User profile for Robert C

Nick: strange indeed! Do you see inetinfo.exe? Which isolation / protection level are you running at? Joe: Aha, I think there lies the explanation! Currently ANTS assumes that the ASP.NET worker process is running as the ASPNET account, and sets the necessary environment variables as such. I suspect that if you were able to run it as the ASPNET user, it should work - is this something you're able to do? / comments

Nick: strange indeed! Do you see inetinfo.exe? Which isolation / protection level are you running at? Joe: Aha, I think there lies the explanation! Currently ANTS assumes that the ASP.NET worker pr...

0 votes

Sorry... slight correction there (having checked the relevant bit of code): we do attempt to get the correct credentials if it is not running under the ASPNET user, but this could well be failing (it sounds like it is in this case). I think it's still worth giving it a go under the ASPNET user if you are able to. Rob / comments

Sorry... slight correction there (having checked the relevant bit of code): we do attempt to get the correct credentials if it is not running under the ASPNET user, but this could well be failing (...

0 votes

Hi Nick, I'm afraid I don't know of a straightforward reason why you're having these problems, but if you're happy to do a little debugging, then maybe we can help... If you can download Process Explorer: http://www.microsoft.com/technet/sysint ... lorer.mspx Then, start profiling your ASP.NET application, and run Process Explorer. In the list of processes, you should find "inetinfo.exe", and then further "aspnet_wp.exe". If you choose "Show Lower Pane" from the view menu, then set "Lower Pane View" to "DLLs", you should get a list of the DLLs loaded for the currently selected process. If you select the aspnet_wp.exe (the ASP.NET worker process), could you confirm whether mscoree.dll appears (and that its version is the same as you expect), and further whether RedGate.Profiler.Core2.dll does? Another test is to check the security settings and environment variables for the ASP.NET process. If you right click on it and choose properties, the Security tab should show "User" as being "<your-machine-name>\ASPNET". Similarly, on the Environment tab, could you check whether "COR_PROFILER" is set on the list of variables? Sorry that this is a rather involved process, but with a bit of luck it will let us work out where the problem lies. Many thanks, Rob / comments

Hi Nick, I'm afraid I don't know of a straightforward reason why you're having these problems, but if you're happy to do a little debugging, then maybe we can help... If you can download Process Ex...

0 votes

Hi, Can you confirm whether you're running IIS5 or IIS6, and if IIS6, whether you're running in IIS5 Isolation Mode? (You can tell this by whether ANTS warns you that it's about to restart IIS when you start profiling) Secondly, have you tried profiling your application in the ASP.NET development web server? Have a look at: http://www.red-gate.com/messageboard/vi ... php?t=3472 if you haven't already. Thanks, Rob / comments

Hi, Can you confirm whether you're running IIS5 or IIS6, and if IIS6, whether you're running in IIS5 Isolation Mode? (You can tell this by whether ANTS warns you that it's about to restart IIS when...

0 votes

Hi, There's still no way of generating a single page of output, but what has change since then is that SQL Doc 1.1 and later are able to export to CHM (compiled help) format. If you do that, you can then chose to print "this topic and all sub-topics" from within the help viewer, which might achieve the results you're looking for. Hope that helps, Robert / comments

Hi, There's still no way of generating a single page of output, but what has change since then is that SQL Doc 1.1 and later are able to export to CHM (compiled help) format. If you do that, you ca...

0 votes

Hi, I'm afraid there isn't a direct way of doing this, as we generate a seperate page for each database object. However, the following may be of help... Internet Explorer has an option in the print dialog to "Print All Linked Documents". If you generate documentation using the "HTML with frames" style (default), open the HTML file for the left-hand navigation frame ("tree.html"), and then print that, choosing to print all linked documents. This should print all the individual database object pages. Unfortunately Internet Explorer chooses to do this by printing each page as a separate print job, rather than concatenating them all into one, which means this isn't very useful if you're trying to output to PDF, for example. (Note: you may wish to try this having documented only a small number of database objects the first time, as it could result in a large number of pages being printed!) Hope that help, Robert / comments

Hi, I'm afraid there isn't a direct way of doing this, as we generate a seperate page for each database object. However, the following may be of help... Internet Explorer has an option in the prin...

