Understanding the design of the Raspberry Pi
The Pi has two identifiable microchips on the PCB, which are described as follows:
- In the center, the one clearly marked as BCM2835 is the main processor
- Near the USB port, there is a smaller chip that is either a USB hub or USB/LAN chip, depending on the model
The BCM2835 is actually a high performance OpenGL ES GPU (VideoCore IV) with a built-in 700 MHz ARM6 processor by its side. It is a System on Chip (SoC), which means that there is a small amount of space for code that executes when it gets turned on. This is known as Stage 1 in the boot process.
Boot process
Some network actions need to be done during the boot process, and it is good to understand the various stages in case you need to troubleshoot something. The boot process is as follows:
- Stage 1 begins on the GPU and executes the code SoC firmware, which starts to load the Stage 2 code to the L2 cache.
- Stage 2 reads
bootcode.binfrom the SD card. It initializes Synchronous Dynamic Random Access Memory (SDRAM) and loads Stage 3. - Stage 3 is the
loader.binfile. This loadsstart.elf, which starts the GPU. - During
start.elf, it prepares to loadkernel.img. - Then, the kernel image reads
config.txt,cmdline.txt, andbcm2835.dtb. - If the
.dtbfile exists, it is loaded at 0 × 100 and the kernel is loaded at 0 × 8000 in memory. - The kernel image is the first binary that runs on the ARM CPU, and it can be compiled with the custom support for a specific hardware.
- The operating system starts to load.
All the source code in Stages 1 to 3 are closed source and protected by Broadcom. These closed source files are compiled and released by Broadcom only. You can update them on your SD card by running a firmware upgrade in Raspbian, which is covered later.
The kernel.img file connects the application to the hardware. Any computer with an operating system has a kernel of some sort. It is possible to compile your own kernel in Linux, and it might be the first file that you might want to amend yourself. This allows you to change the boot screen, load custom drivers, or perform other tasks that you might need. This is an advanced task and is not covered in this book.
Other capabilities
The BCM2835 processor also has dedicated audio hardware together with audio and video encoding/decoding. This allows the Pi to playback HD (MPEG-4) content such as videos or games. You can buy additional encoder/decoder licenses for extra functionality like MPEG-2 used in DVD video encoding and VC-1 that is used by Microsoft's WMV formats. The SD card is also directly interfaced by the Broadcom chip using the dedicated hardware inputs/outputs and interrupts.
All that dedicated hardware means that while some sections of the chip are fully utilized, the ARM CPU will be idle or hardly used. This will allow you to compute other transactions synchronously, and this is what makes the Raspberry Pi truly a unique, single board, credit-card-sized computer!
Hardware limitations
All this hardware crammed into one tiny space has its drawbacks. Some are deliberate and others are not. You should consider that these are theoretical calculations and that the real-world performance may vary. Usually, it is slower than the theoretical estimation.
It may be disappointing that the Raspberry Pi foundation decided to use a 100 MBps LAN chip instead of a gigabit one. But, we need to crunch some numbers to justify this decision. Let's convert megabits to more familiar megabytes. To get to "megabytes per second" from "megabits per second", we divide 100 MBps by 8 (there are 8 bits in a byte). This equates to 12.5 megabytes per second at 100 percent LAN capacity. For a single user, this is only roughly 20 percent of what the USB hub can handle. That means that, by design, this is an unchangeable bandwidth limitation for networking.
If you plan to share files with several users at the same time, each new user will bump down the other user's bandwidth to accommodate their own. As a work-around, you could add a gigabit USB LAN peripheral to increase the bandwidth. But, due to the speed constraints of the USB hub, you will only use approximately 48 percent of the gigabit LAN. To make matters worse, the hard drives running on the USB port will start to fight for bandwidth. The USB Controller has to share 480 MBps across all the four ports! One port is used by the 100 MBit network card, and the other connects the hub to the GPU. For one user, this means a maximum bandwidth of 240 MBps. Why 240 MBps? This is because 240 MBps goes to LAN and 240 MBps goes to the hard drive; and theoretically, there is no USB bandwidth left for anything else.
This can be a problem for a multiuser environment, but for home use, you would not run into any major problems as the bandwidth can accommodate HD video streams while serving other clients. This is why the cheaper 100 MBit version was used.
As it was made clear by the bottlenecks found with LAN, the worst thing about USB bottlenecks is that there is no way to work around this problem! This is because the USB Controller connects to the Broadcom chip and the LAN chip on the PCB, without any possibility of expanding or replacing this chip.
The Pi does not come with a real-time clock, so timekeeping is left to Internet-based time servers. For most people, this might not cause a problem. But if you want to create a remote, disconnected device that depends on events that are recorded at various times of the day, you might be left a little bewildered.
One easy and reliable way to do this is to connect a USB or I2C RTC that runs on a small battery. There is an easier and free option though, but it is not as accurate. You may want to install the fake-hwclock package. All you need to do is set the time once and the software will keep track of time by using a file. If you have a power outage, the software will read the file and set the time back to the last known time. The drawback is that you lose that time as there is no way to determine how long the outage lasted.