Unlike a cut-through switch, which starts to transmit a frame before it has completely received it, a store-and-forward switch waits until it has received a whole packet into its buffer before forwarding it. By waiting to read the entire frame, a store-and-forward switch not only makes more involved routing decisions but can also filter out bad packets and shield destination LANs from corrupted or truncated frames. But there is a penalty for waiting to examine whole frames: long latency.
The store-and-forward switches tested have different configurations and expandability options. Thirteen of them are stackable, and the other 10 are chassis-based. Most stackable switches work with additional modules for switching between faster and slower wire speeds. OST's Xcellys-S is the only store-and-forward switch that offers a pluggable swi
tch module. This design would save your investment in the housing and power supply if your switch module fails.
The stackable switches range in expandability from Xedia's six-port MADswitch 10, which has one high-speed expansion slot, to 16-port switches based on the PlainTree WaveSwitch 100 (including the Fibronics FX8616 Switching Ethernet Hub, Proteon Series 80 p8900, and Xyplex SX-6601), with two high-speed expansion slots. The Alantec 12-port Power-Hub Model 3500 featured two FDDI (Fiber Distributed Data Interface) expansion slots.
Even some chassis-based switches provide expansion options that can satisfy many midrange and some high-end network installations. These range from Cabletron's ESX-MIM, with 12 ports in its maximum configuration, to the Optical Data Systems 12-slot chassis version of the 1094-16EF, supporting up to 176 ports (we tested the seven-slot model). The Lannet LET-36 MultiNet, a larger-chassis version of the LET-10 that we tested, provides up to 120 ports and 64,000 addr
ess tables per switch port. The LET-10 also provides an intelligent, high-performance management application and the most complete virtual LAN support of all the switches we tested.
Variations in the performance of store-and-forward switches are due in part to differing file I/O activities. In our high-level applications tests running under NetWare's standard IPX mode and under burst-mode IPX, average scores for the store-and-forward switches were slightly slower (within 15percent) than the performance of cut-through switches in our Excel for Windows 5.0 and Word for Windows 6.0 tests. However, store-and-forward switches were faster in the FoxPro for Windows 2.5 tests. Surprisingly, performance under FoxPro improves as latency increases, until a point of diminishing returns when latency becomes a hindrance.
This turnabout could have something to do with the timing characteristics or setup of FoxPro under NetWare, or maybe the lock-unlock requests that FoxPro makes to the file server have time-se
nsitive windows that simply appear to exhibit a bad showing under the 50-microsecond delay window. It could also occur if FoxPro bypasses the SPX of IPX/SPX. SPX performs checksums at the transport layer of the OSI (Open Systems Interconnection) model, and if errors aren't detected there, switches (on the lower network and data link levels) find the errors and retransmit packets over a longer connection.
Flow control (see "Putting the Brakes on Runaway Frames") also improves performance. If hubs can't keep up with demand, they waste time dropping and retransmitting packets.
Hardware-based back-pressure flow control is another high-performance feature, as exhibited by two cut-through switches (the NetWiz TurboSwitch-2000 and Ornet's LAN-booster 2000) and also by two store-and-forward switches (Lannet's LET-10 and Nbase's NH208 MegaSwitch). Interestingly, the Xedia MADswitch 10, with software-controlled back pressure, posted one of the highest latency scores.
If your network has a complex
infrastructure with high demand for management support, expanded virtual LAN support, sophisticated filtering, and increased expansion options, you should be looking for a store-and-forward switching hub. But if performance happens to be your major concern, a cut-through switch with flow-control features is the way to go.
WEIGHTING FOR BEST OVERALL
PERFORMANCE
75%
FEATURES
15%
USABILITY
10%
IT DOES IT ALL AND DOES IT WELL
BEST OVERALL:
Lannet LET-10 MultiNet Switching Hub and LANswitch Modules
If cost is no object, one of the MultiNet series (the LET-20 and
LET-36 are similar to the LET-10, just larger) is the switch for you.
The MultiNet switch garners top billing by combining excellent
performance; a wealth of expansion options that support Fast
Ethernet, token ring, and FDDI (Fiber Distributed Data Interface) in
the future; a
nd ease of use through excellent management. It shines
in both the low-level and applications performance tests, thanks to
some impressive latency figures (for a store-and-forward switch) and
a highly effective back-pressure implementation that yields zero
packet loss.
