Hubs vs. Switches vs. Routers -Layered Communication
Network communication models are generally organized into layers. The
OSI model specifically consists of seven layers, with each layer
representing a specific networking function. These functions are controlled
by protocols, which govern end-to-end communication between devices.
As data is passed from the user application down the virtual layers of the
OSI model, each of the lower layers adds a header (and sometimes a
trailer) containing protocol information specific to that layer. These headers
are called Protocol Data Units (PDUs), and the process of adding these
headers is referred to as encapsulation.
The PDU of each lower layer is identified with a unique term:
Commonly, network devices are identified by the OSI layer they operate at
(or, more specifically, what header or PDU the device processes).
For example, switches are generally identified as Layer-2 devices, as
switches process information stored in the Data-Link header of a frame
(such as MAC addresses in Ethernet). Similarly, routers are identified as
Layer-3 devices, as routers process logical addressing information in the
Network header of a packet (such as IP addresses).
However, the strict definitions of the terms switch and router have blurred
over time, which can result in confusion. For example, the term switch can
now refer to devices that operate at layers higher than Layer-2. This will be
explained in greater detail in this guide.
Icons for Network Devices
Hubs are Layer-1 devices that physically connect network devices together
for communication. Hubs can also be referred to as repeaters.
Hubs provide no intelligent forwarding whatsoever. Hubs are incapable of
processing either Layer-2 or Layer-3 information, and thus cannot make
decisions based on hardware or logical addressing.
Thus, hubs will always forward every frame out every port, excluding the
port originating the frame. Hubs do not differentiate between frame types,
and thus will always forward unicasts, multicasts, and broadcasts out every
port but the originating port.
Ethernet hubs operate at half-duplex, which allows a device to either
transmit or receive data, but not simultaneously. Ethernet utilizes Carrier
Sense Multiple Access with Collision Detect (CSMA/CD) to control
media access. Host devices monitor the physical link, and will only transmit
a frame if the link is idle.
However, if two devices transmit a frame simultaneously, a collision will
occur. If a collision is detected, the hub will discard the frames and signal
the host devices. Both devices will wait a random amount of time before
resending their respective frames.
Remember, if any two devices connected to a hub send a frame
simultaneously, a collision will occur. Thus, all ports on a hub belong to the
same collision domain. A collision domain is simply defined as any
physical segment where a collision can occur.
Multiple hubs that are uplinked together still all belong to one collision
domain. Increasing the number of host devices in a single collision domain
will increase the number of collisions, which can significantly degrade
Hubs also belong to only one broadcast domain – a hub will forward both
broadcasts and multicasts out every port but the originating port. A broadcast
domain is a logical segmentation of a network, dictating how far a broadcast
(or multicast) frame can propagate.
Only a Layer-3 device, such as a router, can separate broadcast domains.
Layer-2 devices build hardware address tables, which will contain the
following at a minimum:
• Hardware addresses for host devices
• The port each hardware address is associated with
Using this information, Layer-2 devices will make intelligent forwarding
decisions based on frame (Data-Link) headers. A frame can then be
forwarded out only the appropriate destination port, instead of all ports.
Layer-2 forwarding was originally referred to as bridging. Bridging is a
largely deprecated term (mostly for marketing purposes), and Layer-2
forwarding is now commonly referred to as switching.
There are some subtle technological differences between bridging and
switching. Switches usually have a higher port-density, and can perform
forwarding decisions at wire speed, due to specialized hardware circuits
called ASICs (Application-Specific Integrated Circuits). Otherwise,
bridges and switches are nearly identical in function.
Ethernet switches build MAC-address tables through a dynamic learning
process. A switch behaves much like a hub when first powered on. The
switch will flood every frame, including unicasts, out every port but the
The switch will then build the MAC-address table by examining the source
MAC address of each frame. Consider the following diagram:
When ComputerA sends a frame to ComputerB, the switch will add ComputerA’s
MAC address to its table, associating it with port fa0/10. However, the switch will not
learn ComputerB’s MAC address until ComputerB sends a frame to ComputerA, or
to another device connected to the switch. Switches always learn from the source
A switch is in a perpetual state of learning. However, as the MAC-address
table becomes populated, the flooding of frames will decrease, allowing the
switch to perform more efficient forwarding decisions.
While hubs were limited to half-duplex communication, switches can
operate in full duplex. Each individual port on a switch belongs to its own
collision domain. Thus, switches create more collision domains, which
results in fewer collisions.
Like hubs though, switches belong to only one broadcast domain. A Layer-
2 switch will forward both broadcasts and multicasts out every port but the
originating port. Only Layer-3 devices separate broadcast domains.
