MICHAEL MARTINEZ <Michael_at_xenite.org> WROTE:
> The one thing that most causes me grief in remote
> banner systems is the timeout factor. If the server
> won't timeout quickly, my pages won't load. I don't
> care how much money the other guy loses, *I* lose
> traffic when the page just sits there trying to load
> an ad from a server that won't or can't comply.
So far what we've been doing is using the firewall to
time-out requests after three seconds. But then
somebody's sitting there in front of a broken image or
empty IFRAME - that's no good...
Right now, all things considered it takes around 2
seconds for our current system to serve, that's what
I'm seeing from various boxes around the world,
including everything - even the DNS lookup to the ad
server. It's in development so it's hard to measure
really, it'll get faster as the code is optimized, but
slower as it goes live and is serving real traffic.
Lately I've been thinking - why not have a "fall-back"
mechanism of some sort... that is, inside the "object
selection" algorithm where it is determining which
objects to serve, it has a "queue" of valid items to
serve, perhaps it could attempt to present the
first-priority item, and if it times out in 3 seconds,
go on to the next, etc... there could be a "global"
time-out variable so that if it diddn't load in X
seconds total it would fall back to some "fall-back
There is also the prospect of caching ads that come
from third-party servers. But when I mentioned this the
Clients didn't seem to like the idea.
Mainly the delay is caused either by one particular
third-party server, or else by the crazy functions the
clients are asking for - which proxy content around, or
do other network-intensive things. Even a DNS lookup
can take 10 seconds, way too long.
Making all the ad servers fast will never happen, the
servers are always maxed-out as a matter of policy, but
I think the key is being able to control this aspect,
to be able to detect slow service and to manage
delivery based on that, then the server operators will
invest in faster servers and more bandwidth and start
to be concerned more with performance, since the
clients will be tracking it too.
We got asked to be able to detect if an ad image is
broken, so we might as well detect boggy serves and
stuff too, and log it all up for the admin to have a
look at, like "ad #5 had 1,540 time-outs over 3 seconds
waiting for content delivery this month"... We should
probably draw the line somewhere and make it a default
feature. That'd be nice, because nobody'd be going
around screaming about our software being slow, hehe! ;)
> Is it a problem with the Internet? I don't think so.
> I'm running code from 3 or 4 services right now, and
> only 1 of them consistently drags down my pages'
> performance. I'm sure it's a server-side issue. I did
> have ads from another service for about a month that
> had similar problems. I've heard similar complaints
> from other people about both services (which I see no
> point in naming).
The more features popped in, the more it loads the
network and processors - - but we should still be able
to set a minimum acceptable serve speed.
The system currently let's you reserve a set of
"super-user" objects, which would normally be stored
and served from the same server (fast), this is what
gets displayed if a remote image is broken etc...
perhaps it could be made to select these if content is
taking too long to load...
So far, leveraging the firewall has been an amazing
help. It can detect and correct most of the nasty
things that were happening on the Client's "old
system" - such as super load spikes from a single
client, or thousands of SYN_RECD connections open which
never time out, and stuff like that.
What then should the minimum acceptable serve speed be?
I am thinking 3 seconds but that's fairly optimistic.
Mabye - 3 seconds for an individual serve, 9 seconds
total (so it can try up to 3 before falling back). And
it can log the time-outs for the administrator - in the
future if it's a huge issue we could put something in
that will automatically de-select items which pass a
threshold of time-outs...?
Hey - IDEA! Why not record "time to serve" statistics
too? Sounds like people would like that... so they
could see per-object what the average serve time is and
get a sorted list. They could then really easily see
what sort of performance the different servers are
delivering. And have solid numbers to show...
Thanks very much for the suggestion, we'll definitely
be looking into this stuff!
Zak Power / ZENCOR
(800)-759-9826 / (416)-820-3304
Received on Wed Oct 11 2000 - 13:53:14 CDT
HOW TO JOIN THE ONLINE ADVERTISING DISCUSSION LIST
With an archive of more than 14,000 postings, since 1996 the
Online Advertising Discussion List has been the Internet's leading forum focused on professional discussion
of online advertising and online media buying and selling strategies, results, studies, tools, and media
coverage. If you wish to join the discussion list, please use this link to sign up on the home page of the Online Advertising Discussion List.|
Online Advertising Industry Leaders:
Local SEO with Video
Austin Web Design
Add your company...