Loadrunner stuff
What is new in loadrunner 8.1
Loadrunner quickstart –
documentation step by step
Monitors – new integration
of sitescope monitors
New protocols
VUGen
Workflow wizard, thumbnails and transation editor,
debugging capabilities
Recording options:
HTML based script –
generates a separate step for each HTML user action.( for browser applications)
URL based script –
captures all http requests sent to the server as a result of each user action
( for non browser application)
When you
want to scale your environment for higher number of vuser. You want to record HTTP or analog mode. Context sensitive or HTML
recording heavily rely on caching mechanism. This may help your script look easier to read, however they will not scale to
greater vusers. Analog (HTTP) scripts do record each and every request independently. This will make your script look lot
bigger and cumbersome. Since HTTP statements are indepentent of each other they can help you simulate larger load (as caching
not required).
Complex system doesn't warrant for any http or html recording. It's your call based on hardware you have.
Hope this helps
In loadrunner
7.0, HTML scripts are nearly as scalable as HTTP scripts. HTML scripts are shorter, higher level, require less corrolation
and are easier to maintain. They now also run on UNIX. So HTML should be used whenever possible.
However in some cases
HTML will not work, for instance if your transactions do not come from HTML, eg. Axtive X, Java applets, or javascript. In
these cases HTTP mode is a good fallback.
Bottom line: Try HTML first. If this does not work, you can always use HTTP.
How do I obtain the routing table for my PC?
From an MS-DOS prompt type: "netstat -r" OR "route print". Both
of these commands will display the routing table similar to the one below.
Active Routes:
Try running the "webtrace" command from the command prompt of the FROM
machine.
C:\>webtrace < server >
See what the results are and verify that no errors are generated.
Example:
D:\>webtrace
superman
1 ) retry: 0 err: 0 000.000000
172.16.1.1 ? [000.000000]
2 ) retry: 0 err: 0 000.000000 172.16.2.10 SUPERMAN [000.000000]
The results here show that
the timing is smaller than 000.0000000 seconds -- very fast.
Also, using standard traceroute
shows similar data (although not as fine grained):
D:\>tracert superman
Tracing route to superman
[172.16.2.10]
over a maximum of 30 hops:
1
10 ms <10 ms <10 ms 172.16.1.1
2 <10 ms <10 ms <10
ms SUPERMAN [172.16.2.10]
Trace complete.
In VuGen:
The lr_log_message function prints messages to the output
log. This log file is in the output.txt file within the script directory.
In the Controller:
The lr_log_message function prints messages to
the log folder in the results directory as specified in Results -> Results Settings in the Controller. Note that this will
NOT send messages to the Controller window.
To use lr_log_message, you need to make sure that the following is set
in the Run-Time Settings:
1. Under General:Log, select "Enable Logging."
2. For the Log Options, select "Always send
messages."
Process/Multi threading
If you run each Vuser as a process, then the driver
(The Controller uses a driver program (e.g., mdrv.exe, r3vuser.exe) to run your Vusers) program is launched (and loaded) into
the memory again and again for every instance of the Vuser. Loading the same driver program into memory
uses up large
amounts of RAM (random access memory) and other system resources. This limits the numbers of Vusers that can be run on any
load generator.
When there is no such a huge load to distribute among three load injectors, it is using only two web
servers.
Run Vuser as a thread.
Run time setting
1.Bandwidth Networking speed
– select maximum bandwidth
2.
Multi threading – ability to run more vusers per load generator – only threadsafe protocols to be
run as multi threading
If we run vuser as thread, the contoller launches
only one driver program for every 50 users . This program launces several vusers , each vuser as thread.
The contoller uses a driver program ie mdrv.exe or r3vuser.exe to run
vusers.If we run each vuser as a process , it launches drivers program in to memory again and again for every instance of
vuser.This uses large amount of RAM and other system resources.