Because of this, Layer-2 switches are poorly suited for large, scalable
networks. The Layer-2 header provides no mechanism to differentiate one
network from another, only one host from another.
This poses significant difficulties. If only hardware addressing existed, all
devices would technically be on the same network. Modern internetworks
like the Internet could not exist, as it would be impossible to separate my
network from your network.
Imagine if the entire Internet existed purely as a Layer-2 switched
environment. Switches, as a rule, will forward a broadcast out every port.
Even with a conservative estimate of a billion devices on the Internet, the
resulting broadcast storms would be devastating. The Internet would simply
Both hubs and switches are susceptible to switching loops, which result in
destructive broadcast storms. Switches utilize the Spanning Tree Protocol
(STP) to maintain a loop-free environment. STP is covered in great detail in
Remember, there are three things that switches do that hubs do not:
• Hardware address learning
• Intelligent forwarding of frames
• Loop avoidance
Hubs are almost entirely deprecated – there is no advantage to using a hub
over a switch. At one time, switches were more expensive and introduced
more latency (due to processing overhead) than hubs, but this is no longer
Layer-2 Forwarding Methods
Switches support three methods of forwarding frames. Each method copies
all or part of the frame into memory, providing different levels of latency
and reliability. Latency is delay - less latency results in quicker forwarding.
The Store-and-Forward method copies the entire frame into memory, and
performs a Cycle Redundancy Check (CRC) to completely ensure the
integrity of the frame. However, this level of error-checking introduces the
highest latency of any of the switching methods.
The Cut-Through (Real Time) method copies only enough of a frame’s
header to determine its destination address. This is generally the first 6 bytes
following the preamble. This method allows frames to be transferred at wire
speed, and has the least latency of any of the three methods. No error
checking is attempted when using the cut-through method.
The Fragment-Free (Modified Cut-Through) method copies only the first
64 bytes of a frame for error-checking purposes. Most collisions or
corruption occur in the first 64 bytes of a frame. Fragment-Free represents a
compromise between reliability (store-and-forward) and speed (cut-through).
Layer-3 routing is the process of forwarding a packet from one network to
another network, based on the Network-layer header. Routers build routing
tables to perform forwarding decisions, which contain the following:
• The destination network and subnet mask
• The next hop router to get to the destination network
• Routing metrics and Administrative Distance
Note that Layer-3 forwarding is based on the destination network, and not
the destination host. It is possible to have host routes, but this is less
The routing table is concerned with two types of Layer-3 protocols:
• Routed protocols - assigns logical addressing to devices, and routes
packets between networks. Examples include IP and IPX.
• Routing protocols - dynamically builds the information in routing
tables. Examples include RIP, EIGRP, and OSPF.
Each individual interface on a router belongs to its own collision domain.
Thus, like switches, routers create more collision domains, which results in
Unlike Layer-2 switches, Layer-3 routers also separate broadcast domains.
As a rule, a router will never forward broadcasts from one network to
another network (unless, of course, you explicitly configure it to).
Routers will not forward multicasts either, unless configured to participate in
a multicast tree. Multicast is covered in great detail in another guide.
Traditionally, a router was required to copy each individual packet to its
buffers, and perform a route-table lookup. Each packet consumed CPU
cycles as it was forwarded by the router, resulting in latency. Thus, routing
was generally considered slower than switching.
It is now possible for routers to cache network-layer flows in hardware,
greatly reducing latency. This has blurred the line between routing and
switching, from both a technological and marketing standpoint. Caching
network flows is covered in greater detail shortly.
Collision vs. Broadcast Domain Example
Consider the above diagram. Remember that:
• Routers separate broadcast and collision domains.
• Switches separate collision domains.
• Hubs belong to only one collision domain.
• Switches and hubs both only belong to one broadcast domain.
In the above example, there are THREE broadcast domains, and EIGHT
VLANs – A Layer-2 or Layer-3 Function?
By default, a switch will forward both broadcasts and multicasts out every
port but the originating port.
However, a switch can be logically segmented into multiple broadcast
domains, using Virtual LANs (or VLANs). VLANs are covered in
extensive detail in another guide.
Each VLAN represents a unique broadcast domain:
• Traffic between devices within the same VLAN is switched
(forwarded at Layer-2).
• Traffic between devices in different VLANs requires a Layer-3
device to communicate.
Broadcasts from one VLAN will not be forwarded to another VLAN. This
separation provided by VLANs is not a Layer-3 function. VLAN tags are
inserted into the Layer-2 header.
Thus, a switch that supports VLANs is not necessarily a Layer-3 switch.
However, a purely Layer-2 switch cannot route between VLANs.
Remember, though VLANs provide separation for Layer-3 broadcast
domains, and are often associated with IP subnets, they are still a Layer-2