[Planetlab-users] Re: Networking-related issues

[Sapan Bhatia]

Hi Zichun,

I'm going to post a message on devel at planet-lab.org describing the
situation with respect to this particular problem.

To cut a long story short, you'll have to do one of two things:
(i) Write a pcap filter that maintains a packet buffer and that sorts
the incoming packets in the right order.
(ii) Use the AF_INET/SOCK_RAW interface, in which case the packets
should arrive in the order in which they are supposed to.

Sapan

On Tue, Aug 12, 2008 at 1:08 PM, Hecker W <heckerme at gmail.com> wrote:
> Hi Sapan,
>
> Any updates on the tcpdump issue? Do you know when the problem can be solved?
>
> I recently figure out a way to "sort" the pcap trace, so that I can
> correct the packet reordering introduced by Planetlab. Therefore if
> you guys can deploy the current beta version (with packet reordering)
> to all the node will be helpful to me.
> NSDI deadline will be in about two months. It is best if we can have
> some working system soon.
>
> Thanks,
> -Zhichun
>
> On Mon, Jul 28, 2008 at 12:11 PM, Sapan Bhatia <sapanb at cs.princeton.edu> wrote:
>> Well, somebody else actually observed it on planetlab-7 -- which is a better
>> proof than several successful runs :)
>>
>> We'll keep you posted on any further developments.
>>
>> Sapan
>>
>> On Mon, Jul 28, 2008 at 3:07 PM, Hecker W <heckerme at gmail.com> wrote:
>>>
>>> I did my experiment in many runs. On planetlab-7 it always OK. However
>>> on alice the packet always reordered. It looks like not a matter of
>>> chance.
>>>
>>> For my analysis it is important I can get the CORRECT pcap traces, so
>>> that I can recover the files and other messages in the traces. My code
>>> does not work on these reordered traces. I searched online and  did
>>> not find any tool which can fix the reordering problem for the traces.
>>> If possible, please fix the problem. Thanks a lot!
>>>
>>> -Zhichun
>>>
>>> On Mon, Jul 28, 2008 at 11:50 AM, Sapan Bhatia <sapanb at cs.princeton.edu>
>>> wrote:
>>> > Hi Zichun,
>>> >
>>> > Somebody was experimenting with delta. Sorry about that. It has been
>>> > restored to a working state, but in case you still see problems, please
>>> > continue to use planetlab-7 and alice.
>>> >
>>> > planetlab-7 and alice have more-or-less the same software -- that you
>>> > didn't
>>> > notice the reordering issue on planetlab-7 is probably a matter of
>>> > chance.
>>> > We noticed the packet-reordering problem earlier today. How important is
>>> > it
>>> > for your analysis? Is it easy to work around, given that the packets are
>>> > timestamped?
>>> >
>>> > Sapan
>>> >
>>> >
>>> >
>>> > On Mon, Jul 28, 2008 at 2:40 PM, Hecker W <heckerme at gmail.com> wrote:
>>> >>
>>> >> Hi Sapan,
>>> >>
>>> >> I have tested my code on alice,delta and planetlab-7. I got totally
>>> >> different results on these three nodes!
>>> >> 1) For planetlab-7, the problem has been solved. I have been
>>> >> successfully run the TCPDUMP. I got the TCP SYN/ACK and all the
>>> >> payloads. I have done tcp reassembly and recovered the file I
>>> >> transfer. They are indeed same!
>>> >> 2) For alice, I think I might get all the payload packets. However,
>>> >> the TCP SYN/ACK packet comes AFTER the TCP ACK packet from the
>>> >> senders, although TCP SYN/ACK packet's timestamp is earlier. In other
>>> >> words, the tcpdump trace is not sorted by the timestamp which make my
>>> >> analysis tool fail.
>>> >> 3) For delta, I only get outgoing direction traffic (same for TCP and
>>> >> UDP).
>>> >>
>>> >> If you guys can make every node like planetlab-7, my problem is
>>> >> solved. Thanks a lot!
>>> >>
>>> >> If you guys need me to provide any trace or other information, please
>>> >> just let me know.
>>> >>
>>> >> -Zhichun
>>> >>
>>> >> On Sun, Jul 27, 2008 at 11:14 PM, Sapan Bhatia
>>> >> <sapanb at cs.princeton.edu>
>>> >> wrote:
>>> >> > Hello,
>>> >> >
>>> >> > The following issues have (hopefully) been resolved on a set of alpha
>>> >> > nodes (listed below). If you complained about any of them, then
>>> >> > please
>>> >> > kindly retry your code to verify the fixes.
>>> >> >
>>> >> > * Intermittent loss of TCP payload packets with packet sockets
>>> >> > * Syn/Ack packets not showing up in tcpdump
>>> >> > * TCP-related ICMP error messages not showing up in tcpdump.
>>> >> >
>>> >> > The nodes are: [alice,delta].cs.princeton.edu and
>>> >> > planetlab-7.cs.princeton.edu.
>>> >> >
>>> >> > Please verify all three nodes, if possible.
>>> >> >
>>> >> > Thanks in advance,
>>> >> > Sapan
>>> >> >
>>> >
>>> >
>>> >
>>> > --
>>> > Sapan Bhatia
>>> > www.cs.princeton.edu/~sapanb
>>> >
>>
>>
>>
>> --
>> Sapan Bhatia
>> www.cs.princeton.edu/~sapanb
>>
>



-- 
Sapan Bhatia
www.cs.princeton.edu/~sapanb