netdev - many sockets, slow sendto

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-Id: <E1HObHK-0000WB-00@osiris.swi.hu>
Date:	Tue, 06 Mar 2007 16:08:38 +0100
From:	Zaccomer Lajos <zacco@...hu>
To:	netdev@...r.kernel.org
Subject: many sockets, slow sendto

Hi,



I'm playing around with a simulation, in which many thousands of IP

addresses (on interface aliases) are used to send/receive TCP/UDP

packets. I noticed that the time of send/sendto increased linearly

with the number of file descriptors, and I found it rather strange.
I

searched for some related topics, but I couldn't find any (I might

have chosen the wrong keywords).

Though, I'm a real kernelnewbie, I made up my mind to follow what is

happening in the deep. I used the printk and do_gettimeofday

functions to print timestamps at certain points of the execution
path

to identify the bottleneck. Each socket is bound to a dedicated IP

address and each socket is using the very same port number, with the

help of SO_REUSEADDR socket option. I use the 2.16.19-r5

gentoo-sources on my Compaq Evo N1020v laptop. So far, I could get
to

the following point:



Function/macro: Location:

sys_send net/socket.c

sys_sendto net/socket.c

sock_sendmsg net/socket.c

__sock_sendmsg net/socket.c

(*socket_sendmsg)=inet_sendmsg net/ipv4/af_inet.c

(*sendmsg)=udp_sendmsg net/ipv4/udp.c

udp_push_pending_frames net/ipv4/udp.c

ip_push_pending_frames net/ipv4/ip_output.c

NF_HOOK include/linux/netfilter.h

ip_output net/ipv4/ip_output.c

ip_finish_output net/ipv4/ip_output.c

ip_finish_output2 net/ipv4/ip_output.c

dst->heighbour->output=dev_queue_xmit net/core/dev.c

rcu_read_unlock_bh include/linux/rcupdate.c

local_bh_enable kernel/softirq.c



At this point, I felt really helpless, and I don't even know if I'm

at the right track in this investigation. So, can someone please

help me with advice or links pointing to the info I need? Let me
know

if you need any further input.



I expect that sending a packet should not depend on the number of

file descriptors (sockets), it should be a more or less constant
time

operation. Can you confirm this, or is there something in the kernel

or elsewhere preventing it?



Any hints are welcome. Thanks a lot!

/Zacco 




_____________________________________________________________________
 50Mbyte kereskedelmi webtárhely + korlátlan számú e-mail + php + sql
       havi 1,365 Ft-ért!  Bővebben:  http://www.swi.hu
  
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html