| Time |
S |
Nick |
Message |
| 03:13 |
|
|
bruce-zu joined ##javaee |
| 03:14 |
|
bruce-zu |
##java |
| 03:16 |
|
pdurbin |
it's growing on me |
| 03:17 |
|
tjsnell |
you should have that removed |
| 03:17 |
|
tjsnell |
it'll become malignant |
| 03:19 |
|
pdurbin |
in Madison, WI I listened to WORT. the tagline was "it grows on you" |
| 03:20 |
|
tjsnell |
heh |
| 03:26 |
|
sfisque |
and yes TTYS == talk to you soon |
| 03:26 |
|
sfisque |
:-) |
| 03:32 |
|
* tjsnell |
bows |
| 03:50 |
|
|
toshiba joined ##javaee |
| 09:32 |
|
|
jenue joined ##javaee |
| 10:07 |
|
|
jenue joined ##javaee |
| 10:32 |
|
|
Quest joined ##javaee |
| 10:32 |
|
Quest |
I think tomcat has 200 (can be increased to 2000) requests limit. what about jboss or any other free server (maximum limit I mean) ? |
| 10:39 |
|
acuzio |
I think its 2001 |
| 10:40 |
|
Quest |
hm |
| 10:45 |
|
* Quest |
waits for seniors |
| 10:47 |
|
acuzio |
i just checked its 1999 |
| 10:50 |
|
Quest |
we need not less than 16000 concurent threads limit for requests. we are using 8 tomcat servers on one machine with 2000 limit each with a load balancer / proxy. are there any better options ? |
| 11:08 |
|
Quest |
elaboration = we need not less than 16000 concurent threads limit for requests. we are using 8 tomcat servers on one machine with 2000 limit each with a load balancer / proxy. are there any better options ? we are with hasle. need better ways. why one tomcate cant take 16000 or more. are there any other servers that can ? jboss? |
| 11:45 |
|
|
Xabster joined ##javaee |
| 12:19 |
|
tjsnell |
uhm, you don't need 16000 concurrent threads |
| 12:20 |
|
tjsnell |
that won't benefit you at all |
| 12:20 |
|
tjsnell |
oh spread over servers |
| 12:21 |
|
tjsnell |
16k concurrent request is crazy massive load |
| 12:47 |
|
|
jieryn joined ##javaee |
| 12:47 |
|
|
jieryn joined ##javaee |
| 13:05 |
|
tjsnell |
woah! |
| 13:05 |
|
tjsnell |
I didn't see 8 tomcats on ONE server |
| 13:05 |
|
tjsnell |
yikes, that's massive fail |
| 13:05 |
|
tjsnell |
oh damn, he's gone |
| 13:05 |
|
pdurbin |
he'll be back |
| 13:14 |
|
acuzio |
tjsnell: are you stirring up trouble again ? |
| 13:15 |
|
tjsnell |
every chance I get |
| 13:15 |
|
acuzio |
Where are the admins on this channel ? |
| 13:15 |
|
acuzio |
How come i dont have a + before my name ? |
| 13:16 |
|
acuzio |
Why is tjsnell not banned in here ? |
| 13:16 |
|
pdurbin |
heh |
| 13:17 |
|
acuzio |
i am going to take this case up right to the top |
| 13:18 |
|
acuzio |
start a petition or something - |
| 13:19 |
|
pdurbin |
well, the only admin is Fubar^ :) |
| 13:22 |
|
tjsnell |
Fubar^ and I are pals |
| 13:22 |
|
acuzio |
well - goes without saying |
| 13:23 |
|
|
zhee_tuxer joined ##javaee |
| 13:24 |
|
Fubar^ |
i'm not banning anyone! |
| 13:27 |
|
Fubar^ |
And i agree with tjsnell that 16k threads seems a bit excessive :) |
| 13:29 |
|
pdurbin |
Fubar^: want to make sfisque and whartung admins again? |
| 13:30 |
|
tjsnell |
haha |
| 13:30 |
|
tjsnell |
I thought Fubar^ was a bot :) |
| 13:30 |
|
Fubar^ |
hehe :) |
| 13:31 |
|
Fubar^ |
pdurbin: I don't think i can make anyone admin. Ask quest |
| 13:34 |
|
pdurbin |
Fubar^: ok. neuro_sys is a founder so he could probably do it too |
| 13:50 |
|
|
Naros joined ##javaee |
| 13:53 |
|
|
zhee_tuxer joined ##javaee |
| 14:19 |
|
sfisque |
