ASP.net Coding Practices vs Performance

Apogeum wrote:



Concern 2: I am storing rather big structs within hashtables (again several hundred entries in one hashtable)
 

"storing" How?  in a DB?  in a session var?

as in is your hashtable just a local class var or is it a Session[] var ??

also are you seeing the perf issue as one client hit's the server or when many are hitting the server ?

as other said It seems like you may want to re-work the XML and string builder stuff to use the .net xml classes. that might help a lot.

also if you are recusing are the string builders emiting a string when you exit ??

if you have say 10,000 string builders emiting strings that you then re-combine then you are losing much of the benefit of the string builder.... and generating a lot of string garbage to clean up.

Apogeum wrote:


Hashtables:

I am using strings as keys. the values are structs.
The plan was to store the structs within the cache but I had problems with that:

in the production environment it lost values (for no apparent reason).

since there wasn't enought time to investigate I switched to the hashtable.


I would prefer cache tough.

WOW... missed that before....

BIG TIME ISSUE SHOWN HERE!

PLEASE READ the docs about how ASP.NET works!

Cache is like a memory pool, the way it is designed is to hold data for a limited time.
Cache will get flushed when:

a) the item "times out"

b) the system needs RAM to process other tasks

c) you clear the item

d) the system re-cylces the IIS/ASP.Net Process


from all that has been posted b and or d are happening to your app.

how many users/requests?  how much ram does the server have?
how much ram is each request using?

I bet your app is grabiing a ton of ram and / or the cpu % is to high wich will cause IIS to re-start / cycle your app.
are users losing pages?  getting errors? timeouts?

thats one of the side efects of a re-start ....

aside from other things you might want to look at re-designing the app to not be so bound to asp.net.

for example:  build a seperate windows service that gets a "request" from a database record, processes the request and then saves the results.

then have the web app store the new request and give the user a way to check back and get the results of the processing.

that way you can stack up as many requests as you want and not have users sitting with a session open waiting for output.
and your app will leave iis / aps.net at a much nicer level of use.

then you can do the other work in the background of the web server or if you need to you can setup 1 or more servers to just crunch the data as fast as you need it.

also a single pass may not be the "optimal" way to process the data.
you might save a huge amount of ram by just doing the job in more than one pass...

do not assume that the only tool is a Hammer, not all problems are Nails....
more than one way to solve a problem.

web servers are *NOT* the best way to do long complex jobs.

AndyC wrote:



Apogeum wrote:  I am building the structure with a recursion, so the StringBuilders are created within each recursion.

Recursion is bad, m'kay. Recursive algorithms can always be re-written to be iterative.

In addition to what everyone else has said....

No recursion is a good thing when it's used well.

just that you need to know what it's good for and what the trade offs are speed,time,ram,code complexity and so forth.

and the different ways to handle things ....

loop unrolling, tail recusion and other stuff can be usefull or harmfull if not understood and tested.....

like picking a sort ....  qsort is good but for a small array a simple bubble sort might be the "best choice" while in other cases a shell sort may be the one to use.....

it's all a mater of having the tools and knowing how and when and where to use each one.