SUBJECT D11)

Questions about ATM efficiency

SUBJECT D11-1) Question:What is Goodput?

Answer: When ATM is used to transport cells originating from higher-level protocols (HLP), an important consideration is the impact of ATM cell loss on that protocol or at least the segmentation process. ATM cell loss can cause the effective throughput of some HLPs to be arbitrarily poor depending on ATM switch buffer size, HLP congestion control mechanisms, and packet size.

This occurs because during congestion for example, and ATM switch buffer can overflow which will cause cells to be dropped from multiple packets, ruining each such packet. The preceding and the remaining cells from such packets, which are ultimately discarded by the frame reassembly process in the receiver, are nevertheless transmitted on an already congested link, thus wasting valuable link bandwidth.

The traffic represented by these "bad" cells may be termed as BADPUT. Correspondingly, the effective throughput, as determined by those cells which are successfully recombined at the receiver, can be termed as GOODPUT.

One method of increasing the efficiency of ATM over AAL5 is to drop all remaining cells for a given packet if one of the cells is lost. This functionality is sometimes referred to as "early packet drop."


SUBJECT D11-2) Question:What is Partial/Early Packet Discard (PPD/EPD)

Answer: PPD stands for Partial Packet Discard and EPD stands for Early Packet Discard. These two are actually ATM cell discard techniques which maximize "goodput" by taking advantage of the notion that some types of ATM traffic are made up of large packets that are segmented into a series (or burst) of ATM/AAL5 cells. This notion holds true for classic IP over ATM and for LAN emulation (LANE).

These mechanisms work in concert with traffic policing. In a way they are cleaning up after QoS decisions have been made. If some cells which are part of a larger packet, are dropped for some reason, then why bother sending the other cells that were a part of the same fragmented packet since that entire packet will have to be retransmitted anyway. The act of discarding all other cells under this circumstances is called PPD. Now if all the cells that are the result of fragmenting a large packet will not fit into the available buffer space (and some will be dropped) then why continue sending only some of the cells. Just drop the entire packet (burst of cells), which is called EPD.

So EPD acts *before* cells belonging to an AAL5 frame are admitted to the output buffers. If a switch buffer occupancy threshold is exceeded, then frames are discarded by EPD without even being queued in the output buffers. On the other hand, PPD acts *after* cells of an AAL5 frame have been admitted to a buffer. If any one cell of a particular frame is discarded, then the rest of the cells are also discarded, since the frame is now errored and will require retransmission anyway.


SUBJECT D11-3) Question:PPD/EPD interaction with Traffic Policing?

Answer: One action of traffic policing is to CLP=1 mark (TAG) cells which exceed a VCs specific traffic parameters. As these cells traverse an ATM network they will be discarded IF congestion occurs at some place in the network. Implicitly this gives CLP=0 (not TAGed) cells priority in that the CLP=1 cells will be dropped first.

It is the result of traffic policing and the operation of CLP tagging that causes cells to be discarded, which can then trigger EPD/PPD. However it is also possiblt for policing to be doing the right thing and, for example, not tagging any cells, yet still output queues are congested and the need for EPD emerges.


SUBJECT D11-4) Question: How does ATM perform in transporting IP packets?

Answer: It is important to consider payload size when estimating IP efficiency.

  • Typical packet size on Internet is between 0 - 512 bytes
  • Almost 70% of TCP datagrams are approximately 40bytes(TCP ACKs)
  • Almost 80% of UDP datagrams are less than 48 bytes Unfortunately, the transmission efficiency of ATM networks in transporting typical Internet Traffic does not seem all that encouraging. This is primarily because ATM chops a variable size packet into fixed-size 48-bytes cells. Usually the last cell of the packet will be padded with dummy bytes (Cell padding) adding to inefficiency.

    The efficiency drops considerably as the payload size nears 48-bytes. The type of Physical link. (SONET/DS3 etc) does not seem to have any effect on the efficiency factor.

    Following table gives an indication of this for some of the commonly used datagram sizes.
     IP datagram Efficiency
     40 bytes  ~ 42%
     576 bytes  ~ 92%
     1500 bytes  ~ 97%
    Notes:
        - The effect of Physical layer is not considered in this calculation.
        - LLC encapsulation is assumed for the AAL5 packets.

    The efficiency is low for payload of less than 20 to 48 bytes

    For more info reference:

    1. IEEE paper "Efficiency of ATM networks". By Ramon Caceres.
    2. Book "Performance Issues for TCP over ATM". By M.Hassan & M.Atiquazzaman


    [ Back to Index | FAQ Index | Cell Relay Retreat ]

    Maintained by Cell Relay Retreat
    Last Changed 24 November 2002