| Fusun, attached is a recent competitive discussion about LANNET.
I would suggest that you get on the NET_PARTNER distribution list by
sending mail to META4::WEISS.
Thx, Stan
From: META4::NET_PARTNER "19-May-1995 1055" 19-MAY-1995 11:44:40.30
To: @NET_PARTNER
CC:
Subj: Can we guarantee zero packet loss with the HUB 900 !(From: BSS::RIGGEN "Sales Engineer or customer punching bag")
I am in a competitive situation with a Lannet hub and they have
a guarantee of zero packet loss between switched ethernets. I know we
received rave reviews from Scott Bradner(sp) with the DECswitch products
as well as store and forward switching rather than cut through. The customer
is sold on this guarantee and if I could state the same would help sell
about 18-20 HUB 900's.
Jeff
From: META4::NET_PARTNER "20-May-1995 0854" 20-MAY-1995 08:59:29.26
To: @NET_PARTNER
CC:
Subj: 0 packet loss discussion from Karl Pieper/thanks jeff riggen! !(From: PTOJJD::DANZAK "Pittsburgher �")
From: PTOVAX::BSS::RIGGEN "Sales Engineer or customer punching bag" 19-MAY-1995 17:18:51.36
To: ANGLIN::BERNDT, PTOVAX::DANZAK, USFHSL::LOPEZ
CC:
Subj: Packet loss answer from Karl Pieper
From: DELNI::PIEPER "Hub-based High Speed Networks & PE Program Manager" 19-MAY-1995 15:08:38.50
To: BSS::RIGGEN
CC: PIEPER
Subj: Packet Loss
NO switch can guarantee zero packet loss without some sort of "push-back"
kludge. If three ports are trying to get to a fourth port constantly, packets
will eventually get dropped when all buffer space is consumed or the switch
must somehow stop of slow down the three ports trying to send data to the
fourth (a "push-back scheme").
"Push-back" schemes may look like a good idea until you look at what impact
they have on LANs. If the switch in the following picture gets too busy, it
pushes back on ports coming into the switch by raising the Ethernet carrier
signal. This gives a "busy signal" (just like the telephone) to stations
connected to the switch ports. Remember that Ethernet is a CSMA/CD technology
where all stations LISTEN before attempting to transmit. This is not too big
an issue for Station A or Station B but is a major problem for LAN #1. Now
Stations #1 thru Stations #N-1 cannot get to their Server (Station #N) because
the Ethernet LAN is busy!
+---------------+
| |
| SWITCH |
| |
+-+-+-+-+-+-+-+-+
| | |
+---------+ | +---------+
| | |
+---+---+ +---+---+ +----+----+
|station| |station| |repeater |
| A | | B | | LAN #1 |
+-------+ +-------+ +-+-+-+-+-+
|.....|
|.....|
Stations 1 to N
So push-back schemes may be OK with an individual station per switch port but
push-back schemes work VERY POORLY with shared LANs. The switch with a push-
back scheme effectively slows the thruput of the shared LAN down to whatever
it needs (8 Mb, 6 Mb, 4 Mb, whatever) to clear its buffers. By the way it
also slows down the effective communication from Station A or Station B to
whatever it needs (8 Mb, 6 Mb, 4 Mb, whatever). This is how they can guarantee
no packet loss - slow down the thruput/speed of Ethernet!
LANnet has a push-back scheme and guarantees no packet loss but slows down
Ethernets connected to the switch to keep that guarantee. Our switches do
NOT support any push-back scheme (since we believe that CURRENTLY more folks
want to connect LANs to switches as opposed to individual stations - this will
change in future years), so all Ethernets will run at a full 10 Mb unconstrained
but packets could be lost by the switch (Don't worry per the Ethernet protocol,
the sending station will retransmit the "dropped" packet so it does not stay
lost) Make sure your customer really understands what he gets with the zero
packet loss guarantee - slower networks!
