The MPLS WG Archive

[Enhui Liu <leh10814@huawei.com>]

Hi,

There is another requirement for MPLS header compression: MTU problem in MPLS L2 VPN.

In MPLS layer2 VPN model, customer premises' data link layer frames are transported 
 transparently on  virtual circuits.
CE1<-------------->PE1<------------->P<--------->PE2<------>CE2
 |                                                                                                        |
 |----------------------a   MPLS Virtual Circuit----------------------|

The frame format forwarded between PE1 and PE2  is the following.
+---------------------+-------------+----------------------+---------+------------+
 |data link layerA head |  mpls head   | data link layerB head  | IP head  | IP payload  |
+---------------------+-------------+----------------------+---------+------------+
|-------14 bytes------+---8 bytes---+---------------1518 bytes ------------------|

If  two data link layers are ethernet,  ethernet MTU(1518 bytes) + mpls head(8 bytes) 
+ ethernet head(14 byte) is greater than ethernet MTU(1518 bytes).  This equation 
is always valid if  data link layer has MTU limit. 
So it's a problem that PEs or CEs must deal with. 

There are three solutions to this MTU problem.
1.To request that all customers' IP packets are smaller than 1482 bytes, 
    though  the default is 1500bytes.
2.To split and recombine IP packets greater than MTU in PE.
3.To compress and decompress head information of MPLS payload in PE,  
    including data link layerB head, IP head, UDP/TCP head and RTP/RTCP head etc.

I think MPLS head compression is a good solution. compression algorithms can be 
acknowledged through RSVP or LDP  signaling.

Regards,   
Enhui Liu