Simultaneous Requests

[Newsgroup_User]

Hi
I'm after some up-to-date info on how many simultaneous requests each CPU
(or CPU core) can handle for CF, these days.

The old yard stick was - say - three per CPU:
http://kb.adobe.com/selfservice/viewContent.do?externalId=tn_18540&sliceId=2.

More recently people have pointed out that doc is a few years old (it's
older than the "last updated" timestamp suggests: it's from around 2003-4,
from memory), and CPUs these days can handle maybe 8-10 per CPU, and for
multi-core CPUs, that's per *core* not per physical CPU.

And that's always been in relation to the standard CF install, not the
multi-server version. For the multi-server version I've never seen
anything except "it defaults to 25".

Now I'm faced with a virtual server set up using this technology:
http://en.wikipedia.org/wiki/Logical_Domains. So what do I count as "CPUs"
in this case?

Obviously I realise the most accurate answer to this sort of question is
"it depends", where it's depending on the type of app (CPU intensive, DB
intensive, etc), but I'm just after a rule of thumb, or people's own
experience.

I've been working on eight-per-CPU (-core), but that's really not an
infomred position it's just based on something I read somewhere, at some
stage.

Thoughts?

--
Adam

[cesarmejia]

You could set up as many as you would like the question is: Do you have the RAM to do it! Every instance of JRun does consume a 512MB and ColdFusion alone another 512MB, that means that you would need 1GB just for the software to run, plus anything else for your apps, sooo, the question is more ram or more boxes to run more instances and more apps under one big cluster? As Ben Forta describes under scalability and resources, there's no use to have more Processors since 2 are good enough than 4, since the curve for 2 servers is higher than the one for 4, another article that you could use is the following: http://coldfusion.sys-con.com/read/42050.htm

hope it works!

[Newsgroup_User]

> You could set up as many as you would like the question is: Do you have the RAM
> to do it! Every instance of JRun

Cheers for taking the time to respond, but you don't seem to be answering
the question I'm asking.

I'm not talking about JRun instances, I'm talking about the number of
simultaneous requests CF is configured to process (see the settings screen
in CF Administrator).

--
Adam

[cesarmejia]

hmmmm, ok, uhm...we have manage as much as 500 a ms, but that is because we pushed the servers to the limit. What area are you planning, meaning, what kind of traffic or app are you planning to run on CF, because we could handle that amount only because we let SQL do a lot of work for us and CF just the light part!

[Newsgroup_User]

> hmmmm, ok, uhm...we have manage as much as 500 a ms,

OK, look, you don't understand my question, so it's perhaps best you leave
it. I do - honestly - appreciate your efforts though.

I think if you actually read my initial post (and the included links)
rather than just picking up a couple of words and guessing what I might
mean, then... it might have helped. At least helped you understand what
I'm asking anyhow.

[deep - melodramatic - breath]

There's a setting in CF Administrator which sets how many simultaneous
requests CF will service. It's loosely analogous to how many threads the
JVM will allow to start to process said requests. A rule of thumb number
for this was published some years ago (three per CPU), but it is out of
date, given the advances in CPU performance and architecture. As time
passed, people have stated that it's more like 8-10 per CPU, although that
info is also a couple of years old now. I am asking what people's findings
are currently.

I'm *not* asking "how many requests per second I can make my CF server
process", nor am I asking "how many JRun instances I can run on my server".

Now. I'm all ears if you actually understand the question and have
anything to offer. But, to be honest, I don't think you get it. So never
mind.

Thanks for taking the time to respond though. As I said, I do appreciate
it.

--
Adam

[Grizzly9279]

Unfortunately I don't believe there is a one-size-fits-all answer. This blog has a good discussion on it (with helpful links):
http://www.coldfusionmuse.com/index.cfm/2006/5/4/higher.simultaneous.requests

If you're in a multi-server environment with high end hardware, I would start with the default of 25. Run some load tests and benchmark the performance impacts of ratcheting this setting up or down. The "sweet spot" is going to be different for everyone.

Generally speaking, the old rule of thumb applies to CPU cores, not physical CPUs.

[Guest]

Adam - I've always used the general rule of thumb of 6-8 threads per CPU (core, not socket). I called Support not all that long ago (8-12 months or so) and they said that was pretty much the standard when I asked.

That setting has served me well over the years. The thing you need to watch out for is running out of threads. Keeping an eye on your logs will help.

But as Grizzly suggested you may want to do load testing.

And the rule of thumb is the same for CF Standard versus multi-instance.

Finally, for the record, I've got a single box dealing with 3.5M page hits/day on a pair of single core Xenon 3.4GHz processors with 2 J2EE installed instances using 14 threads each and we're averaging around 0.25 seconds to render pages.

[Newsgroup_User]

Cheers to both you and Grizzly.

I was beginning to give up hope :-)

--
Adam