[Contiki-developers] csma operation with full queue
Mohammad Abdellatif
iam using csma in a network where packets are always generated and sent to the sink.
when the nodes in the middle have their queue full they broadcast the new received packet directly "normal"
however since more packets are always coming, i find out that as soon as the queue gets full, it will never be emptied and new packets will always be directly sent. which affects the performance of the whole system.
i think there should be a mechanism to avoid this
has anyone faced this problem before, i think there is a bug here somewhere but i cant put my hand on it just yet.
any help is welcomed,
Thanks,
Mohammad Abdellatif
David Kopf
Mohammad Abdellatif
------------------------------------------------------------------------------
10 Tips for Better Server Consolidation
Server virtualization is being driven by many needs.
But none more important than the need to reduce IT complexity
while improving strategic productivity. Learn More!
http://www.accelacomm.com/jaw/sdnl/114/51507609/
_______________________________________________
Contiki-developers mailing list
Contiki-d...@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/contiki-developers
------------------------------------------------------------------------------
10 Tips for Better Server Consolidation
Server virtualization is being driven by many needs.
But none more important than the need to reduce IT complexity
while improving strategic productivity. Learn More!
http://www.accelacomm.com/jaw/sdnl/114/51507609/
_______________________________________________
Contiki-developers mailing list
Contiki-d...@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/contiki-developers
Mohammad Abdellatif
my question is that, shouldn't there be a limit on how much time a packet stays in the queue?
the packet should not be dropped only if it was transmitted or passed the max retransmit times
Mohammad Abdellatif
Simon Duquennoy
However, as David wrote, this will be useful only if your data rate isn't steady. If it is steady, you have to work on improving the throughput of your network or decreasing the application data rate.
Cheers,
Simon Duquennoy, Postdoc @ SICS http://www.simonduquennoy.net
Mohammad Abdellatif
thanks,
Mohammad Abdellatif
Simon Duquennoy
------------------------------------------------------------------------------ Learn Windows Azure Live! Tuesday, Dec 13, 2011 Microsoft is holding a special Learn Windows Azure training event for developers. It will provide a great way to learn Windows Azure and what it provides. You can attend the event by watching it streamed LIVE online. Learn more at http://p.sf.net/sfu/ms-windowsazure
Mohammad Abdellatif
when enabling the debug in the csma file, i find out that it says:
csma: could not allocate packet, dropping packet
without the queue getting full.. i had to reduce the sending rate much much more than earlier "with the older csma version" in order to avoid this
i dont know what makes the csma drop the packet when the queue is not full?
anyone has any suggestion?
Mohammad Abdellatif
Simon Duquennoy
1) How do you know it isn't?
2) What configuration do you use? If you enabled queueing in flash, you have to set COFFEE_CONF_DYN_SIZE as done the project-conf.h of examples/udp-stream
Regards,
Mohammad Abdellatif
1) How do you know it isn't?
2) What configuration do you use? If you enabled queueing in flash, you have to set COFFEE_CONF_DYN_SIZE as done the project-conf.h of examples/udp-stream
thanks for your help
Mohammad Abdellatif
Mohammad Abdellatif
nodes drop packets because they cant allocate them in the memory, even though the queue is not full!
i dont know where is this problem is coming from and even with a very low rate it still happens.
any one knows why this is happening?
thanks
Mohammad
David Kopf
there are a lot of broadcast strobes, and with rpl they fly back and forth
for a few seconds. That could be the trigger for the queue filling. Are you
using a recent git? there was a problem with the callback that unloaded
packets a couple of months ago.
the
Mohammad Abdellatif
Mohammad Abdellatif
Regards,
Mohammad Abdellatif
http://sourceforge.net/projects/contiki/develop
Mohammad Abdellatif
Mohammad Abdellatif
------------------------------------------------------------------------------
Write once. Port to many.
Get the SDK and tools to simplify cross-platform app development. Create
new or port existing apps to sell to consumers worldwide. Explore the
Intel AppUpSM program developer opportunity. appdeveloper.intel.com/join
http://p.sf.net/sfu/intel-appdev
_______________________________________________
Contiki-developers mailing list
Contiki-d...@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/contiki-developers
------------------------------------------------------------------------------
Write once. Port to many.
Get the SDK and tools to simplify cross-platform app development. Create
new or port existing apps to sell to consumers worldwide. Explore the
Intel AppUpSM program developer opportunity. appdeveloper.intel.com/join
http://p.sf.net/sfu/intel-appdev
Mohammad Abdellatif
i am showing the pritfs from csma and i can see that the packets r being dropped while the queue is not full.
going through the csma.c i can see that this print is triggered when q = memb_alloc(&packet_memb); returns NULL
i am not using contikimac.. i am using sicslowmac.. rpl operation is working ok.. the problem happens when i start sending data..
the problem is that it happens randomly..
David Kopf
list or pointer.
