3 latency – HP 2424M User Manual
Page 26
HP ProCurve Switch 4000M / 8000M / 2424M / 1600M Reviewer’s Guide
while backpressure is an appealing idea, in real-world environments it is not necessary and may
actually impede network traffic in times of high segment utilization.
In many respects this is true of 802.3u use as well. The one area where 802.3u can be beneficial and
simple to implement is on links between the switch and an end node, such as a server. In times of
network congestion this will allow the switch to hold off traffic originating on the server. The network
card in the server has to support 802.3u for this to work.
Long term a better solution for congestion control than either backpressure or Flow Control will be
implementation of QoS priority queues and policy-based networking, particularly in the network core.
3.2.2 Bi-directional Congestion Testing (X-Stream)
Another congestion test being used is the bi-directional, multi-stream spray test.
15
In this test bursts of
packet streams are sent into all tested ports at the same time. Collisions and congestion result
simultaneously on all ports. Throughput and packet loss rate (PLR) are recorded for each port.
This is an ardent stress test because the conditions created in this test would not be seen in a properly
designed network. Interpretation of the results of this type of test is not straightforward. It is important
to look at both throughput and PLR. Switches with passive congestion control may have a non-zero
PLR but throughput should be high. Switches that use backpressure will tend to have a 0% PLR but
throughput will tend to be lower than passive congestion control switches. Switches that have 0% PLR
and high throughput either have extensive packet buffers (an expensive proposition) or, more likely,
are using some form of forwardpressure and are not within Ethernet specifications.
With the results of this test keep in mind that switches that are less expensive will tend to have higher
PLRs and lower throughputs. This does not mean these switches are poorly designed, but rather that
they are designed for lower traffic level areas such as workgroup settings where cost is a primary
concern. For switches with active congestion control (both backpressure and forwardpressure), this
type of test doesn’t measure the possible negative effects on the rest of the network that occur in real
environments.
With the unrealistic conditions present in this test, interpretation is very difficult, particularly with
regards to expected performance in a real-world network.
3.3 Latency
Latency is the amount of time the switch takes to process a packet. In store-and-forward switches, such
as with the HP ProCurve Switches 4000M/8000M/2424M/1600M, latency is reported by vendors on a
LIFO basis (last bit in, first bit out), i.e. the time elapsed between the end of the last bit of the packet
going into the switch to the beginning of the first bit of that packet emerging from the switch. Cut-
through switch latency is measured on a FIFO basis (first in, first out), time measured from the
beginning of the first bit of the packet going in to the switch to the beginning of the first bit of the
outbound packet coming out of the switch. Latency figures should be used with care when comparing
store-and-forward switches versus cut-through switches.
To add to the confusion, many testing houses report store-and-forward latency as FIFO because the
test equipment reports it that way. Following the actual definition of latency (RFC1242), which is the
time taken by the switch to decide where to send a particular packet, store-and-forward switch
latencies should be reported as LIFO. At any rate, latency figures should be labeled as to whether they
are LIFO or FIFO.
Occasionally store-and-forward switch latency will be reported on a FIFO basis. To convert FIFO to
LIFO results, the time associated with the packet itself must be subtracted out. For 10Mbps Ethernet
frames the following formula can be used:
LIFO = FIFO - (packet length (bytes) * .8 µsec)
15
The SmartBits version of this test is the X-Stream test in the Advanced Switch Test suite. SmartBits (NetCom Systems) is a
network tester commonly used for switch testing.
©1998, 1999, 2000 Hewlett-Packard Co
Revision 3.2b – 1/15/2000
Page 26 of 36