Karl
From: DELNI::KOPEC "NETWORK PRODUCTS BUSINESS GROUP 508-486-6404" 20-MAY-1995 22:12:59.50
To: META4::NET_PARTNER
CC: BSS::RIGGEN,KOPEC
Subj: RE: Can we guarantee zero packet loss with the HUB 900 !(From: BSS::RIGGEN "Sales Engineer or customer punching bag")
Hi Jeff,
LANNET *DID* in fact perform well in the Bradner Ethernet-Ethernet tests;
however here are a few points that they "forgot" to mention in their
attached article from the May 8 FIRST wire service.
First off, LANNET achieved 0% pkt loss through active congestion control (i.e.
backpressure) which is fine for Personal Ethernet users connected to the
switch but tends to disrupt LANs connected to the switch (by introducing
artificially created collision situations). In fact if they did not achieve
0% pkt loss then that would have been surprising since that is one of their
design goals. (Note that we do not use backpressure for the above reason,
we use very deep buffers on each port, often referred to as passive
congestion control).
LANNET's latency and thruput numbers were better than ours, but Bradner's lab
acknowledges that their "one pkt at a time" latency test was not real world,
and their thruput vs ours was 99.9% vs 98.5% of the theoretical maximum for
64-byte packets. So.... Everybody wins on throughput at a more real-life
128-byte pkt size and they inch by us in a non-real-life test.
(BTW, in this quarter's 10-100 tests, Scott agreed to perform a more real-life
latency test by transmitting a "packet train" through the switch, tagging a
packet somewhere in the middle and measuring its latency - this would of
course be performed many times to get a representative sample.)
Lastly, another part of Bradner's test (and subsequently SNC's analysis)
included MIB accuracy test. We comply
with all 10 key counters looked at, they comply with 6 out of 10. In the
area of MIB variable verification we had only 27 missing public MIB entities
(lowest was 22) and LANNET had 127 missing MIB entities (by far the
highest/worst). Anything less than 40 missing variables is classified as
very good, more than 40 missing variables may mean either they are missing
functionality or they implement private/propietary MIBs instead of the
public ones. In summary, DEC received a S+ (close adherence to stds and
significant extensions), while LANNET recieved a P for clearly propietary
MIB objects. Highest score was S+, lowest was a P.
There is more that can be dug into but these are the highlights that we can
bring up with as much bravado as LANNET did with their carefully worded
chest-beating article below, which came across one of those wire services on
May 8.
Jeff, call if you have questions on the above...
Good selling!
Stan Kopec
===========================================================================
SUBJECT: LANNET LANSWITCH WINS MAJOR PERFORMANCE TEST OF 18 SWITCHING HUBS
SOURCE: Business Wire via First! by Individual, Inc.
DATE: May 8, 1995
INDEX: [4]
---------------------------------------------------------------------------
IRVINE, Calif.--(BUSINESS WIRE) via First! -- LANNET Data Communications
Ltd., a leading supplier of scalable switching hubs for computer networks,
Monday announced that its MultiNet switching hub and LANswitch Ethernet
switching modules achieved top performance scores in a major test of 18
switching hubs, including products from 3Com, Bay Networks, Cabletron, and
DEC. It was the largest field of vendors yet tested in a published
switching hub comparison.
LANswitch easily swept the four performance tests in the evaluation,
coordinated by Strategic Networks Consulting of Rockland, Mass. with new
testing by Dr. Scott Bradner of the Harvard Internetworking Laboratory.
Results of the tests, which include benchmarks for switching hub
throughput, latency, packet loss, and handling of data congestion, were
made public in a press conference by Strategic Networks Consulting last
Friday. The results were also made available on the World Wide Web at
http://www.snci.com.
Strategic Networks Consulting's 160-page report on the test concluded,
"The LANNET LANswitch offers outstanding performance, scalability and a
wealth of value-added features for the network manager who is considering
widespread deployment of Ethernet switching to the desktop. High speed LAN
interfaces and the ATM-ready nature of the technology add to the
flexibility of the product to adapt to future requirements. With its
traffic priority features, the LANswitch can also more readily accommodate
multimedia applications than most other switches. The LANswitch has the
lowest unicast and broadcast latencies measured in the evaluation --
approximately 12 microseconds for packets of all sizes. The LANswitch
achieves full wire speed throughput at 100 percent of theoretical rates for
30 simultaneous Ethernet streams (the maximum that the test lab equipment
can accommodate). In all cases no packets were lost."