PRICE PER OVERALL
PORT TEST EVALUATION
VENDOR/MODEL CONFIGURATION SCORE
=======================================================================
BEST Lannet Data LET-10 MultiNet $283 9.45
RUNNER-UP NBase Switch NH208 MegaSwitch $615 8.59
RUNNER-UP OST Xcellys-M $832 8.30
RUNNER-UP OST Xcellys-S $574 8.31
VENDOR/MODEL PERFORMANCE FEATURES USABILITY
=============================================================================
BEST Lannet Data LET-1
0 MultiNet ***** **** *****
RUNNER-UP NBase Switch NH208 MegaSwitch **** *** ***
RUNNER-UP OST Xcellys-M **** *** ****
RUNNER-UP OST Xcellys-S **** *** *****
BUFFER MAX. MAX.
SIZE PORTS MAC SWITCH-
PER PER ADDRESS/ ING
VENDOR/MODEL PORT SWITCH PORT RATE
===============================================================================
BEST Lannet Data LET-10 MultiNet 16 KB 120 64,000 2 million
RUNNER-UP NBase Switch NH208 MegaSwitch 24 KB 8 2048 14,881
RUNNER-UP OST Xcellys-M # 24 4048 240,000
RUNNER-UP OST Xcellys-S # 6 1024 60,000
# Shared-RAM architecture.
KEY
Ratings from 1 to 5: * is the lowest; ***** is the highest.
FOR SMALL, STABLE WORKGROUPS
BEST STACKABLE:
Nbase NH208 MegaSwitch
The competitively priced NH208 MegaSwitch targets small workgroups
that don't need expansion options and multiprotocol LAN support. The
eight-port switch doesn't excel in the frills department, but it has
the performance network administrators need to break through
bandwidth barriers.
PRICE PER OVERALL
PORT TEST EVALUATION
VENDOR/MODEL CONFIGURATION SCORE
======================================================================
BEST NBase Switch NH208 MegaSwitch $615 8.59
RUNNER-UP OST Xcellys-S $574 8.31
RUNNER-UP 3Com SuperStack LinkSwitch $1165 8.21
VENDOR/MODEL
PERFORMANCE FEATURES USABILITY
===========================================================================
BEST NBase Switch NH208 MegaSwitch **** *** ***
RUNNER-UP OST Xcellys-S **** *** *****
RUNNER-UP 3Com SuperStack LinkSwitch **** ** ***
BUFFER MAX. MAX.
SIZE PORTS MAC SWITCH-
PER PER ADDRESS/ ING
VENDOR/MODEL PORT SWITCH PORT RATE
=============================================================================
BEST NBase Switch NH208 MegaSwitch 24 KB 8 2048 14,881
RUNNER-UP OST Xcellys-S # 6 1024 60,000
RUNNER-UP 3Com SuperStack LinkSwitch 128 KB 6 169 89,280
# Shared-RAM architecture.
KEY
Ratings from 1 t
o 5: * is the lowest; ***** is the highest.
FOR LARGE LANS THAT KEEP GROWING
BEST RACKMOUNTABLE:
Lannet LET-10/20/36 MultiNet
In addition to excellent performance, the Lannet MultiNet switching
hubs provide unmatched modular network expansion. A single hub can
house up to 18 full-height or 36 half-height modules and support up
to 120 Ethernet ports. A 1.28-Gbps internal bus can handle
bandwidth-intensive transmissions, such as voice and video.
PRICE PER OVERALL
PORT TEST EVALUATION
VENDOR/MODEL CONFIGURATION SCORE
======================================================================
BEST Lannet Data LET-10 MultiNet $283 9.45
RUNNER-UP OST Xcellys-M $832 8.30
RUNNER-UP Xyplex SX-6001 $600 8.21
VENDOR/MODEL PERFORMANCE FEATURES USABILITY
===========================================================================
BEST Lannet Data LET-10 MultiNet ***** **** *****
RUNNER-UP OST Xcellys-M **** *** ****
RUNNER-UP Xyplex SX-6001 **** **** ****
BUFFER MAX. MAX.
SIZE PORTS MAC SWITCH-
PER PER ADDRESS/ ING
VENDOR/MODEL PORT SWITCH PORT RATE
==============================================================================
BEST Lannet Data LET-10 MultiNet 16 KB 120 64,000 2 million
RUNNER-UP OST Xcellys-M # 24 4048 240,000
RUNNER-UP Xyplex SX-6001 256 KB 16 1000 238,000
# Shared-RAM architecture.
KEY
Ratings from 1 to 5: * is the lowest; ***** is the highest.
photo_link (7 Kbytes)

photo_link (5 Kbytes)

photo_link (7 Kbytes)