0 votes

Hi Dave, Thanks for the comments. Some thoughts... 1) CHM is certainly something we're considering doing in the future. 2) Depending on how much of a "table of contents" you're looking for, you could try generating the document using the "HTML With Frames" style, then just loading the right-hand frame and ignoring the frameset entirely. In this case you'll want to load the "index.html" file in the output directory. Obviously navigation is then harder, but it might suit your needs better than always having the tree. 3) This is possible to some degree already by modifying the CSS stylesheet used by the documentation. If you're familiar with CSS, then you can find the master stylesheet (used whenever you generate a new set of documentation) in the SQL Doc installation directory, then "Style1", "Styles", and finally "Master.css". Hopefully this will let you make the documentation appear closer to your own house style. Many thanks, Robert / comments

Hi Dave, Thanks for the comments. Some thoughts... 1) CHM is certainly something we're considering doing in the future. 2) Depending on how much of a "table of contents" you're looking for, you cou...

0 votes

Hi Yevgeny, The chances are, that if the UI controls require Finalization (quite common), then on the first garbage collect (triggered by 4), they will be marked as needing finalization, and moved to the finalize queue. Only when you take the second snapshot, at 5, do they actually get garbage collected. If, between stages 4 and 5, your application is not actively allocating new objects, then it's entirely possible that no other garabage collections will be triggered, which is why you don't see the UI controls getting released until you take the snapshot (which forces a GC). Hope that helps, Robert / comments

Hi Yevgeny, The chances are, that if the UI controls require Finalization (quite common), then on the first garbage collect (triggered by 4), they will be marked as needing finalization, and moved ...

0 votes

Hi, I've just tried out your code on ANTS Profiler 3, and I see the effect you're reporting. However, after looking a bit closer, let me explain why it's there. When profiling at line-level, we obviously introduce an amount of overhead into your application - this is necessary, as we need to record timing data when every sequence point (analogous to a debugger breakpoint) is hit. In order to try and give you the most realistic idea of how your application will perform outside the profiler, we subtract the amount of time we spend in our code from the amount of time your method spends executing. This means that methods may be reported as taking less time in the results than they actually did when executing in the profiler - this is entirely correct. However, this approach can run into problems when there are multiple threads waiting on each other. When you asynchronously invoke a method, it is queued for processing on a worker thread. Calling EndInvoke() then causes the current thread to block until the result of that invocation is ready. Now, the problem here is that ANTS isn't directly introducing any overhead into that call to EndInvoke(), since it's effectively a Thread.Sleep, so we don't subtract any time from the wall clock time. However, the reason the EndInvoke takes longer than it would outside the profiler is of course that ANTS is introducing the overhead into the method being executed on the other thread. You can probably see that it would be difficult to detect these cross-thread dependencies! There is, however, something that can help: reduce the amount of overhead ANTS introduces. This is now easy to do in version 3 - profile in Fast Mode instead of Detailed Mode. This will only give you method level timings, but that is almost always sufficient to get you to the rough area of the problem, at which point I would recommend switching to detailed mode, and using a custom filter or the Visual Studio Add-in to profile only the methods or classes that you're most interested in. I'd also recommend using the hierarchy view - and in particular, the "time with children" columns. This lets you quickly navigate down to the areas of code causing you the most pain - there will almost always be one or two methods taking the vast majority of the time (in my experience at least). Note that any asynchronous invokes will "split" the hierarchy - in your example, AsyncInvoke() has no children, and AsyncCalc() has no parents, whereas SyncInvoke() shows SyncCalc() as a child. This is expected - we define method calls in the same way as the .NET runtime, which doesn't include cross-thread invocations (there's actually a lot of other stuff going on behind the scenes when you do that!). I hope that goes some way to explaining the results you're getting. Regards, Robert / comments

Hi, I've just tried out your code on ANTS Profiler 3, and I see the effect you're reporting. However, after looking a bit closer, let me explain why it's there. When profiling at line-level, we obv...

0 votes

Hi Glen, We fixed all the cases we could reproduce here in V3. One think you might wish to try is profiling in fast mode instead of detailed mode - do you still see the problem? If not, then it narrows it down to a smaller set of things that could be causing it. Do you think it would be possible to send us a copy of an application that exhibits this bug? We don't need the source code, just the application and associated PDB (and any libraries needed to get the thing to run). Alternatively, if you can come up with a small test case that fails, that'd be even better! I dare say if you can help us out here then a free ANTS Profiler T-shirt would be in order... Thanks, Rob / comments

Hi Glen, We fixed all the cases we could reproduce here in V3. One think you might wish to try is profiling in fast mode instead of detailed mode - do you still see the problem? If not, then it nar...

0 votes

How can we help you today?

Robert C

Comments