well considering the fact that threads have been native since 1.3, it would be bad to spawn that many threads on an OS. talk about process starvation |
| 14:36 |
|
acuzio |
sfisque: 1.1 i thought |
| 14:42 |
|
sfisque |
they were added as an option in 1.1 (u could switch between green and native) |
| 14:42 |
|
sfisque |
1.3 they removed green threads completely |
| 14:43 |
|
sfisque |
the issue was not every OS back in 96 was posix or had posix threads |
| 14:44 |
|
acuzio |
dont remember that far back - but you seem to know this , so i will take your word for it |
| 14:44 |
|
sfisque |
by 2000 every major os had posix thread support |
| 14:44 |
|
acuzio |
but sfisque the important question is - can you kick tjsnell from here ? |
| 14:44 |
|
sfisque |
yeah, 1.1 the added native support. 1.2 native became default on some platforms, 1.3 green was removed |
| 14:44 |
|
sfisque |
i was -o'ed sometime ago. no op privs |
| 14:45 |
|
acuzio |
damn it sfisque |
| 14:45 |
|
* sfisque |
shrugs |
| 14:45 |
|
acuzio |
so other than Fubar^ and Quest no one has Op privs is it ? |
| 14:45 |
|
sfisque |
it would appear as such. |
| 14:45 |
|
acuzio |
i am deeply disappointed |
| 15:38 |
|
Naros |
like myself, I lurk here and speak up now-and-again but not enough to be oped :) |
| 15:38 |
|
Naros |
work comes before irc :P |
| 15:40 |
|
sfisque |
aye. and personally, i don't have the time to become familiar with all the nuts and bolts of "Op-ing" a channel. i know enough to function as a user, and that's alright by me :-) |
| 15:40 |
|
Naros |
yeppers |
| 15:45 |
|
|
[[thufir]] joined ##javaee |
| 15:49 |
|
acuzio |
nonsense- irc is life surely |
| 15:52 |
|
Naros |
LOL |
| 15:57 |
|
|
ariel__ joined ##javaee |
| 15:59 |
|
ariel__ |
hello |
| 16:00 |
|
Fubar^ |
hi |
| 16:02 |
|
ariel__ |
my eclipse luna no funcina with java jdk 8 |
| 16:02 |
|
ariel__ |
in ubuntu 13.04 |
| 16:02 |
|
ariel__ |
help |
| 16:02 |
|
ariel__ |
my english is so so |
| 16:03 |
|
Fubar^ |
any errors? |
| 16:05 |
|
ariel__ |
my eclipse does not work with ubuntu 13.10 java 8 open wide and start editing errors I closes |
| 16:07 |
|
Fubar^ |
what are the errors? |
| 16:07 |
|
ariel__ |
The Eclipse executable launcher was unable to locate its |
| 16:07 |
|
ariel__ |
companion shared library. |
| 16:09 |
|
Fubar^ |
maybe try reinstalling |
| 16:26 |
|
|
ariel__ left ##javaee |
| 17:42 |
|
|
Quest joined ##javaee |
| 17:45 |
|
Quest |
tjsnell, I saw your comments on chat logs about the tomcat. i was disconnected. sory |
| 17:46 |
|
Quest |
am. any senior arround? |
| 17:46 |
|
|
manish_chhabra joined ##javaee |
| 17:54 |
|
Naros |
What's up Quest? |
| 17:54 |
|
Quest |
hey! |
| 17:54 |
|
Quest |
Naros, well, sharing views here. did that in an other channel in the morning too. |
| 17:55 |
|
Quest |
we need not less than 16000 concurent threads limit for requests. we are using 8 tomcat servers on one machine with 2000 limit each with a load balancer / proxy. are there any better options ?elaboration = we need not less than 16000 concurent threads limit for requests. we are using 8 tomcat servers on one machine with 2000 limit each with a load balancer / proxy. are there any better options ? we are with hasle. need better ways. why one tomcate cant t |
| 17:55 |
|
Quest |
ake 16000 or more. are there any other servers that can ? jboss? or changing the mentioned strategy? |