Avi Fogel, executive vice president of global marketing and business
development at LANNET, commented, "LANNET has proven again that it offers
the top-performing switching hub in the industry, and that's important to
users since increasing bandwidth is the number-one reason for buying
switches. No one provides more bandwidth or performance than LANNET."
Performance drives switching purchases
In a February 1995 study by the market research firm Infonetics Research,
an overwhelming 71 percent of network equipment buyers at 170 companies
responded that increasing bandwidth was the number one reason for adding
switching hubs to their networks. The next highest-scoring reason,
relieving network congestion, scored 22 percent response.
LANswitch extends its winning streak in high growth market
The win extends LANswitch's string of top performance results in the
fast-growing switching hub market to six victories, including editors
choice awards from Data Communications and LAN Times magazines. According
to the market research firm International Data Corporation (IDC), the
market for switching hubs is expected to grow from $327 million in 1994 to
approximately $1 billion in 1996. Port shipments in the market grew 492
percent and revenues jumped 262 percent from 1993 to 1994, IDC reported.
LANNET officials said they expect continuing highly-publicized performance
victories of LANswitch technology to fuel the company's sales in the
market.
LANNET scores highest in all four performance tests
Throughput
LANswitch demonstrated the highest throughput of any switch in the test,
handling 30 simultaneous 10 Mbps Ethernet streams with 100 percent
throughput. The throughput was several hundred percent higher than most
switches in the test and 25 percent higher than the next-ranking
competitor. LANswitch was the only switch to exceed "the maximum that the
lab equipment can accommodate," according to SNCI. The test lab ran out
of
equipment while utilizing only one-fourth of the total switching capacity
of LANswitch's 1.28 Gbps Cellenium Bus (128 Ethernet sessions).
Throughput
measures the total number of dedicated point-to-point data transfers a
switching hub can handle at full wire speed.
Latency
LANNET achieved the lowest packet forwarding latency of any switch in the
test -- approximately 11.5 microseconds. Latency measures the amount of
delay a switching hub adds when it forwards packets -- lower is better.
High latencies could cause a switching hub to interrupt network
applications. LANNET's latency was several hundred percent lower than most
of the other switches tested. LANNET also measured consistently lower
latency over all packet sizes, and maintained the lowest latency for
broadcast packets, which are common in real-world networks and seriously
degraded the performance of other switches in the test.
Packet Loss
LANNET demonstrated the best results in the packet loss evaluation,
showing zero percent packet loss rate while handling 30 simultaneous 10
Mbps streams of Ethernet. Again, 30 Ethernet streams was several hundred
percent more traffic than handled by most of the other switches in the
test.
Handling Data Overload
Only LANNET LANswitch demonstrated a sustained 100 percent theoretical
maximum throughput under 4:1 and 6:1 overload conditions. Fourteen
switches in the test demonstrated only finite numbers of packets they
could
handle before packet loss occurred under overload conditions. Packet loss
results in packet retransmissions, which occur much more slowly than
original packet transmissions and degrade the performance of real-world
networks. Not all vendors elected to participate in this test.
About LANNET
LANNET, with headquarters in Tel Aviv, Israel, designs, manufacturers,
and markets a complete line of intelligent switching hubs and platform-
based graphical network management software. For enterprise networks,
LANNET's award-winning MultiNet(r) fault-tolerant multi-slot concentrators
support multi-segment Ethernet, Token Ring, FDDI, and LocalTalk networks;
advanced high-speed switching capabilities; ATM; and integrated bridging
and routing, all managed within a platform-based UNIX application
environment. For small businesses and remote sites, LANNET offers easy-to-
use, cost-effective LANstack(tm) stackable hubs and Microsoft Windows(tm)-
based network management software. LANNET Data Communications Ltd.'s stock
is traded on NASDAQ, under the symbol LANTF. LANNET's Irvine, Calif.-based
U.S. operations may be reached by phone at 714/752-6638 or by fax at
714/752-6641.
|