Extremely tough competitions in all three key lines of business:
GBit, highspeed switch/router and SP ATM switches.
1. In GBit, AOL installed Foundry GBit switches, Compaq with Extreme switches won top honors and CSCO is no where to be seen.
2. In highspeed switch/router, Juniper M40 is gaining good grounds and there are another dozen companies with very good products. This is CSCO bread and butter with big margins and when this sector gets squeezed so will be CSCO overall margins.
3. In SP's ATM switche arena, FORE and ASND ATM swiches are much better than CSCO current ATM WAN swicthes.
So, as those Financial Houses said a few weeks ago, Competition is getting really tough and with that margins will come down. If CSCO is at LU's margin it will probably be trading at around half of current price. imo.
========================
February 15, 1999, Issue: 829
Section: CRN Test Center
Lab Analysis -- Measuring switching speed layer by layer
Edward J. Correia
All of the products in this roundup performed at wire speed in throughput tests
and experienced zero packet loss during Layer 2 switching capabilities testing.
No surprise there. The difference was in latency tests.
The lowest average latency-the time it takes to read the packet, determine the
designation, set up circuit and send the packet-was delivered by the SW5450
Enterprise Gigabit Ethernet Switch from Compaq Computer Corp. Built for
Compaq by Extreme Networks, the SW5450 clearly was the performance
winner, with an average latency of 19.3 microseconds.
And while the ProCurve 8000 from Hewlett-Packard Co. delivered the
lowest single-stream latency measurement of 4 microseconds, the product's
average switchwide latency was measured at around 92 microseconds,
placing it third overall. Put another way, the ProCurve's average latency was
nearly five times greater than the average latency of the first-place Compaq
switch.
Also impressive in L2 tests was the Lucent Cajun P550 switch. The
chassis-based unit turned in one of the lowest single-stream measurements of
9.5 microseconds, on par with the speedy Compaq unit's 8.9-microsecond
low. But in Gigabit Ethernet latency measurements, Lucent held a decided
edge.
The Cajun P550 delivered a roundup-leading 61-microsecond average
latency for Gigabit Ethernet. Average switchwide latency of all streams
combined was 60.9 microseconds, around three times longer than the leader,
but still respectably fast. The Cajun unit's longest latency was around twice
that of the Compaq at 183 microseconds, still good enough for second place
overall. Gigabit latencies were not charted.
Layer 3 tests told a different story. When routing IP packets between
VLANs, the Compaq and Lucent switches were worlds apart. The Compaq
SW5450 delivered an average latency of 29.3 microseconds, with its longest
latency reaching 61.1 microseconds. The fastest single-stream latency for the
switch was 11.5 microseconds. By contrast, the fastest single-port latency
exhibited by the Layer 2 media modules of the Lucent Cajun P550 was 23.9
microseconds, more than twice that of the Compaq unit, but still respectable.
Colossal differences were seen in maximum and average latencies between the
two products. When configured with Layer 2 media modules, which are its
most cost-effective setup, the Cajun P550 exhibited an average switchwide
latency of 4,269.9 microseconds per stream.
This performance lag was due to an oversubscription of the Layer 2 interface
to the backplane, which is limited to 1.5 million packets per second. Since the
Layer 2 media module was not capable of Layer 3 routing, packets had to be
forwarded to the management module for routing. This round-trip across the
backplane introduced latencies that would not be exhibited by a Layer 3
media module.
To prove this theory, engineers requested from Lucent two 12-port Layer 3
media modules and repeated the Routing VLAN test. Results were quite
different. While maximum latencies stayed about the same, the average latency
across the switch was 1,915.2 microseconds. And while this time still was
much slower than the Compaq unit, it represented an improvement of around
55 percent. The Cajun's minimum single-stream latency was 16 microseconds,
nearly 40 percent slower than the Compaq unit's minimum. Lucent's best times
appear in charts.
Though all reviewed products featured some degree of Layer 3 capability,
only units from Compaq and Lucent could route IP traffic across VLANs.
