|
Previous
|
Content
|
Next
|
|
|
7.0.- ACK-compression |
|
|
|
|
In 1992, J.C.Mogul did some investigation observing TCP
dynamics in real networks specially about a phenomenom called
ACK-compression. Here is a summary of this work. |
|
The essence of the congestion avoidance algorithm is the observation that
data packets arrive at the receiving host at the rate that the bottleneck
link will support. If the receiver's ACK arrive at the sender with
the same spacing, then by sending new data packets at the same rate the
sender can avoid overruning the bottleneck link. It is by correctly
exploting this self-clocking property of TCP that congestion
may be avoided. |
|
Zhang, Shenker & Clark studied the slightly more complex case of a single
link with TCP data flowing in both directions at once. Their
simulations showed that several surprising phenomena could arise in such a
situation, even when Jacobson's algorithms were employed. |
|
|
|
On such phenomenom they called ACK-compression. A TCP sender's
self-clocking depends on the arrival of ACKs at the same
spacing with which the receiver generated them. If these ACKs spend
any time sitting in queues during their transit through the network,
however, their spacing may be altered. When ACKs arrive closer
together than they were sent, the sender might be misled into sending more
data than the network can accept, which could lead to congestion and loss of
efficiency. |
|
|
The Mogul's paper describes the result of a trace-based study
of large numbers of uncontrived connections through the Internet. These were
obtained by monitoring the packets flowing in and out of a busy gateway
system, widely used by sites all over the Internet for electronic mail and
similar protocols. The experiments show that ACK-compression
can indeed be detected automatically from packet traces, and that it does
happen in real networks. Other phenomena besides ACK-compression,
such as synchronization of losses between several connections,
out-of-order packet delivery, and some forms of improper TCP
behavior, can also be automatically detected from traces, as will be
described in this paper. Several simulation studies have identified other
phenomena that might be visible to a trace-based analisys. For
example, Shenker found that when multiple connections use the same path, the
packets from a given connection tend to be clustered together, rather
than interleaved with those of other connections. |
|
|
Previous
|
Content
|
Next
|