Mike Kelp wrote on 08/23/09 11:24 AM

Great post!

I especially agree with your points on CFC based frameworks except that unlike many, I disagree that CFCs are a big hog in CF. I argue that similar to Java, creating hundreds of objects to do nothing but display data and not use any of the logical functionality of the objects is pointless. There is a reason primitives, collections, hashtables, etc. exist and you certainly don't create a Boolean object everytime you need a bit for true/false for example.

In a service based architecture, which is the way the web is designed and handles the load it does, there should be little reason to have the same functionality instantiated so many times by returning arrays of objects / cfcs, etc. with every event. This is how the memory usage of an application becomes more volatile. Point being, while cf objects are no doubt heavier and more powerful than POJOs, I think what the community has mistaken is that the cost is not the problem, but the cost/benefit understanding people in general have as it is a bit different from the general purpose usage of Java objects.

I hope in my rambling, that I've added something to the conversation haha.

charlie arehart wrote on 08/23/09 12:06 PM

Mike, glad to hear you repeat this idea that sometimes jvm "tuning", per se, may not be quite as critical for some problems as just making some modest configuration changes like these.

That said, while setting the heap to be preallocated (xms equal to the xmx) does solve some problems and so is a good recommendation, I find it curious that for the permgen you've focused only on setting a min for it, rather than raising its max.

I've just seen plenty of instances of CF getting an outofmemory error which ultimately proved to be in the permgen, so raising the maxpermsize above the default seems a more common solution to many problems in my experience than setting its min (permsize). As you say, of course, raising either the xmx or the maxpermsize should be done with caution, ensuring there is adequate memory.

But I'm assuming you have some experience in mind that led you to say mention just the permsize, so perhaps that's something you'll elaborate on in the future threads you hint at.

Also, it seems worth repeating here (as the entry should get a fair bit of views) that another great example of an easy jvm improvement that doesn't require "tuning" is simply updating the JVM from the default version implemented at installation by CF. I know you know it, but since it was unsaid, I wanted to add it.

Of course, we should clarify that Adobe doesn't formally support anything but the version implemented at installation, but the bug fixed by 1.6_10 is just so important (for CF 8) that many can't wait for Adobe to update the supported JVM version in CF9. They just have to understand the risks.

And even in 6 and 7, the minor updates after the base 1.4.2 version were substantial, and at least the update to 1.4.2_11 was supported by Adobe for the DST change (http://kb2.adobe.com/cps/447/d2ab4470.html). They should certainly at least consider that. Updates to the JVM can have important impacts on CF. Of course, there are many moving parts and no recommendation is always best for every case.

Hope those thoughts are helpful to your readers, and certainly I mean no disrespect in dropping this in here. I will welcome your thoughts on all this, and I'm sure your readers will, too. :-) Thanks for having the discussion.

charlie arehart wrote on 08/24/09 7:36 AM

@Mike, thanks. I meant to add at the end of my (already long) comment that I did realize your main point was more on why to focus on caching and cause less work against the JVM in the first place. Yes, good point (just like caching of data can reduce the need to focus on at least some tweaking SQL and the DB connection).

If only then on this entry, I hope we can expand a little on these other aspects that have been raised.

For instance, it's interesting to hear that permgen filling is not something you've seen often and that 80m has been enough. In that case, what was the benefit of setting the permsize (since the maxpermsize would seem enough). Like I said, I figured you had some reason to mention adding the former and not changing the latter.

Not trying to sap your energy/time from the other caching entries you want to focus on. If it may be a while before you can attend to commenting on this point, no worries.

@Gary, good point. Sorry I didn't elaborate. Naturally, one should tend to apply updates/hotfixes (to CF or the JVM) only when you know you have a problem they fix. The one that the 1.6.0_10 update fixed was a class loading bug that most noticed as affecting CFC loading (I've seen it also affect just page loading on servers with many sites that didn't even use a lot of CFCs.)

There's a good blog entry discussing jvm updates for both CF 6 and 7 at http://www.trunkful.com/index.cfm/2009/1/16/Coldfusion-JVM-Version. Hope that helps.

charlie arehart wrote on 08/24/09 9:48 PM

Thanks, Mike. Interesting to hear that setting a permgen startsize has fixed problems (but there are some who have had problems solved by increasing the permgen.) Just goes to show there are no one-size-fits-all solutions. And as many will attest, we often find a lot of different proposals to how to solve problems. Can be knotty stuff.

One last point: when I pointed ou tthe blog entry at the end of my last comment, I referred to it as discussing both CF 6 and 7 updates, but I should have said 7 and 8. Didn't want people on 8 to ignore it as not applying to them.

Gary F wrote on 08/31/09 4:09 PM

Mike, apologies, I didn't realise you replied to my comment. Each server has 2GB RAM and runs 2003 Std 32bit. The jrun.exe instance running the site never eats more than 520Mb physical ram and 520Mb virtual (approx). There's 800Mb physical that never gets touched. I guess that's down to the default heap settings that come with CF?

Lance Saldanha wrote on 06/10/10 11:21 AM

Anyone have recommended stable heap sizes/JVM settings for the 64-bit flavor of ColdFusion 8?

Coach bags wrote on 06/22/10 11:24 PM

A military exercise, http://www.guccihandbagsoutlet.com/ gucci handbags outlet a shell from far away, and sent to check that the soldiers fell on farms, shells, TianZhongZheng http://www.guccihandbagsoutlet.com/ gucci outlet stand you, your tears brimming with dark clothes break ground say: stealing star cabbage with http://www.guccihandbagsoutlet.com/ gucci handbags shells are shot?

Coach bags wrote on 06/23/10 7:27 PM

For toilet fly flies restaurant said: you've been http://www.coachoutletmalls.com Coach factory outlet chased away all day, I smell xing looks to drink hot, http://www.coachoutletmalls.com Coach bags come on! Toilet fly: are a poor, eat in good light http://www.coachoutletmalls.com Coach outlet
store fart beauty pass, you have seen several?