Taking into account the addressing method of the frames used
Store-and-Forward switches store each frame in a buffer before exchanging information to the output port. While the frame is in the buffer, the switch calculates the CRC and measures the frame size. If the CRC fails, or the size is too small or too large (an Ethernet frame is between 64 bytes and 1518 bytes) the frame is discarded. If everything is in order, it is directed to the port of departure.
This method ensures error-free operations and increases network confidence. But the time used to save and check each frame adds a significant delay time to their processing. The total delay is proportional to the size of the frames: the larger the frame, the longer this process takes.
Cut-through switches were designed to reduce this latency. These switches minimize the delay by reading only the first 6 bytes of data from the frame, which contains the destination MAC address, and immediately route it.
The problem with this type of switch is that it does not detect corrupt frames caused by collisions (known as runts), or CRC errors. The greater the number of collisions in the network, the greater the bandwidth it consumes when routing corrupted frames.
There is a second type of cut-through switch, the so-called fragment free, which was designed to eliminate this problem. The switch always reads the first 64 bytes of each frame, ensuring that it has at least the minimum size, and avoiding the routing of runts through the network.
These are the switches that process frames in adaptive mode and support both store-and-forward and cut-through. Either mode can be activated by the network administrator, or the switch can be intelligent enough to choose between the two methods, based on the number of errored frames passing through the ports.
When the number of corrupted frames reaches a certain level, the switch can switch from cut-through to store-and-forward mode, returning to the previous mode when the network normalizes.
Cut-through switches are most used in small workgroups and small departments. In these applications, a good volume of work or throughput is necessary, since potential network errors remain at the segment level, without impacting the corporate network.
store-and-forward switches are used in corporate networks, where error control is necessary.
Taking into account the way the subnets are segmented
They are traditional switches, which function as multi-port bridges. Its main purpose is to divide a LAN into multiple collision domains, or in the cases of ring networks, to segment the LAN into several rings. They base their sending decision on the destination MAC address contained in each frame.
Layer 2 switches enable multiple simultaneous transmissions without interfering with other subnetworks. Layer 2 switches cannot, however, filter broadcasts, multicasts (in the case where more than one subnet contains the stations belonging to the destination multicast group), or frames whose destination has not yet been included in the addressing table.
These are switches that, in addition to the traditional layer 2 functions, incorporate some routing functions, such as path determination based on network layer information (layer 3 of the OSI model), validation of the integrity of layer 3 wiring by checksum and support for traditional routing protocols (RIP, OSPF, etc.).
Layer 3 switches also support the definition of virtual networks (VLANs), and depending on models, they enable communication between the various VLANs without the need to use an external router.
By allowing the union of segments from different broadcast domains, layer 3 switches are particularly recommended for the segmentation of very large LAN networks, where the simple use of layer 2 switches would cause a loss of performance and efficiency of the ADSL, due to the excessive number of broadcasts.
It can be said that the typical implementation of a layer 3 switch is more scalable than a router, since the latter uses level 3 routing and level 2 routing techniques as complements, while switches superimpose the routing function on top of routing, applying the former where necessary.
Within the layer 3 switches we have:.
Basically, a packet by packet switch is a special case of a Store-and-Forward switch because, like this switch, it stores and examines the packet, calculating the CRC and decoding the network layer header to define its path through the adopted routing protocol.
A layer 3 Cut-Through switch (not to be confused with a Cut-Through switch), examines the first fields, determines the destination address (through the information in the layer 2 and 3 headers) and, from that moment on, establishes a point-to-point connection (at level 2) to achieve a high packet transfer rate.
Each manufacturer has its own design to enable correct identification of data flows. As an example, we have "IP Switching" from Ipsilon, "SecureFast Virtual Networking from Cabletron", "Fast IP" from 3Com.
Additionally, a Layer 3 Cut-Through switch, from the moment the point-to-point connection is established, will be able to operate in "Store-and-Forward" or "Cut-Through" mode.