[Planetlab-users] TCPDUMP response side traffic lost

[Sapan Bhatia]

Hi Hecker,

Greg Banks wrote a script that reproduces the case in which a raw
socket intermittently misses TCP payloads. I hope that this problem
will be fixed in the next RC, which Stephen is going to build and roll
out to the beta nodes today. I'll ask you to test your program again
on these freshly installed beta nodes.

Sapan




On Mon, Jul 21, 2008 at 2:09 PM, Hecker W <heckerme at gmail.com> wrote:
> Hi Sapan,
>
> Please let me know if you understand the problem well now. Or you need
> anything from us. Thanks.
>
> -Zhichun
>
> On Tue, Jul 15, 2008 at 5:07 PM, Hecker W <heckerme at gmail.com> wrote:
>> Thanks Sapan. I just saw your message today.
>>
>> The silence period I mentioned is not only for the TCP handshakes but
>> also for the payload packets.
>>
>> I did some tests on the machine you mention in another email
>> (plantlab-7 and echo). I have put the traces I produced by downloading
>> files in different size from my university online at
>> http://mot.cs.northwestern.edu/Planetlab/.
>>
>> You can take a look at the traces to see the problem.  I have tried
>> different files with different sizes on the planetlab-7 and echo. .
>> For the traces captured on echo.cs.princeton.edu, the incoming traffic
>> always have loses. For the
>> traces of planetlab-7, in some cases I can get full traces even with
>> SYN/ACK. But in other cases it repeated the behavior I mentioned. I
>> notice after the stabilization period, there are no noticeable packet
>> loss even when the downloading speed is high, so it seems not due to
>> the disk bottleneck.
>>
>> Here are the traces and corresponding file sizes
>> TRACE           FILESIZE
>> testfull*pcap   4572765
>> testsmall*pcap  25376
>> testtiny*pcap   2219
>>
>> For planetlab-7 node
>> testfull2.pcap, testsmall1.pcap have obvious packet loss at the
>> initial silence period.
>>
>> Thanks,
>> -Zhichun
>>
>>
>> On Mon, Jul 14, 2008 at 2:42 PM, Sapan Bhatia <sapanb at cs.princeton.edu> wrote:
>>>
>>> Hi Zhichun,
>>>
>>> I can't reproduce this problem. I tried some simple tests (wget a
>>> small file, wget a big file) but I could capture all incoming packets.
>>>
>>> Is the initial period of silence that you've encountered restricted to
>>> the TCP handshake? If that's the case, then it's expected - and you
>>> need to do something special to be able to see it. But you should not
>>> be missing any payload packets because of this.
>>>
>>> If not, then could you give me a test case to work with?
>>>
>>> Sapan
>>>
>>> On Thu, Jul 10, 2008 at 9:47 PM, Hecker W <heckerme at gmail.com> wrote:
>>> > I need to issue HTTP requests from Planetlab nodes and
>>> > to dump the whole TCP connection traces. I found it worked
>>> > on the FC4 version. But for the FC8 version, after the initial
>>> > SYN packet sent out, I cannot capture the server response
>>> > traffic until 0.6 second later. All the packets responded from
>>> > web servers are lost within this 0.6 second time window. After that
>>> > I can get all the traffic from both side.  I tried the big web objects to
>>> > make
>>> > sure its downloading time is longer than 1 seconds, and find this problem is
>>> > persistent.
>>> >
>>> > I am not sure whether it is due to the new version of Planetlab needs
>>> > 0.6 seconds to recognize that the new connection is issued so that it begin
>>> > to
>>> > allow TCPDUMP to see the packets?
>>> >
>>> > I guess it is not expected behavior, doesn't it? I did not use any raw
>>> > socket API.
>>> >
>>> > The problem is similar to the problem mentioned by Greg in the previous
>>> > post.
>>> >
>>> > If anyone needs the pcap trace to see the problem please reply the post. I
>>> > will make it available online.
>>> >
>>> > Thanks
>>> > -Zhichun
>>> >
>>> >
>>> >
>>> >
>>> > _______________________________________________
>>> > Users mailing list: Users at lists.planet-lab.org
>>> > https://lists.planet-lab.org/mailman/listinfo/users
>>> >
>>>
>>> _______________________________________________
>>> Users mailing list: Users at lists.planet-lab.org
>>> https://lists.planet-lab.org/mailman/listinfo/users
>>
>
> _______________________________________________
> Users mailing list: Users at lists.planet-lab.org
> https://lists.planet-lab.org/mailman/listinfo/users
>