Using the Routing VLAN test, which is part of NetCom Systems' VLAN
Advanced Switch Test suite, data streams are routed by IP subnet from each
VLAN to every other VLAN within each switch. But unlike the Layer 2 tests,
which measure latency of a single delivered packet embedded with the
10-second test stream, the Latency Per Stream Histogram captured during
Layer 3 tests measured the latency exhibited by all the packets in each stream.
The high, low and average latency of all streams of each tested device are
shown in charts.
----
METHODOLOGY
Test Center engineers and the invited reseller from ComputerHelpers Inc.,
Farmingdale, N.Y., evaluated Gigabit/Fast Ethernet LAN switches with Layer
3 IP forwarding capability. Minimum port requirements called for 20 Fast and
two Gigabit Ethernet ports-chassis and fixed configuration models were
acceptable. Qualifying switches were capable of automatically sensing
10-Mbit-per-second or 100-Mbps Ethernet and could perform routing of
packets based on IP addresses on all Fast Ethernet ports.
Aside from gauging performance compared with the theoretical maximum wire
speed, engineers also evaluated switch features, flexibility of connection
options, ease of setup, and management and utility software, if included.
Layer 2
To prepare each switch for Layer 2 testing, engineers initiated a management
session through the console port of each unit under test (UUT). Spanning tree
and flow control were disabled to ensure maximum wire speed. Category 5
patch cables were run from each Fast Ethernet test card to consecutive ports
on the UUT, and multi-mode fiber-optic cables were used to patch the
Gigabit Ethernet test cards to the corresponding ports on the UUT.
Engineers measured throughput, packet loss and switching latency with the
help of a Smartbits tester and an additional expansion chassis from NetCom
Systems Inc. (www.netcomsystems.com). The Smartbits was configured with
20 card slots filled with NetCom ML-7710 10/100 Fast Ethernet test cards
with built-in RJ-45 transceivers. Two additional slots contained GX-1405B
Gigabit Ethernet test cards.
The Smartbits tester is accompanied by Smart Applications v.2.6, a suite of
benchmark utilities designed to stress switched network ports. Engineers
configured the Smart Applications to produce 10 continuous 10-second data
streams of Fast Ethernet traffic and one stream of Gigabit Ethernet traffic and
direct them into corresponding ports on the UUT at full duplex. The remaining
UUT ports receive the data streams and analyze the data and its path. Prior to
each test run, the software sends three "learning packets" to each port to allow
the switch to learn the MAC addresses of all its connected nodes.
Test packets varied in size from 64 Kbytes to 1,518 Kbytes each in a
doubling pattern (64 Kbytes, 128 Kbytes, 256 Kbytes, etc.). Fill pattern for
data packets was set to all zeros. The switch wire speed was compared with
the theoretical limit for Fast Ethernet (148,810 packets per second for
64-Kbyte packets) and results were analyzed as percentages of that total. All
Smartbits tests were repeated twice to ensure accuracy.
Layer 3
On devices with Layer 3 routing capability (units from Compaq Computer
Corp. and Lucent Technologies), engineers made use of another NetCom
utility, the VLAN Advanced Switch Test (VAST) version 2.1. This
benchmark measures switch latency when routing packets using Layer 3 IP
routing. To prepare each UUT, engineers created a VLAN (virtual LAN)
inside each switch, assigned each VLAN a Class A IP address and added
one of the UUT's ports to it, 20 in all. Each of these ports was connected to a
corresponding port on the Smartbits. Layer 3 testing involved only Fast
Ethernet ports.
Engineers employed the Routing VLAN test, one of several available in
VAST, and one of its most rigorous. In the Routing VLAN test, each port
sends IP data to all other selected ports, each of which is a Class A VLAN.
Therefore, IP traffic that passes from one VLAN to another must be routed
by the switch. The program reports test results in terms of throughput, packet
loss, flooding and latency by stream. Latency results are illustrated in the
charts (www.crn.com).
Copyright ® 1999 CMP Media Inc. |
