Hey Jim
One maybe-off-the-wall suggestion would be to have a quick global variable
somewhere that you can use to switch off code you suspect may bog you down. I
don't know what *you* plan on doing that night, but if watching the web site is
one of them, it might be nice to be able to turn off the session tracking stuff
to keep the site running, in a pinch. (Lots of the session tracking stuff you
mention seems like things you should be able to do after the fact using log
file analysis. That is always preferred to live tracking, in my opinion.)
Turning off robots that day (morning?) might be a good idea too, but yeah, it
might suck to get kicked out of google right when you might really want to be
there. (I really doubt they could react that quickly.) I do know that many
robots read robots.txt in advance and use the copy they have for some (short)
time, so changing any later could end up doing nothing.
Mark
-----Original Message-----
From: Jim Davis [mailto:[EMAIL PROTECTED]
Sent: Monday, December 27, 2004 2:18 PM
To: CF-Talk
Subject: Slow (sorta) site issues - Ideas?
I've a site that while not slow, is most likely not going to take the traffic
it's facing. The site is www.firstnight.org - it will get absolutely pounded
on New Year's Eve.
Since this will be the warnest New Year's Eve in recent memory AND is a weekend
AND is a holiday for nearly everybody now I fully expect this to be one of the
busiest years ever.
Unfortunately we're still a non-profit. We can't afford the kind of iron that
this kind of traffic would require for only one or two days a year. Right now
we're on a shared hosting plan at CrystalTech which we nearly overran last year.
I'm just going through and looking for savings here.
1) An average page runs anywhere from 80-300 ticks. I plan to address some of
that by caching the navigation HTML (right now it's dynamically generated from
a cached CFC).
What do you think? Too high? Way too high? Way, way too high?
Many of the pages are quite large and complex (for example the "All Events"
list here:
http://www.firstnight.org/Content/NewYears/Artists/Explore/Events.cfm?Type=All
) but it's exactly those pages that are the most popular.
As an aside, you can see the current number of active sessions and the current
page's tick count at the bottom of any page.
2) My session manager is worrying me. The site doesn't use CFs built in
Session management. This allows me to capture user information at the end of a
session, but means that I have to manually check and destroy sessions. When a
session ends it's saved in a database along with the pages viewed during the
visit, information about the user agent and several other things.
This process requires several database calls (perhaps a minimum of 8, but a
maximum determined by the number of pages visited) and averages in the range of
40-80 ticks per session cleaned.
That would be fine, except I may be cleaning several thousand sessions at a
shot on the 31st. The system is SQL Server and I've optimized it about as much
as I know how (there are indexes on the major columns, I've cleaned out all
unneeded data, etc).
Any thoughts on using multiple CFQuery statements vrs one more complex SQL
call? Right now, for example, I make a call to the DB to see if the session
already exists, if it does I do an update, if not I do an insert.
Could it actually be faster to do an IF statement in the SQL using only one
CFQUERY tag? It seems to me that with "maintain connections" on this wouldn't
make a difference... but I'm not sure (and want to use the time I've left
wisely).
I am using CFQUERYPARAM and caching what queries make sense.
Some other thoughts:
I've actually considered placing some of the site on another CrystalTech
account (on a different server of course) and using redirects to move the
traffic off. Of course there's no way I could get the URLs to stay the same
(outside of frames) and I wouldn't want it there all the time - just for a day
or two.
It would also royally screw up my log statistics.
Any other ideas for kludgy, cheap load balancing?
I can easily turn off the end-of-session handlers. I know that this process
will take a while, but I'm not sure if it's really the performance hog that I
fear. It is, after all, spending nearly all of it's time waiting for the
database - sitting effectly idle. So what if the clean up process takes two
minutes if the thread isn't dominating the CPU? (I will also decrease the
number of clean-ups to one every 15 minutes or so in an attempt to clear out
old sessions as quickly as possible.)
Many (sometimes VERY many) of the sessions in memory are generated from bots.
I'm considering creating a ROBOTS.TXT that would prevent bots from indexing the
site for our busy time, but I fear that would inhibit them more than I want -
if you tell robot's to bugger off do they come back?
I'm open to any other ideas. My gut says that with the resources we have we'll
just have to live with overloads unless they want to create a much simpler site
(and they don't want to do that).
Anybody got some heavy iron and bandwidth they're willing to donate for three
days a year. ;^)
Sorry for the babbling - I'm entering my normal end-of-year paranoid phase.
Jim Davis
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~|
Special thanks to the CF Community Suite Silver Sponsor - New Atlanta
http://www.newatlanta.com
Message: http://www.houseoffusion.com/lists.cfm/link=i:4:188824
Archives: http://www.houseoffusion.com/cf_lists/threads.cfm/4
Subscription: http://www.houseoffusion.com/lists.cfm/link=s:4
Unsubscribe: http://www.houseoffusion.com/cf_lists/unsubscribe.cfm?user=89.70.4
Donations & Support: http://www.houseoffusion.com/tiny.cfm/54
