In LoadRunner 7.x under Recording Options, you can use the HTML-based script or URL-based script for recording purposes. But I notice that when you gather a series of transaction times and do a standard deviation, the HTML response times are always much shorter (cumulatively) than the URL ones and the STDEV always seems to be around 0.005 or so. This seems to suggest that the HTML-based recording is not the way to record because some algorithm is getting smoothed over. Given the descriptions of the two (URL says it records "all requests and resources from the server"), I have to wonder why HTML recording would even be an option. (I have noticed this same problem with the HTML-based recording whether or not you set the option to "A script describing user actions" or "A script containing explicit URLs only".) The other difference seems to be that the URL-based scripts do pull down things like style sheets but the HTML-based scripts do not mention these. I wonder if somewhere this is causing the variance. (Note that I am recording non-HTML generated elements when I do HTML recording so I know the style sheets are not just being left off because of a bad setting on my part.)
So I guess my questions are: Has anyone else noticed this discrepancy? If so, how do you account for it? If you use HTML-based recording, what benefit does that specifically give you?
I must admit I never noticed the smoothing effect of using html mode. Have you tried this out on several websites and for different lengths of scripts? Maybe this is just unique to your situation????
I switched to url mode because of the control it gives you over what the script is doing. You can control what items get downloaded which ones do not. You also do not have to worry about the flow of functions. For instance a web_link only works if the previous call had the url explicitly stated in it. I like to modularize my long term scripts so placing each one of these calls in seperate action statements did not work.
I'll give my two cents. In simple world, the two modes do the same. HTML mode just doesn't record the downloading of gifs, jpegs, css, etc. But when you execute an HTML recorded script it still handles those objects. The main reason that you use URL mode if for those non-standard objects that HTML mode doesn't seem to work well with. For example, adobe attachments. HTML mode does not seem to work well with file downloads, etc. So Mercury suggested using URL mode. URL mode also does allow greater flexibility of setting so that you can control file types. But variances in your responses must mean that your site has something unique that HTML mode can't. So to make a long story short in your case URL mode maybe better, but there are certain cases where its ok to use HTML mode. For one thing it makes your scripts alot shorter!!
Thanks for the help guys. I guess what I'm noticing is it also has to do with the "Resource=1" or "Resource=0" tag that gets added in. Example, I have some .css and .js files that make sense as Resource=1 but I also have some .jpg files that VuGen automatically set to Resource=1. I am not sure why the image files would be that since you would always want those to be brought down unconditionally, no? (Granted there might be a cache but my understanding is that Resource=1 is not the same thing as simulating the cache.) I'm more concerned with this because I want to make sure that the user I'm simulating is actually representative but that Resource=1 leads me to believe I might have some issues given how the manual defines what that means. Either way, when I normalized all the Resource= tags, the smoothing over effect vanished and the STDEV of both recording methods matched up.