How to find which protocol is threadsafe
Solution: Vuser type
with "EnableThreads=1" is threadsafe
Note: Do not modify this setting.
For LoadRunner 7.6 or above:
1. Go to <LoadRunner>\dat\protocols
2. Open up the relevant .lrp file (e.g. for Web, open the QTWeb.lrp file )
3. Search for "EnableThreads". If it is
1, the Vuser is threadsafe
Vusers per load generator
50 users per load generator for machine with 512 mb memory
The primary factors affecting the script execution are Memory and CPU
Utilization
Others are Network Bandwidth and Hard Disk speed
Some rules of thumb for monitoring load would be
- CPU usage should stay below ~80-90% continuous
usage.
- Memory usage should stay within the physical
limits of the machines memory. This means that the "Commit Charge Total" should be less than the "Physical Memory Total" which
will ensures that minimal paging is done to simulate physical memory.
Memory footprint for web users
Memory required for each web vuser
|
Operating system |
Win2000 Advance Server |
XP |
|
Protocol |
|
Process |
Thread |
Process |
Thread |
|
(MB) |
(MB) |
(MB) |
(MB) |
|
WEB |
Infra |
6.82 |
13.74 |
8.77 |
8.98 |
|
Next |
5.02 |
0.84 |
4.61 |
0.83 |
Monitors info
Run time monitors
– number and status of vusers
Transaction monitor –
response time
Web resourse monitor –
measures statistics on web servers - hits per second,
throughput,http
responses, pages downloaded per second,
System resources - average load, cpu utilization, to monitor we need to configure
rstatd deamon on unox servers.
Network delays
Firewall monitors
Web server resource monitors
- Apache – http:// server name/serverstatus?auto
- # of busy servers
- # of idele servers
- Apache cpu usage
- Hits per sec
- Kbsent per second
-
Web application server resource
Database server resource
– cpu used by this session
- bytes recived via SQL * net from client
- logon current
- user calls
- dead locks
- opened cursors current
- total files open
-
Weblogic - Currently (LoadRunner 8.0), WebLogic 5.x - 8.x are supported
Loadrunner provides two different ways to monitor
weblogic server depending on weblogic version that is SNMP based moniotor and JMX monitor
WebLogic monitor (SNMP- based) gathers measurement based on information returned by the WebLogic SNMP agent.
·
Availability: LoadRunner 6.0
·
Supported Server Version: WebLogic 5.x
·
Supported Server Platform: Windows NT, Windows 2000
WebLogic monitor
(JMX) gathers measurement based on Java API. It uses the JMX interface, (which is a Sun extension to Java.
·
Availability: LoadRunner7.02 or above
·
Supported Server Version: WebLogic 6.x –7.x. For WebLogic8.x,
please refer to the additional setup information below .
·
Supported Server Platform: Windows NT, Windows 2000, Solaris 2.6,
AIX 5.1, HP-UX 11
1. Click the WebLogic (SNMP) graph in the graph tree, and drag it into the right pane of the Run view.
2. Right-click the graph and choose Add Measurement(s), or choose Monitors à Add Online Measurement.
3. In the Monitored Server Machines section of the WebLogic (SNMP) ( or WebLogic) dialog box, click Add
to enter the server name or IP address of the machine you want to monitor. Select the platform on which the machine runs,
and click OK.
4. In the Resource Measurements section of the WebLogic (SNMP) dialog box, click Add to select the
measurements that you want to monitor. The WebLogic SNMP Resources dialog box displays the available measurements.
5. Browse the WebLogic
SNMP Objects tree.
6. To measure
an object, select it, and click Add. The following tables describe the measurements and server properties that can
be monitored
The
SNMP-based monitor can only monitor up to 25 measurements.
Database
monitors
Oracle
monitor has the following tables to see V$STATNAME
V$INSTANCE
V$SESSION
V$SESSTAT
V$SYSSTAT
V$PROCESS
By default, only V$SESSSTAT
and V$SYSSTAT are available in the Oracle monitor. Using the custom query feature, you can define your own query for V$STATNAME,
V$INSTANCE or V$SESSION to the Oracle database and view the result of this query (a single numerical value) in the Oracle
online monitor graph.