| 17:55 |
|
Naros |
on a single server? |
| 17:55 |
|
Quest |
yup |
| 17:55 |
|
Naros |
how many cores are you throwing at the machine? |
| 17:56 |
|
Quest |
8 |
| 17:56 |
|
Fubar^ |
What makes you think you need 16k threads? |
| 17:56 |
|
Naros |
You believe you're going to have 16k users simultaneously doing stuff on the server at precisely the same moment in time? |
| 17:56 |
|
Quest |
Fubar^, hard stats. becuase 14000 are used right now |
| 17:57 |
|
Quest |
Naros, yes. Fubar^ we have very huge trafic |
| 17:57 |
|
Naros |
That's a lot for 1 machine to handle not to mention the fact that regardless of load balancing, you've got a single point of failure. |
| 17:57 |
|
Quest |
no. Naros its not the hardware issue. its the software bottleneck |
| 17:57 |
|
Naros |
I understand that, but making 1 machine handle 16,000 threads is overworking a tiny 8 core processor |
| 17:58 |
|
Quest |
if we run one tomcat on that machine. it cant serve |
| 17:58 |
|
Quest |
so we have to do 8 |
| 17:58 |
|
Naros |
Create 8 servers |
| 17:58 |
|
Quest |
all is fine now. but |
| 17:58 |
|
Fubar^ |
What is the bottleneck then, are you sure it's the amount of threads? |
| 17:58 |
|
Naros |
each running an instance of tomcat. |
| 17:58 |
|
Quest |
8 machines ? |
| 17:58 |
|
Naros |
Yes |
| 17:58 |
|
Quest |
Fubar^, Naros whats the max core machine available |
| 17:58 |
|
Naros |
Assuming each core is 2 hardware threads, that gives you 16x8 hardware threads of simultaneous operations. |
| 17:59 |
|
sfisque |
Q i would first find out what the optimal thread pool size is for the cpu/OS combination you are running. then tune that way. i'm guessing you are overtaxing what the OS can schedule |
| 18:00 |
|
Naros |
so while you might take 16k connections / 8 servers = 2000 threads per machine - you'd still be limited to running 16 threads at precisely the same nanosecond of computation. |
| 18:00 |
|
sfisque |
remember, each thread has an overhead cost that the kernel and OS have to pay |
| 18:00 |
|
Quest |
ok |
| 18:00 |
|
Quest |
8 machines ? |
| 18:00 |
|
Quest |
sory. ignore 6 |
| 18:00 |
|
Quest |
^ |
| 18:00 |
|
Quest |
Fubar^, Naros whats the max core machine available |
| 18:00 |
|
Fubar^ |
Quest: Have you actually done any profiling on this? All time is probably spent context switching |
| 18:00 |
|
Naros |
Quest, in most cases you split your load up across multiple servers to give you the max capacity you need without overtaxing a particular machine's CPU operations. |
| 18:01 |
|
Naros |
e.g. cluster |
| 18:01 |
|
Naros |
but as fubar indicated, you arrive at the max capacity and number of servers based on profiling the application's usage. |
| 18:01 |
|
Quest |
I think its not the machine issue. as its serving 14000 at one time now. and can boost to 30000 |
| 18:02 |
|
Naros |
14,000 threads on a single machine feels wrong |
| 18:02 |
|
Naros |
your cpu is context switching like a crazy horse |
| 18:02 |
|
sfisque |
and to be honest, unless you are running somekind of realtime hub (chat server?) i would argue that you can probably cut the thread pool way down. ultimately, if a user has to wait 0.25 seconds for a thread to be freed up in a pool, i do no think they will care |
| 18:02 |
|
Naros |
assuming those threads are all busy at some capacity |
| 18:03 |
