[Ns-developers] tag rework for next release

[Tom Henderson]

Mathieu,

The below message revisits my previous comments, with a couple more.

Mathieu Lacage wrote:
> On Fri, 2009-04-03 at 11:04 -0700, Tom Henderson wrote:
> 
>> One thing that I noticed is that packet tags are left out of the 
>> Packet::Serialize() process (while byte tags are included).  This is a 
>> bit subtle, but it basically means that packets that are sent to another 
>> node via an ns-3 channel will retain packet tags but if that ns-3 
>> channel happens to utilize a real IP tunnel (in a distributed simulation 
>> use case) the packet tags will be stripped.
>>
>> It seems like we ought to support packet tags in Serialize, but is there 
>> a reason why not?
> 
> Lazyness: I can fix this late next week but feel free to beat me to it.

What is the status of serializing tags?  I see this in the code:

   // write tags
   //XXX
   //reserve = m_tags.GetSerializedSize ();
   //buffer.AddAtStart (reserve);
   //m_tags.Serialize (buffer.Begin (), reserve);

   // aggregate byte buffer, metadata, and tags
   Buffer tmp = m_buffer.CreateFullCopy ();
   tmp.AddAtEnd (buffer);


> 
>> We ought to document somewhere the interaction between application-level 
>> tags and NSC, while this is not yet fixed for NSC.
> 
> I doubt it will ever be really fixed in nsc.

this would be nice to document somewhere in NSC, tracker, and/or manual. 
   If a packet enters an nsc stack with tags, should some log message be 
raised?  Do you have a suggestion on how to document this?

> 
>> samples/main-packet-tag.cc could be enhanced with a byte example and 
>> more commenting.  I would volunteer to contribute this patch if you want.
> 
> please, do.

Since I suggested it previously, I will volunteer to do this and fix the 
manual before the release.

> 
>> Also, in udp-echo-server.cc:
>>
>>             NS_LOG_INFO ("Received " << packet->GetSize() << " bytes 
>> from " <<
>>               address.GetIpv4());
>>
>> -          packet->RemoveAllTags ();
>> +          packet->RemoveAllPacketTags ();
>>
>>             NS_LOG_LOGIC ("Echoing packet");
>>             socket->SendTo (packet, 0, from);
>>
>>
>> Why not remove all byte tags also?  Conceptually, this is a new packet 
>> that is being echoed; i.e., it should behave the same as if the packet 
>> were freed and a new one generated.
> 
> Yes, this makes sense.

the above still needs fixed.


also:

\packetperf Doxygen is stale; it would be nice also to include some 
brief comment about the performance implication of enabling metadata, 
and the two different types of tags

There is a distinction between packet and byte tags in terms of removal 
that could be more explicit in the doxygen.  Namely, packet tags can be 
removed one-by-one, while byte tags must all be removed.  So, maybe 
document that byte tags are assumed to have a lifetime of a packet and 
not removed until the packet is conceptually freed, while packet tags 
added somewhere should be removed by the consumer of the tag (such as a 
cross-layer tag internal to a node that should be stripped before 
sending on an ns-3 channel).

- Tom