The memb_alloc routine just does
for(i = 0; i < m->num; ++i) {
if(m->count[i] == 0) {
/* If this block was unused, we increase the reference count to
indicate that it now is used and return a pointer to the
memory block. */
++(m->count[i]);
return (void *)((char *)m->mem + (i * m->size));
}
}
the
Mohammad Abdellatif
Mohammad Abdellatif
Regards,
Mohammad Abdellatif
http://sourceforge.net/projects/contiki/develop
Mohammad Abdellatif
q->buf = queuebuf_new_from_packetbuf();
and if this q in NULL it means that the queue is full
and this never happens.. because the packet is dropped before it goes there
i think something is messing with the memb like u say but i dont know how to catch it.. i used the stack watcher in cooja and no stack overflow happened..
do you know of anyway to check what is messing with it?
thanks
Mohammad Abdellatif
David Kopf
every case. Did you enable debug in quebuf.c?
Mohammad Abdellatif
Mohammad Abdellatif
David Kopf
abruptly.
Mohammad Abdellatif
thanks,
Mohammad Abdellatif
David Kopf
overflow would affect other things first (the last entry is uip_udp_conns).
However writing off the end of packetbuf would set the bufmem count array to show no free entries. Do you memcpy or sprintf to that
using a variable size?
00800cba l O .bss 00000002 packetbufptr
00800cbc l O .bss 000000b2 packetbuf_aligned
00800d6e l O .bss 0000000f bufmem_memb_count
00800d7d l O .bss 0000001e bufmem_memb_mem
00800d9b l O .bss 00000002 refbufmem_memb_count
00800d9d l O .bss 0000006a refbufmem_memb_mem
00800e07 l O .bss 0000000f buframmem_memb_count
00800e16 l O .bss 00000c4e buframmem_memb_mem
00801a68 l O .bss 00000002 lastport
-----Original Message-----
From: Mohammad Abdellatif
Sent: Thursday, December 22, 2011 9:15 AM
To: Contiki developer mailing list
Subject: Re: [Contiki-developers] csma operation with full queue
i am pretty sure this is the problem.. can u help me with doing that if u have the time?
thanks,
Mohammad Abdellatif
Phd student at Faculty of Engineering University of Porto (FEUP)
Researcher at Inesc Porto, UTM, WiN
http://win.inescporto.pt/mma
On Wed, Dec 21, 2011 at 6:47 PM, David Kopf <dak...@embarqmail.com> wrote:
Do your own scan for free slots before adding and after removing an entry. A stack overwrite would make that number change
abruptly.
-----Original Message-----
From: Mohammad Abdellatif
Sent: Wednesday, December 21, 2011 12:08 PM
To: Contiki developer mailing list
Subject: Re: [Contiki-developers] csma operation with full queue
after enabling the debug in queuebuf.. i can see that the queue gets full.. however csma still says that the queue is not full which
i cant seem to understand why.
Mohammad Abdellatif
Mohammad Abdellatif
Mohammad Abdellatif
as soon as the size reaches the max, csma start saying that it cant allocate the packet.
Mohammad Abdellatif
Mohammad Abdellatif
the node sends the packets from one queue only while the the other queue backs up until it is full which causes the problem..
in my case, each nodes r arranged in a row, what happens is that the neighbor queue in the direction of the sink is not full while the other one is "in the direction of the other nodes" , then this node cant forward the packets to the sink however it can still generate its own packets and reply to messages coming from the sink.
can any one tell me where is the function responsible for sending queued packets and how does it select which neighbor queue to send from first?
thaks,
Mohammad Abdellatif
Mohammad Abdellatif
in csma.c
the function free_first_packet
in the end after the main queue is finished and freed.. i added this in order to send any remaining packets in the other queue.
else {
/* This was the last packet in the queue, we free the neighbor */
ctimer_stop(&n->transmit_timer);
list_remove(neighbor_list, n);
memb_free(&neighbor_memb, n);
// my addition
struct neighbor_queue *n = list_head(neighbor_list);
if(n != NULL) {
transmit_packet_list(n);
}
//
}
so far the problem hasnt appeared, but i am still doing some tests.
will keep u posted
Mohammad Abdellatif
Simon Duquennoy
As there is one transmit_timer per neighbor queue, you shouldn't have to do this. In the code snippet you're showing, the timer of the current neighbor is stopped before freeing the neighbor structure. Other neighbor queues remain active, and their timer will eventually trigger.
To answer the question in your previous email, there is no centralized decision selecting the next neighbor to send to. Instead, each neighbor has its own ctimer implementing the csma backoff, and the transmissions are triggered independently.
Regards,
Simon
Mohammad Abdellatif
it is still did not solve the problem completely, i will try to do more work on it and keep u posted if i reach something solid.
thanks,
Mohammad Abdellatif
------------------------------------------------------------------------------
Ridiculously easy VDI. With Citrix VDI-in-a-Box, you don't need a complex
infrastructure or vast IT resources to deliver seamless, secure access to
virtual desktops. With this all-in-one solution, easily deploy virtual
desktops for less than the cost of PCs and save 60% on VDI infrastructure
costs. Try it free! http://p.sf.net/sfu/Citrix-VDIinabox
Mohammad Abdellatif
i am still having the same error
the csma freezes at some point "random" and so the queue gets full
it sends new generated packets however it doesnt forward the ones who are already in the queue and drops any new incoming packets.
i tried reducing the sending rate to 1 packet per minute and still this error sometimes happen.
not sure what to do more.. any ideas?
Mohammad Abdellatif