|
Naros |
We handle well over 15,000 concurrent users with a thread pool of around 325 |
| 18:03 |
|
Naros |
on a single instance |
| 18:03 |
|
Quest |
Fubar^, Naros yes. mee too. some one advised me to use async and reactor pattern with only 200 threads instead of 16000. 200 with async will serve more than 16000. now why is that? |
| 18:03 |
|
Quest |
sfisque, it is similar to chat server |
| 18:03 |
|
Quest |
its not a web server |
| 18:04 |
|
Naros |
A request comes in, gets dispatched to an internal queue. |
| 18:04 |
|
Naros |
Then another dispatcher processes this queue |
| 18:04 |
|
sfisque |
aye, decouple the thread from the handling |
| 18:04 |
|
Naros |
When the queue element gets processsed, it puts it back on another queue to be delivered to the caller. |
| 18:04 |
|
sfisque |
so you can do multiple feeder / worker queues |
| 18:04 |
|
Naros |
Caller (original dispatcher) feeds the data to the socket |
| 18:05 |
|
Quest |
Naros, but...... the requestS coming IN needs to be in some concurrent threads. those are 16000 |
| 18:05 |
|
Naros |
this way your responding to sockets as quickly as possible, doing the I/O and other computation asynchronously as it finishes. |
| 18:05 |
|
Naros |
You can accept 16k connections, that isn't the issue |
| 18:05 |
|
Naros |
But each connection shouldn't be a 1=1 thread |
| 18:05 |
|
Quest |
hm. decoupling |
| 18:06 |
|
Naros |
This is pretty much server programming 101 :P |
| 18:06 |
|
sfisque |
keep in mind Q, not every connection is doing i/O at exactly the same time, and if they need it, they can block until a thread frees up |
| 18:06 |
|
sfisque |
if a user has to block for a fraction of a second, they won't care |
| 18:06 |
|
Naros |
nor will they be able to notice it in most cases. |
| 18:06 |
|
* sfisque |
nods to naros |
| 18:07 |
|
sfisque |
well, i'm out, gotta pick up one of the spawnlings… code strong! |
| 18:07 |
|
Quest |
hm |
| 18:07 |
|
Naros |
My MMO server alone in C++ can handle upwards of 20k connections per instance but uses only 16 threads for internal thread pool work |
| 18:08 |
|
Naros |
key is to make sure that the jobs you dispatch to the thread pool are small in nature and fully async |
| 18:08 |
|
Naros |
e.g. decoupling the network layer from the I/O and processing layer |
| 18:08 |
|
Naros |
which is what your suggestion of using reactor pattern means |
| 18:09 |
|
Quest |
how much delay time will be expected for A) 15,000 concurrent users with a thread pool of around 325 B) 15000 actual threads with 8 tomcats with a load balancer |
| 18:09 |
|
Naros |
Acceptable delay is a business question :) |
| 18:09 |
|
Fubar^ |
That's not enough information to expect anything |
| 18:09 |
|
Quest |
which one will be faster. A or B |
| 18:10 |
|
Naros |
Do you have 15k users doing something at the exact same nano-second? |
| 18:10 |
|
Quest |
yes. yes.. |
| 18:10 |
|
Naros |
Or just that 15k users are connected but aren't sending messages at the same nanosecond? |
| 18:10 |
|
Quest |
thats the issue.. |
| 18:10 |
|
Quest |
15k doing stuf |
| 18:10 |
|
Quest |
its a chat/ xml based update for cricket matches |
| 18:10 |
|
Naros |
To find the answer to your question, it's called profiling :P |
| 18:10 |
|
Quest |
you know. comentry. ads, |
| 18:10 |
|
Quest |
score updates |
| 18:11 |
|
Naros |
