Forum OpenACS Q&A: photo.net on linux /nfs problems

Collapse
Posted by Rajeev Surati on
I was curious to understand if any of you folks are running aolserver
configurations where the files are nfs mounted onto the front end
servers. (we need to do this generally because our photodatabase
needs to be served across servers.)

Currently we have a solaris server serving as the db and fileserver.
(I've ordered another box to be the nfs system)
And a 2.2.x redhat linux box serving as the front end. While this has
mitigated a lot of load issues we were experiencing running it all
off the e450) we are having end user performance issues.

Behaviors we are observing that are disturbing:

1> files will get served up sporadically empty. E.G. HTML files with
non zero length will show up empty having been processed by ad-html
as totally empty files e.g. <html> ... </html> weher ... is just a
valid head and body with no content. Indicating it was empty to start
with.

2> IE6 reloading pages as it gets to the end of the file for some
reason.

3> On not so high bandwidth connections netscapea dn IE (mostly
netscape returning errors suggesting the connection was reset by peer
or that a page is returned with 0 data)

We are running aolserver 3.3 ad13. The e450 serving behind the load
balancer is perfectly happy and we dont' get similar results for end
users.

Does anyone have any suggestions. I would love to know if people have
working linux front end installations that load the aol over nfs. I
know aD has used solaris  x86 in this configuration successfully.

Love to hear some feedback. I have reproducibly observed this problem
going to /pvt/home.tcl using netscape.

The other problem I mentioned where the page just reloads is
often happens on these pages whereby the page will "load and display"
and then "reload" and return an error. A user using IE6 has had these
issues and brought this up. I had it happen on IE5.5 and subsequent
requests during that session seem to work out okay.
http://www.photo.net/photo/building-an-slr-system
http://www.photo.net/photo/color-printers

5> connection will be reset by peer on a lot of netscape browser and
sometimes IE. Usually the folks have different types of browsers.
What concerns me is that if the load balancer were causing this then
why doesnt' it happen when we are ruunign the servers exclusively on
the sun box?

-raj

Collapse
Posted by David Walker on
1.  Are you certain the html you're seeing was processed by ad-html?
IE will show some valid head and body html if you "view source"
during the page load.

I haven't been able to reproduce your errors from my machine but I
guess if you're behind a load balancer I might not be hitting the
right machine.

Collapse
Posted by Rajeev Surati on
Interseting point about IE.
All the loadbalancing is going across separate instances of aolserver on the 1  linux box with multiple IP addresses. Aolserver has much better performance with fewer threads and fewer db handles.

I was observing this behavior with just two such instances. Right now there are 7.  While this might be ther problem I am not sure. I am planning on instrumenting ad-html to make sure this is the case.

-raj