Hardware Reference
In-Depth Information
receiving end, the reverse process takes place, with the link header and trailer
being stripped and the result being given to the transaction layer.
The concept of each layer adding additional information to the data as it works
its way down the protocol has been used for decades in the networking world with
great success. The big difference between a network and PCI Express is that in the
networking world the code in the various layers is nearly always software that is
part of the operating system. With PCI Express it is all part of the device hard-
ware.
PCI Express is a complicated subject. For more information see Mayhew and
Krishnan (2003) and Solari and Congdon (2005). It is also still evolving. In 2007,
PCIe 2.0 was released. It supports 500 MB/s per lane up to 32 lines, for a total
bandwidth of 16 GB/sec. Then came PCIe 3.0 in 2011, which changed the en-
coding from 8b/10b to 128b/130b and can run at 8 billion transactions/sec, double
PCIe 2.0.
The PCI bus and PCI Express are fine for attaching high-speed peripherals to a
computer, but they are too expensive for low-speed I/O devices such as keyboards
and mice. Historically, each standard I/O device was connected to the computer in
a special way, with some free ISA and PCI slots for adding new devices. Unfor-
tunately, this scheme has been fraught with problems from the beginning.
For example, each new I/O device often comes with its own ISA or PCI card.
The user is often responsible for setting switches and jumpers on the card and
making sure the settings do not conflict with other cards. Then the user must open
up the case, carefully insert the card, close the case, and reboot the computer. For
many users, this process is difficult and error prone. In addition, the number of
ISA and PCI slots is very limited (two or three typically). Plug-and-play cards
eliminate the jumper settings, but the user still has to open the computer to insert
the card and bus slots are still limited.
To deal with this problem, in 1993, representatives from seven companies
(Compaq, DEC, IBM, Intel, Microsoft, NEC, and Northern Telecom) got together
to design a better way to attach low-speed I/O devices to a computer. Since then,
hundreds of other companies have joined them. The resulting standard, officially
released in 1998, is called
USB
(
Universal Serial Bus
) and it is now widely imple-
mented in personal computers. It is described further in Anderson (1997) and Tan
(1997).
Some of the goals of the companies that originally conceived of the USB and
started the project were as follows:
1. Users must not have to set switches or jumpers on boards or devices.
2. Users must not have to open the case to install new I/O devices.
3. There should be only one kind of cable to connect all devices.
