ARP Cache/Table

The Address Resolution Protocol or ARP is very important in data network communication. It is used to translate Layer 3 address (IP) to Layer 2 address (MAC). To do so, host or switch/router will create one ARP table which contains the mapping between IP address and MAC address.

Like MAC table, ARP table could be statically configured and dynamically learned. ARP protocol is used for learning. Basically, when the host wants to send IP packets to 192.168.100;  It will first query whether corresponding entry in ARP table. If found,  the corresponding MAC address will be the L2 address used. Or else,  the host will broadcast one ARP request saying: whose IP address is 192.168.1.100, and what's your MAC address? Every host on the same LAN will compare its IP address with the request; If they matches, it will respond with its own MAC address. Then the host will update its local ARP table for future usage.
To prevent ARP table becoming too big, ARP table entry will age out. The aging time could be configured in
modern operating system or switch/routers.

MAC Address

MAC address or Media Access control address is a 48-bit/6-byte address associate with a network interface card on a host or a network interface on a router/switch. Mostly, MAC address is permanent on the hardware. So sometimes it is called physical address or burn-in-address.
MAC address consists of 3-Byte of organization identifier and 3-byte of Network Interface Identifier. 

On switch, MAC address table is used to associate MAC address, VLAN to specific physical port so that Ethernet packet could be correctly forwarded by MAC address look-up.
MAC address table could be configured statically or learned dynamically. The entry will age out because of
MAC address table is a rare hardware resource and the number of entries are limited.
On host, MAC address is used to identify the specific physical network interface on the LAN network.
For further details, see MAC address wiki page.
To understand how MAC address works in switch, read this article.

Low-Latency

On eWeek.com,  the following news just published a couple of days ago:

Cisco, Arista Unveil Low-Latency Ethernet Networking Switches 

What's the latency? How to measure it?  Google and found a couple of very good resources to explain latency.  Just list them here for reference.

1. Understanding Switch Latency from Cisco Systems Inc
2. Cut-Through and Store-and-Forward Ethernet Switching for Low-Latency Environments from Cisco Systems Inc
3. It's time for low latency from Standford University

VN-Tag and VEPA

One good article about VN-Tag and VEPA is here:
 Also found a couple of resources on the Internet: 

1. One blog article  is here
2. VNTAG 101 slides from IEEE 802
3. Another good link - IO Virtulization Overview

Both are network interface virtulization technologies,
which makes the switch be aware of VMs on the host,
so that traffic could be correctly forward.

VN-Tag is implemented on Cisco Nexus Series
Data Center Switches; while VEPA is implemented
by data center switching solutions from other vendors
like HP.

802.1p vs 802.1Q

802.1p defines the class of service (Cos) in Ethernet MAC layer, and uses 802.1Q's 3-bit field to indicate 8 different traffic priority. The higher the value, the higher priority the packet has. When the packet is tagged with VLAN 0, it means it is 802.1p packets only.
802.1Q defines the virtual LAN tagging method in Ethernet MAC layer and is used to divide one physical network into multiple virtual LANs. It adds 4 bytes into original Ethernet frame. 2 bytes are used to indicate 802.1Q packets, which are  0x8100. The left 2 bytes consist of 3 fields: 3-bit Priority Code Point (PCP), which is defined by 802.1p; 1-bit Canonical Format Indicator (CFI), which indicate whether it is Ethernet or Token Ring; 12-bit VLAN identifier, which is used to indicate which VLAN the packet belongs to. VLAN 0 is used by 802.1p to deliver Cos priority only, while 4096 is reserved. The available VLAN range is 1 ~ 4095 .

Gb/s, GB/s and GT/s

In data communication or computer industry, Gb/s, GB/s or GT/s is used to describe the data transferring rate or bandwidth between endpoints. Gb/s is gigabits per second, GB/s is gigabytes per second, while GT/s is gigatransfers per second. The conversion between Gb/s and GB/s is simple: dividing Gb/s by 8 will get GB/s.
To convert GT/s to Gb/s, the encoding of the data has to be looked at. For example, PCIe 2.0, it mentions the interconnect bit rate is 5 GT/s. Considering that encoding scheme in PCIe is 8b/1b, the effective bit rate is 5 Gb/s * (8/10) = 4 Gb/s. Basically, GT/s is raw data rate, while Gb/s is effective data rate.

19 Inch Rack and Rack Unit (RU)

In telecommunication, computer network or audio/video industry, 19-inch rack chassis cabinet is widely used.19-inch means the width of the equipment front panel, and not the distance between holes.19 inch is about 48.26cm or less than half one meter. As to how tall the chassis is , it is measured in Rack Unit (RU).Generally, one chassis could hold multiple RUs. Here, 1 RU equals to1.75 inch, which is about 44.45 mm. 1.75 inch is the not height of front panel. The fact is the height of the front panel is 1/32 inch less --- about  (44.45 - 0.79) = 43.66 mm. Such design will give some spaces between RUs in the rack.

When buying chassis based switch/routers or servers, good to buy those standard chassis. When looking into
the data sheets or production information next time,  you know what those terminology means.

23 inch or 24 in rack also exist in industry, but not as popular as 19 inch rack chassis.

BPS (Bits per Second) verus PPS ( Packet per Second)

BPS is also written as b/s. Both BPS and PPS are used to describe the network bandwidth usage. The relationship between BPS and PPS depends on the packet size. However, the defacto packet size in industry is 64 bytes. To convert  PPS to BPS using the formula:  BPS = PPS * (64 * 8 ) bits

bps has Kilo bits per second (Kbps),  Mega bits per second (Mbps), Giga bits per second (Gbps) and Tera bits per second (Tbps); While pps has Kilo packets per second (Kpps), Maga packets per second (Mpps),  Billion packets per second (Bpps) .

The reason why we need pps is that switch/router's packet processing capability depends on packet size. That is, switch/router can't do wire speed for all packets size, but should be able to process most packet sizes at line rate.