I'm currently performing stress tes

[sebroj]

I'm currently performing stress test on our web server(SilverStream 3.7.2 & Oracle 8.0i) using the MS Web application stress tool (aka WAS).
After running one of the recorded scripts, (ca 1 hr) with 100 users, 2 threads(multiplier X10), we received a lot of hits. but the problem is that 80% of the total hits were invalid. Among few error codes(404) we come across somethign like: Code 302 Service temporarily moved or even code "NA" HTTP result code not given.
My question is what really happen with silverstream when 100 users simultaniously start too run the same tasks for a period of time. Why do we encounter 80% failures in the hits?
here is one of the results i got by running a test for 1 hr:
Hit Count: 75893
% Total Hits: %100.00

Result Codes
Code Description Count
--------------------------------------------------------------------------------
200 OK 6034
500 Internal Server Error 147
NA HTTP result code not given 69712

Socket Statistics
--------------------------------------------------------------------------------
Socket Connects: 70417
Total Bytes Sent (in KB): 25812.07
Bytes Sent Rate (in KB/s): 7.17
Total Bytes Recv (in KB): 88189.83
Bytes Recv Rate (in KB/s): 24.50

What is the Error Code NA means?

 

[ihbrune]

i'm now performing also some stress tests on our server to find a long lasting bug. we are using apache jmeter.

i did not get error messages of the kind you described. But i had some problems in jmeter with pages doing redirections.

also i received not very helpful messages when urls are wrong.

 

[sebroj]

Guess what!
It better if you update the silverstream
we are using right now the 3.7.4 version with oracle 8i
with some patches also,a nd the servers perform much much better. even our clients are happy

In addition I tried to setup oracle9i with Silverstream3.7.4
it's useless!!! don't even try! the apps are not compatible.
As for the oracle8i make sure thatis installed on Pentium3. I got many trouble with install ation on P4 and even when i got it, everything seems to suck, including Silverstream

 

[ihbrune]

Hi,

we are using 3.7.4 too. Runs on a 2 CPU solaris box. Oracle runs on a different, even more powerful server also with 2 CPU's under solaris.

So: speed is not the issue here. i work for a university and we don't have more then 300.000 hits per day at this time and about 1.000 concurrent session.

I am only testing to find this nasty bug which causes the server to block sometimes. Yesterday I made a stunning discovery:

if I start with 200 concurrent users in one second then my test server (also solaris with 2 cpu's) almost for sure blocks and has to be restarted. And this even happens when I use random and useless urls !

Now i am frightend that someone will do this on our production server. Did you make similar experiences while you r testings ?

-Henning

 

[sebroj]

Yes indded, I made this kind of testing, and yes it's true, the server had to be restarted.
the difference between ur servers and ours is that u got a PC with 2cpu I got only one.
the DB is running only on P4, the rest (3 PC) got only silverstream. Before I had to instal db and silverstream on each machine to perform testing.. ouch!
now when it takes to the hits, yes, I got a lot of problems with this b4 specificaly when 200-300 user were performing the same task over and over.. it was madness!

 

[ihbrune]

sounds bad to me. we are using ss now for 2.5 years and since half a year we have this problem from time to time. what makes it even worse is that after the server crashes under heavy load one can't get it back online because it crashes again immediately.

i send my testing results to silverstream support and wait for their answer.

but if there is no solution then I guess we will have to leave silverstream even if i like the tool very much.

what is your plan ?

-Henning

 

[sebroj]

Recently, our company decided to update the server applications, and we going with jboss
more stable and no need of SS. Plus Jboss supports more DB versions, not only Oracle.

The joboss is much more stable but I didn't do any testing yet whatever abt the user hits. our application is still in dev.
as soon as I got some good new I let you know

 

[ihbrune]

Looking for it ;-))

 

[sebroj]

Btw as far as i can see thew good stuff abt jboss is that you can have 2 differen application served from one machine:
ex. SS and jboss are installed on the same machine
SS serves one application and is running as a service
jboss serves otehr application (i'm not sure yet but youcan run it a s a service too)

 

[ihbrune]

I have no idea how ss behaves on nt but on our solaris installation we are doing something similar:

we have two ss servers running under different user accounts and different ports on the same solaris box. Both servers are completly independant though at this time they work on the same oracle db.

further in one of the ss servers there are several different application areas (this maps in ss to different oracle db users). These applications live independently on the server and only share the user authentication facilities of silverstream.