Owner's manual

HP ProCurve Switch 4000M / 8000M / 2424M / 1600M Reviewer’s Guide
Throughput testing using just the installed 24 10/100 ports of the HP ProCurve Switch 2424M shows it
to be media speed under these conditions for all packet sizes.
3.1.3 HP ProCurve Switch 1600M
For the HP ProCurve Switch 1600M, where the port count is lower than the Switch 8000M, throughput
is at full media speed for all ports. The throughput numbers in Appendix A for the HP ProCurve
Switch 1600M are for twenty-four 10/100Mbps ports at half-duplex. The test used to derive this result is
the SmartApplications suite, revision 1.7, from NetCom Systems. This version does not support full
duplex operation. In addition, the test will not test the backplane speed of the Switch 1600M with a
Gigabit port installed because it cannot operate at full duplex. Backplane throughput was tested with
the Gigabit port by using SmartWindows from NetCom Systems, another testing program that operates
at a low level much like a LAN analyzer. SmartWindows, which can handle full duplex traffic, was set
up to send theoretical maximum traffic into the Switch 1600M with the Gigabit module, while
monitoring traffic coming out of the switch. All traffic sent into the switch was observed to also leave
the switch indicating full media speed operation.
3.1.4 Gigabit Port Throughput
As shown in Appendix A, a Gigabit port, on an HP ProCurve switch tests out at ten times the
throughput of a 100 Mbps port. In other words, as far as the throughput tests are concerned, a Gigabit
module is equal to ten 10/100 ports, for both half and full-duplex tests.
3.2 Congestion Control
Congestion occurs any time more packets are destined for a particular segment than that segment can
handle. All switches handle momentary congestion passively through buffering excess packets in
memory. Some switches attempt to handle congestion actively by making a heavily used segment
appear to have numerous collisions, causing all stations on that segment to back off in their sending of
packets. This method is called “backpressure”. Since all nodes on a segment are prevented from talking
during the time the switch is applying backpressure, even traffic not destined for the switch is also
stopped during this time.
Another method of congestion control is to modify the Ethernet backoff algorithm parameters. This
technique, sometimes called “forwardpressure”, makes a switch port more aggressive in getting any
buffered packets out on a segment after a collision has occurred, at the expense of other nodes on that
segment. Since this directly conflicts with Ethernet standards it is strongly criticized and discouraged
by the various Ethernet standards groups. It is, however, very difficult to test for compliance in this
area. HP switches do not employ forwardpressure.
In order to control congestion in a standardized manner, the 802.3x Flow Control standard has been
implemented in the HP ProCurve Switch 4000M/8000M/2424M/1600M. This allows orderly control of
segment traffic by allowing a port on the switch to signal its congestion to the segment and holding off
segment traffic. For 802.3u to work correctly, the node(s) connected directly to switch ports with flow
control enabled also have to support the flow control protocol. Current congestion control testing has
not been done while using 802.3u, but should have similar results to switches employing backpressure.
3.2.1 Unidirectional Congestion Testing
As the HP ProCurve Switch 4000M/8000M/2424M/1600M does not utilize backpressure or
forwardpressure, congestion control is through having adequate buffers available to handle periods of
output port over-subscription or through the use of 802.3u Flow Control. In Section 6 of their 1997
Switched 10Mbps — 100Mbps Evaluation Report, Strategic Networks Consulting Inc., a well-respected
network consultancy, reported that switches with passive congestion control can be as effective as
switches with active congestion control. They further state that such switches “can be deployed
throughout the network with somewhat less regard for the traffic flows and client-server system
configurations.” In other words, deployment of switches with passive congestion control is easier to do
(less upset to the network itself) without any loss in effectiveness of overall congestion control. So
©1998, 1999, 2000 Hewlett-Packard Co Revision 3.2b – 1/15/2000 Page 25 of 36