JBossDeveloper

Okay, now I am running into dropped emails, after you mentioned I should check...

It seems that anything >= approx 135 threads and >= approx 100 or so emails, causes emails to begin to be dropped.

On creating a connection, i get connection refused exceptions.

I tried adjusting all the things I could think of, i.e. mySql, the JVM's, and the jbossservice (increasing the thread pool up to 100, after many failed runs)

still to no avail, now I am a loss, as to what to tweak next...any ideas andrew?

if i create a 1000 threads, and 500 emails, i lose about 400

if i create a 1000 thr and 1000 emails, i lose about 888

500/500=379
500/250=160

so, it appears at the moment, that just the connections are the bottlekneck, any insight or recommendations, greatly appreciated...

lol...no Dos attacks, etc.

My Boss, says we need to be able to handle approx 6,000 emails per second, these emails, every single one(none sent by internal users/all sent to internal users) will all come from external domains

now I know this is unreasonable, of just one machine, and one instance of JBoss Mail, but I am trying to find a threshold of such, a way, if any, to figure out how many instances, machines (a cluster) would be reqd to achieve this feat

it seems though, that JBoss, is not allowing more than twenty or so instances of a SMTP listener, as I see everything go in waves, queuing the connections, and processing them 20 or so at a time.(I could mail the output?)

Umm depending on the system 100 simultaneous connections (which means some many more number of users) may be rather large. 1000 threads would be a pretty big system. Is the system your testing big enough for this kind of load?

Sounds like you hit all the common bottlenecks. Also check your message driven bean pool (server/default/conf/standardjboss.xml Standard Message Driven....). I'd have to really look at your system in a lot of detail to tune it for heavy load (and of course you need a RAID drive). However if you're working on hte load tool now then you don't need that.

Your numbers 1000 threads for 500 emails is totally wack though. I'm not even sure the logic will handle that. I never anticipated "every other thread" will send an email and the others won't.... I do no "validity checking" in the load tool. So you can put in bogus values like that. A burst of 1000 connections comming in at once simultaneously is crazy talk....are you planning a DoS attack against Yahoo?

Lots more mails....less threads. You're not testing threading, you're testing message rate first and concurrency second (many thousands of users will not == many thousands of simultaneous threads for SMTP or POP)

i changed the containerpoolconf MaximumSize ->200 from 100

had no effect after restart

i saw theres about 5 or so more, similar settings, throughout standardjboss.xml as well as in jboss-service.xml

should i start doubling, some of these values, till my machine either blows up, or JBoss Mail Server begins spawning more threads

I am responding in depth, but for the moment, what/where is the JBCS pool size? Is that collaboration server?(BTW: whay has named changed, or name changed in forums)

I've found pool-sizes in
default/conf/jboss-service.xml
default/conf/standardjboss.xml

and in

mail.ear/mail.sar/meta-inf/jboss-service.xml

I touched nothing in def/conf/jboss-service.xml
(though I noticed there are it seems thread settings)

In default/conf/standardjboss.xml I touched, as you said, standard message driven mbean (i beleive it was) and nothing else

And in mail.ear/mail.sar/meta-inf/jboss-service.xml
i increased the SMTPProtocol threads massively

okay, understood on the 4 cpu's/boxes and Raid...

would it help much with postgresql? is it more difficult than mySql, does it run in windows?

ooh, good point, my boss had been wondering, if everything would run better on linux, any thoughts towards this?

but as I said, it seems the app is doing fine, except for the lack of extra SMTP threads

what i see in output, goes in waves, and they seem to be approx waves of 20

i'll see 20 or so connects, up to about a 100 or so, then I start seeing the mail data being processed and displayed, again in spurts of about 20

then there is a nice pause(as you noted, prob relating to mySql)and all database work begins, again, in waves of about twenty

note, during these waves, i do see, intermittent earlier threads finishing up

thas wy I was wondering if there are some more thread settings somewhere, i.e. hence the "waves/spurts"

with regards to being diskbased, hmm, my windows partition, i.e. the one with the swap file, has less that a 750mb of space, as you said, that prob is contributing, but Mail Server, is installed on another drive, but imagine, still swapping memory out of page file(which could definitley hold a grain of truth to it in that I saw that half way through the op, is seemed it was reclaiming memory)

based on last rate
i.e. 1000 messages sent in 8609ms
with 900 dropped

if i figure to retry the dropped, as a normal outside mail server would, I could...

8609ms * 9retries = 77,481ms/1000=77

So 1000 emails could be fully proccessed in approx 77 secs
(on my crap box)

so based on my boss's rate of 6,000/s

11 of my crap boxes, could completely process a 1000 emails in 7 seconds

that means if I could get the above down to 1 second (i.e. fat boxes), it would take 55 mail servers (fat boxes) too process 5,000 emails in a second

thats alot of boxes, but again, this is based on the results gathered from a crap box

settings.props

hostname=192.168.69.251
port=25
threads=100
prefix=test
start=1
end=1400
domain=192.168.69.251

Poor Windows...

My boss, said to worry about it more later, I hassle you later, and as well, as I cant let things drop, as stated will try those other settings, if you think of anything else, please post your ideas...

I think at moment he's happy with 1 per second, which seems feasible. I am just concerned, that if too many servers, try to connect at same time, they'll be lots of retries...

and my, owe so optimistic boss's and investors, do envision 25 million plus users(hint, hint, cell phone industry, carriers, and consumers)

Thanks Andrew!