Setup one instance with the scaled back parameters. |
| 18:11 |
|
Quest |
hm |
| 18:11 |
|
Naros |
Then setup a second one and tune and continue this until you get optimal throughput without overtaxing your hardware. |
| 18:11 |
|
Quest |
Fubar^, Naros sfisque ok. your strategy is very much understandable, and I appreciate. where can i start learning? what keywords? |
| 18:12 |
|
Naros |
Not sure I can offer keywords Quest. |
| 18:13 |
|
Naros |
I mean this is performance tuning |
| 18:13 |
|
Naros |
of a server application based on network load and user load |
| 18:13 |
|
Quest |
what settings should i change in tomcat |
| 18:13 |
|
Naros |
I believe in server.xml is where you can specify the number of threads right? |
| 18:14 |
|
Quest |
how can i decouple the request and response |
| 18:14 |
|
Naros |
max idle thread count, etc. |
| 18:14 |
|
Quest |
ok. |
| 18:14 |
|
Quest |
and ? |
| 18:15 |
|
Naros |
I'd have to research precisely how tomcat handles maxThreads |
| 18:15 |
|
Naros |
I'm not sure whether its designed in a true reactor pattern or not. |
| 18:15 |
|
Quest |
well max threads setting is no issue |
| 18:16 |
|
Quest |
the issue is making a threadpool with request / response decoupled |
| 18:16 |
|
Naros |
Which i don't believe tomcat can do for you. |
| 18:16 |
|
Naros |
A custom web server would |
| 18:16 |
|
Quest |
what settings would have to be changed in order to do that? |
| 18:17 |
|
Naros |
Tomcat accepts connection, dispatches it to a thread and lets it operate until it finishes. |
| 18:17 |
|
Naros |
I'm not entirely sure Tomcat can support that is what I am saying Quest. |
| 18:17 |
|
Naros |
Not in a decoupled fashion. |
| 18:17 |
|
Quest |
Naros, which server do you know can do it? |
| 18:17 |
|
Quest |
jboss? |
| 18:18 |
|
Fubar^ |
netty i think |
| 18:18 |
|
Quest |
so theres no one line answer |
| 18:18 |
|
Naros |
The reactor code I have is C++, not java and is a custom built server platform. In our java app, we just tune the web server settings based on user load and statistics we've gathered. |
| 18:18 |
|
Quest |
Naros, you said you have tuned your server for 325 thread entertaining 15000. how did you do that? |
| 18:18 |
|
Naros |
Making sure that the settings for the java environment exceed the user's demand. |
| 18:19 |
|
Naros |
3 servers, each with 325 threads |
| 18:19 |
|
Quest |
oh ok |
| 18:19 |
|
Quest |
3 servers each with 325 threads entertain how much users? |
| 18:19 |
|
Naros |
But keep in mind, the user requests are HTTP, so short lived type stuff |
| 18:20 |
|
Naros |
Our current user load is 15k users and those 3 servers far exceed that load presently. |
| 18:20 |
|
Quest |
hm |
| 18:20 |
|
Quest |
ok |
| 18:20 |
|
Quest |
Naros, will that be same for a chat service too? |
| 18:20 |
|
Naros |
doubtful, different application with different requirements. |
| 18:20 |
|
Quest |
where each user cant wait more than 1 or 2 seconds |
| 18:21 |
|
Quest |
hm ok |
| 18:21 |
|
Quest |
great help |
| 18:21 |
|
Naros |
again, i'm talking purely from a web application world |
| 18:21 |
|
Quest |
last silly questions. how is embeded objects normally fetched and outputted to html . heres a small spring controller snippet in which a board has boxes and each box may have n number of sub child boxes and each box may have tasks too (sticky notes .). all are objects. http://pastebin.com/yPscDEps i dont know how many boxes each board/ box may have. how to auto get them. thats first question. lets go to the second later |
