickcom, Inc
704-365-9970x101