空间管理您的位置: 51Testing软件测试网 » 子木清风 » 日志

不要追求绝对的公平，红尘之中没有公平而言，人活一世，难得糊涂。　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　it is no use doing what you like, you have got to like what you do.

ERROR 263377

上一篇 / 下一篇 2007-06-04 18:11:28 / 个人分类：Mercury LoadRunner

查看( 713 ) / 评论( 0 ) / 评分( 0 / 0 )

每日随笔

每日随笔:

f it is working in the LoadRunner scrīpt up to a point, then the problemprobably(see footnote 1 at end of post) exists in the system under test. Are you or anyone else monitoring the system under test? The first places to look would be at the web and app-servers:
1) heap sizes
2) number of jvms
3) number of jdbc connections
4) heap pattern
5) container threads
6) web component connections should be equal to or less than the http server connections
... much more if these are not involved in the root cause ...

Are these configured properly?
Is there evidence of maxing out the heap and going beyond?
Is garbage collection running too frequently and for long periods of time?
... and so on...

Footnote 1:
There is still a chance that you may be contributing to the issue on your end (not to say the sys-under-test is not the issue - and that it may require tuning or scalabiity work in 2 dimensions), depending upon the answers to the below questions AND the computing power/configuration of the servers-under-test...
1. How fast are these Vusers executing? (with or without think time)?
2. Are the Vusers simulating a new user each iteration?

Finally...
Increasing the web_set_max_html_param_len to 819200 is just not good - unless one has to deal with .NET and large cumbersome viewstates. Think about that value. Understand the impact of the RAM consumption by the Vusers. This is called "shooting oneself in the foot".
It appears that it is a bottleneck in the environment.
I suggest the following:
Determine how often and how long garbage collection is running. Monitoring by either LoadRunner, SiteScope, or tools used by the SysAdmins should be able to tell you that.
If load-balancing is being used, what is the method? Are the Vusers being load-balanced properly? Do you need to use IP Spoofing?
I recommend changing:
a) both JVMs to 256/768, (what is the reason they are different to begin with? Perhaps there is good reason - if not, change to recommended.)
b) JDBC Max to 75, and
c) Consider using another JVM.

Are the Vusers getting HTTP-500s?
Are you getting any premature connection shutdowns/closures?

Note! At this point, the troubleshooting should involve some key technical people on your end. And, by far the largest point here…
What must the architecture support in terms of transactions and types per second? Are you perhaps putting a load on that goes well beyond what is required?

收藏举报

TAG:

查看全部评论