List:General Discussion« Previous MessageNext Message »
From:Van Date:July 17 1999 6:51am
Subject:Optimizing a P133 for MySQL, Sending the Topic to Pasture......
View as plain text  
Okay, folks, I promised results on this, and, here they are:
A P133 w/ 48 Meg of SIMMS NP 40 RAM on a T-1 running a mixture of static
content and mysql tracking will be capable of about 5,000,00 hits a day
taking 15 million inserts/queries to the mysql dbase.  

I'm not going to beat a dead horse, but, this should be sufficient..... 
Scenario:  Can you take 5MM hits/day?
You get my point, all.  

Regardless, this has been the most exciting discussion I've seen in here
in many a day, and, that ain't trivializing this group.  I look to the
MySQL list the first thing I get home from work 'cause you folx really
know what you're doin'.  

I am most gratefull to you all. 

The reality is, this machine could potential take as many as 13,000 hits
per second, but, how am I gonna draw that kind of traffic?  If it takes
it's max of 100 hits/sec, I'm gleaning over 10 Million hits/day, and, we
all know what happens then.... (demand=a few bucks on a better server).

Any additional comments, always welcome, of course, but, I want to cut
the chatter down.  
Regards, and thanks all.
BTW:  I accomplished the benchmarks using a 486/75, P166, and a K6.2
400Mhz over 2 10Mbps cards.  The reality is this paradigm couldn't make
the P133 work hard enough to send more than 20-25 requests/sec.  The
solution (ironically) was to fire up the requests from within telnets on
the target box (the P133, we originally put to the test).  I set up 2
lynx -traverse -crawl scripts to run for over an hour, and, that
generated sufficient data to indicate a 100 hit/sec expectation was not
unreasonable.  Keep in mind, these client lynx processes consumed
tremendous CPU, and, probably debilitated the server's performance, but,
while running 5 such processes at 100 requests/second, I telnetted to
the outter world and lynx-ed back in and was greeted by the page w/int
less than 3 seconds.  This all logged an additional 2 records to the
dbase with full header information, so, nothing broke in the process.

I'm quite amazed at this beast.  Ironically, when running the
bench-marks, I ran into some drive problems (stuff like slow seek
times), and, just blew it off thinking that a different scenario would
not thrash the drive to that extent.  I haven't seen an error in over 36
hours.  (yes, the machine has yet to be rebooted since I started testing
after the 2.2.9 kernel build 2 nites ago).

I tried to break it, with all of your help and wasn't able to.  Anyone
else have another suggestion, I'm all ears.  

But, I'm thinking:  showtime.

Best regards, and, thanks all of you.  You're some incredibly bright

Linux rocks!!!
Optimizing a P133 for MySQL, Sending the Topic to Pasture......Van17 Jul