kernel-based (Linux) data relay between two TCP sockets -

-

i wrote tcp relay server works peer-to-peer router (supernode).

the simplest case 2 opened sockets , data relay between them:

clienta <---> server <---> clientb

however server have serve 2000 such a-b pairs, ie. 4000 sockets...

there 2 known data stream relay implementations in userland (based on socketa.recv() --> socketb.send() , socketb.recv() --> socketa.send()):

  • using of select / poll functions (non-blocking method)
  • using of threads / forks (blocking method)

i used threads in worst case server creates 2*2000 threads! had limit stack size , works right solution?

core of question:

is there way avoid active data relaying between 2 sockets in userland?

it seems there passive way. example can create file descriptor each socket, create 2 pipes , use dup2() - same method stdin/out redirecting. 2 threads useless data relay , can finished/closed. the question if server should ever close sockets , pipes , how know when pipe broken log fact?

i've found "socket pairs" not sure purpose.

what solution advice off-load userland , limit amount fo threads?

some explanations:

  • the server has defined static routing table (eg. id_a id_b - paired identifiers). client connects server , sends id_a. server waits client b. when , b paired (both sockets opened) server starts data relay.
  • clients simple devices behind symmetric nat therefore n2n protocol or nat traversal techniques complex them.

thanks gerhard rieger have hint:

i aware of 2 kernel space ways avoid read/write, recv/send in user space:

  • sendfile
  • splice

both have restrictions regarding type of file descriptor.

dup2 not in kernel, afaik.

man pages: splice(2) splice(2) vmsplice(2) sendfile(2) tee(2)

related links:

bsd implements so_splice:

does linux support similar or own kernel-module solution?


Comments

Post a Comment

Popular posts from this blog

javascript - Count length of each class -

-
i have below code in this fiddle . html <span class="myname">john smith</span> <span class="myname">john smith</span> <span class="myname">john smith</span> <span class="myname">john smith</span> <span class="myname">john smith</span> <span class="myname">john smith</span> javascript $('.myname').each(function(){ var text = $(this).html().split(' '), len = text.length, test = $(this).val().length, result = []; console.log('outside: ' + test); for( var = 0; < len; i++ ) { result[i] = '<span class="name-'+[i]+'" >' + text[i] + '</span>'; console.log('inside: ' + test); } $(this).html(result.join(' ')); console.log('after
Read more

What design pattern is this code in Javascript? -

-
this question has answer here: why write “.call(this)” @ end of javascript anonymous function? [duplicate] 4 answers what kind of design pattern , significance of using closure? (function(){ // code here }).call(this); edit then difference between above code , following this keyword still refer same object in both ways. (function(){ // code here })(); thats invoked function expression. more info here: http://benalman.com/news/2010/11/immediately-invoked-function-expression/ the purpose run code while protecting scope (so variables declared within don't leak out global scope. update call sets value of this function being applied on. without it, value set window object, there set outer scope.
Read more

hadoop - Restrict secondarynamenode to be installed and run on any other node in the cluster -

-
i have have cloudera cdh3u4 cluster setup. have secondarynamenode daemon running on 1 of nodes. i have ensure user has root access node on cluster (for example, datanode not have other daemons running) should not able start instance of secondarynamenode daemon on node (thereby having 2 secondarynamenodes on cluster) how can achieve this? thanks in advance. add property hdfs-site.xml namedfs.secondary.http.address valuesecondarynamenode-host address or ip:50090
Read more