Correlation
Correlation is fundamental
to replaying scripts against sites that use dynamic values such as Session IDs, Database Primary Keys, and almost all security
done over HTTP
Automtic correlation – we can set rules
to correlatre
Manual correlation : record two scripts and highlight
difference with windiiff
Counters
Average load
Bottlenecks
Functions:
Lr_eval_string – replaces
a paramenter with current value
Lr_start_transaction,lr_end_transaction
Lr_log_message – logs
a message to vuser log file
Load
average
Load Average
is defined as the average number of jobs in the run queue. So a load average of 14.4 indicates that 14.4 processes were on
average waiting to be served by the processors in your system (being 4 in your case if it was on web/app server or 2 if load
average was on your database server).
For a process to be in the 'run queue', it must be waiting on CPU (ie not waiting
on I/O or any other resource).
Memory leak
A memory leak
is the gradual loss of available computer memory when a program (an application or part of the operating system) repeatedly fails to return memory that it has obtained for temporary use. As a result, the available memory for that application
or that part of the operating system becomes exhausted and the program can no longer function. For a program that is frequently
opened or called or that runs continuously, even a very small memory leak can eventually cause the program or the system to
terminate. A memory leak is the result of a program bug.
Some operating systems provide memory leak detection so
that a problem can be detected before an application or the operating system crashes. Some program development tools also
provide automatic "housekeeping" for the developer. It is always the best programming practice to return memory and any temporary
file to the operating system after the program no longer needs it.
How to determine db delays
use another DB monitor such as Omegamon and I can get all kinds of useful stats such
as DB wait time, IO time, connection time, etc. I can drill down and find where
time is spent on a transaction. I would think LRs database monitor provides similar
capability although I am not too familiar with using the tool.
Performance counters
Unix box
-
average load
-
cpu usage
-
cpu system
-
cpu user
-
idle time
-
swap memory
-
memory usage
Web server(Apache, IIS)
CPU usage
KB sent
# of idle servers
# of busy servers
Hits/sec
Database
Logson current
Opened cursors
User calls
Cpu used by
Global locks
Physical reads
Byses sent
Weblogic
Idle threads
Que length
MEmory
|
Function Name |
Description |
|
calloc |
Allocates an array and initializes all elements to zero. |
|
free |
Frees a block of memory. |
|
malloc |
Allocates a block of memory. |
|
realloc |
Reallocates (adjusts the size of) a block of memory. |
Lr_decrypt – to decrypt
passwords
Lr_message – message
to vuser log and controller out put window
Lr_log_message – logs
messages in output.txt in vuser script directory
Lr_output_message –
to only out put window
Lr_error_message –
send error message to out put window
Lr_think_time – for
think time
Parameterization with group number, iteration number, random number
Update method
Each occurance, each iteration, once
Importing data from existing database – with database query wizard from parameter list
- by creating a new query
- by manually entering sql quer y and
connection string through ODBC source
User defined functions
A user
defined function replaces a parameter with a value returned from a function located in an external dll. Create an external
library files (dll) with function
Load dll using lr_load_dll function
lr_load_dll("C:\\MyProject\\MyDll.dll");
To register a DLL or OCX file
Method 1:
1. Open a DOS window.
2. Navigate to the DLL's or OCX's directory.
3. Enter
regsvr32.exe <filename>.<dll/ocx>.
lr_load_dll("C:\\...full_path_to_the_DLL...\\dll_name");
//Call your
own function from dll_name, it should work now
rc = ci_unload_dll(ci_this_context, "C:\\...full_path_to_the_DLL...\\dll_name");
lr_message("\"rc==%d",
rc); // check the return code for ci_unload_dll
//Call your own function from dll_name, it should fail now
