CS 321 Spring 2013 > Lecture Notes for Monday, April 8, 2013 |
This is a way of preventing deadlock by doing acquisition in two phases. First, a process requests all resources in a non-blocking manner. If they are all acquired, they are used and released. If any resource is not acquired, then all are released, and we start over.
Lost messages (for example) can cause a kind of deadlock that is not related to resource allocation. Suppose process A is waiting for a message from process B; when it is received, A will reply. B sends the message to A, and then waits for the reply. Now, if the initial message is lost, then each process will wait for the other forever. We have deadlock, but there are no resources involved.
Like deadlock, but with busy-waiting instead of blocking. Processes are all running, but not accomplishing anything.
If resources are allocated preferentially to certain processes, and these processes continually make requests, then other processes may never get access.
Networking means automatically transferring data from one computer to another. This may not seem complicated, but consider:
Some terminology: A packet is a collection of data that travels together (or appears to) across a network. Sometimes we use the term segment to mean essentially the same thing.
We will use the term reliability. This has two meanings. First, it is the probability that some functionality will perform as desired. Second, it is the property of giving assurances of success. So functionality is reliable if it gives such assurances.
For example, the fundamental protocol of the Internet is IP. IP is not reliable. When a packet is sent via IP, there is no assurance that it is received. A protocol built on top of IP is TCP. TCP is reliable, despite the fact that it sends information via the unreliable IP protocol. TCP achieves this by allowing for acknowledgements of received segments, and retrying if no acknowledgement is received.
Note that the fact that IP is unreliable is not necessarily a bad thing. The decision was made to include reliability at the TCP level. Thus, IP does not need to be reliable; making it reliable would add needless complexity.
For example, as long as we can do IP over some network, we can do TCP on top of the IP, no matter whether the network is actually wires, optical fiber, or a radio link.
The OSI Model is never completely implemented, but it provides a useful framework for understanding networking. Here are the seven layers, from low to high.
Introduction to Networking will be continued next time.
ggchappell@alaska.edu