Forum .LRN Q&A: Re: Help Needed in Setting up .LRN to Scale

Collapse
Posted by Andrew Piskorski on
It would be interesting to find out what mysterious bug or configuration problem is causing this "Heidelberg's dotLRN runs pathetically slowly even when only one user is requesting one page at a time" problem.

But you may well be wasting your time. One, that problem is reasonably likely to be peculiar to your particular installation on that particular Sun box. Two, even if you fix the problem, and especially given that you're running a bunch of other software on the same low-end Sun box, that box is likely to still be far too wimpy to handle the loads you expect.

So why don't you just buy a dual Opteron or dual Xeon Linux box with 4 to 8 GB of RAM and a bunch of fast RAID disks, set that up, and do whatever further debugging and tuning you need to there? Maybe the mysterious performance problem will not reappear, which would be a nice bonus. If it does reappear on the new machine, that also would tell you something and might help your debugging. But the main point is that the few anecdotal reports we have from current high-volume dotLRN users seem to say that your current shared Sun box is unlikely to meet your needs, and that you're going to need a new machine anyway.

Of course, I suspect Mike and Janine know that, so what's the deal? No money in the hardware budget currently to buy a Linux box? Do you really think that shared Sun 280R will meet your client's needs? Or what?

Time constraints? Getting a new machine up and running going to take quite some time, of course. (Even longer if the customer has bureaucratic purchasing rules.) But Furfly already has various other Oracle installations up and working, right? So how about setting up a Heidelberg Dev site on an entirely different machine, using known-good hardware and a known-good Oracle instance? If the mysterious problem re-appears there as well, then you know with about 99% certainty that it's not the hardware or Oracle, that the mysterious problem has got to be in your site OpenACS, dotLRN, or AOLserver code or configuration.

Collapse
Posted by Carl Robert Blesius on
Thanks for the feedback Andrew.

It does indeed look like this has to do this particular setup and has nothing to do with .LRN/OpenACS. We just did a comparison on a MUCH SMALLER Sun box (with cob webs and all) and .LRN was faster than what we are experiencing now.

We are moving forward and will report when we find out the exact problem for posterity.

Carl

P.S. Mike wrote, "we're running several busy Oracle sites, one with a database in excess of 10GB and over 15 million hits per month on less hardware" and I am sure we can do the same thing with a .LRN site, it is just a question of some of the .LRN users cooperating on making it happen with gradual improvements over time.