| 18:22 |
|
Fubar^ |
just loop over the boxes? What are you actually asking? |
| 18:23 |
|
Quest |
i dont know how many boxes each board/ box may have. how to auto get them. thats first question. lets go to the second later |
| 18:23 |
|
Naros |
With that type of hierarchy of data, I usually loop over the data in the controller and construct a DTO type object that I can very easily iterate over in the web UI without too many inner loops, etc |
| 18:23 |
|
Quest |
DTO? |
| 18:23 |
|
Naros |
We did something similar with a multi-tiered tree of data. So I gathered the data in the controller and created a single object that represented the view since the data wasn't view friendly. |
| 18:24 |
|
Quest |
Fubar^, Naros as i dont know how many boxes / task would be in one box and so on. i have to embed objects in lists |
| 18:24 |
|
Naros |
DTO = Data Transfer Object or some people call them VO = Value Objects or others prefer Model |
| 18:25 |
|
Naros |
Right get the lists and then you can use list.size() to determine the count. but keep in mind that if the list size is indeterminate, paging may be useful here. |
| 18:25 |
|
Naros |
or dynamically loading additional rows as the user scrolls |
| 18:25 |
|
Naros |
using jquery/ajax |
| 18:25 |
|
Naros |
i wouldn't want to load a page that contains 1000s of rows :P |
| 18:26 |
|
Quest |
pagination is not a problem here |
| 18:26 |
|
Quest |
http://stackoverflow.com/questions/7722291/data-transfer-object-sample-in-java this? |
| 18:26 |
|
Naros |
conceptually yes |
| 18:26 |
|
Naros |
sometimes you store data in a normalized fashion in the database right? |
| 18:26 |
|
Quest |
its just a simple class |
| 18:27 |
|
Naros |
but the view may need it denormalized. |
| 18:27 |
|
Naros |
so you load it using proper joins and such and shove it into a single object that mimics the view :) |
| 18:27 |
|
Quest |
so my class would just have 2 variables. a box and a task. |
| 18:28 |
|
Naros |
Sounds like a board contains a list of boxes, each box containing a list of boxes. And each box could have tasks. |
| 18:29 |
|
Naros |
At the base level some object takes a list of boxes and a list of tasks |
| 18:29 |
|
Naros |
The level up takes a list of those base level objects |
| 18:29 |
|
Naros |
But design the dto around your view :P |
| 18:29 |
|
Naros |
Not the data itself |
| 18:30 |
|
Naros |
Will save you from scripting stuff in the jsp |
| 18:31 |
|
Naros |
but you can do all this using the el-tags and no DTOs too |
| 18:31 |
|
Quest |
hm |
| 18:31 |
|
Naros |
if you want your jsp to have iteration logic of your domain classes |
| 18:31 |
|
Quest |
wil show you code soon |
| 18:31 |
|
Naros |
but i dislike that approach because if the domain model changes for any reason, the view will too |
| 18:32 |
|
Naros |
and its been my experience the view stays the same but i want to optimize the domain model or view changes but domain model remains the same. |
| 18:33 |
|
Naros |
and anywhere I can encapsulate things I do |
| 19:30 |
|
|
SoniEx2 joined ##javaee |
| 19:59 |
|
|
Quest joined ##javaee |
| 20:40 |
|
|
SoniEx2 joined ##javaee |
| 20:40 |
|
|
SoniEx2 joined ##javaee |
| 20:43 |
|
|
SoniEx joined ##javaee |
