|
|
| Author |
Message |
|
| ostinelli |
 Posted: Wed Apr 08, 2009 8:36 am |
 |
|
|
Joined: 21 Dec 2008
Posts: 3
|
hello all,
i have had many issues with the relative slow speed of message passing between nodes. i have therefore been working on it lately, and have been able to somehow triple the message sending speed between nodes. i've published my results here:
http://www.ostinelli.net/boost-message-passing-between-erlang-nodes/
i'd be delighted to discuss about it, if anyone is interested in digging this deeper.
cheers,
r. |
|
|
| Back to top |
 
|
| klaar |
| Posted: Sun May 17, 2009 11:56 pm |
|
|
|
User
Joined: 06 Oct 2008
Posts: 11
Location: Göteborg/Sweden
|
http://www.erlang-consulting.com/thesis/tcp_optimisation/tcp_optimisation.html
Only had time to skim through it, but it is about Erlang TCP/IP performance on three OSes at least. Although it is about a large number of connections, it should be a good resource for you.
As far as i understood, you are trying to stuff as much as possible through a fat tube. The most likely suspects then would be too small TCP windows and too small buffers.
TCP window scaling is enabled by default on Debian Lenny and OS X Tiger, so that should't be a problem. 
And your test seems a bit off. Why do send 200,000 messages in parallel first, and then process them sequentially on the other node? That's two scenarios at once. |
|
|
| Back to top |
|
| ostinelli |
| Posted: Sun Jun 07, 2009 11:51 am |
|
|
|
Joined: 21 Dec 2008
Posts: 3
|
thank you for your response, klaar.
i am familiar with the study you are referring to, but nothing in that study shows an increase by more than 200% via various optimization, which is what happens on OSX for instance using my queuing mechanism. btw: i already am using kernel polling and such.
this is why i'm feeling curious: i'd love not to use such a mechanism, and would really prefer to go more low level to solve this. as you correctly state, i too do believe this has to be solved at tcp level.
the test is performed in this way because it reflects a real need of an application i'm developing, where loads of client threads have to go through a bottleneck of a single registered process.
cheers,
r. |
|
|
| Back to top |
|
| wuji |
| Posted: Wed Aug 15, 2012 7:33 am |
|
|
|
User
Joined: 10 Aug 2012
Posts: 654
|
her business and she says she nearly lost her kids.Both kids.Both [h3]cheap Ralph Lauren[/h3] kids.Both women are now clean and they have this simple
for any mom considering taking Adderall, "don't.""It's pretty addictive," said said cheap replica *beep* said Degree. "It can happen to anybody."George Zimmerman's Story 'Marginally
With Life Threatening Attack: CopLead Investigator Reassigned to Midnight Shift Shift cheap designer *beep* Shift on PatrolBy MATT GUTMAN and AARON KATERSKYSANFORD, Fla., June
2012 The lead homicide investigator in the shooting death of of [h3]cheap Ralph Lauren[/h3] of Trayvon Martin voiced strong skepticism of George Zimmerman's account
the shooting, describing Zimmerman's injuries as "marginally consistent with a a cheap Ralph Lauren Polo a life threatening violent encounter," according to documents released today |
|
|
| Back to top |
|
|
|
All times are GMT
|
|
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
You can attach files in this forum
You can download files in this forum
|
|
|
Digg It
Del.icio.us
Reddit
Facebook
Stumble Upon
Technorati
