Systems/Networks [223]

• Local-Area Networks (LANs):

• small geographical area

• multiaccess bus (Ethernet), ring, star

• cables: 1 megabyte/sec

• optical networks: 1 gigabit/sec

• one or more gateways to other networks

• Wide-Area Networks (WANs):

• large geographical area (Internet)

• relatively slow (1200 bits/sec to 1 megabyte/sec)

• unreliable

• telephone lines, microwave links, and satellite channels

• communication processors

• routers

DS: Naming [224]

• how does a process locate another host?

• domain name service (DNS)

• host name, identifier

• bob.cs.brown.edu

• request to name server edu for address of server for brown.edu

• edu must be known address

• request to edu.brown for address for cs.brown.edu

• request to cs.edu.brown for address for bob.cs.brown.edu

• implies Internet address 128.148.31.100

• relies on caches for better performance

• name server is in ``wait'' state:

daemon name_server() {              
  while (1) { 
    receive(&name,&pid); 
    send(pid,lookup(name));
  }
}

DS: Circuit Switching [225]

• how do two processes communicate with each other?

• session s is allocated transmission rate r_s bits/sec

• fixed path established between two sites

• each link guarantees portion r_s (say by time multiplexing)

• if path cannot be found, then reject session

• like telephone switching

• which has equal session transmission rates

• very inefficient for data networks

• short bursts of high activity

• lengthy inactive periods

DS: Store-and-Forward Switching [226]

• each session is initiated without reserved allocation

• no multiplexing of links - use full transmission rate

• links are fully utilized when there is traffic

• packets are queued - waiting for transmission

• decreases delay relative to circuit switching

• needs (feedback) control mechanism to reduce queueing delays

• message switching:

• store-and-forward with arbitrary message sizes

• must be broken into packets

• packet switching: store-and-forward

• virtual circuit routing: store-and-forward but on a fixed path

• uses full transmission rate of links

• dynamic routing: store-and-forward but packet finds its own path

DS: Contention [227]

• several sites want to transmit simultaneously (bus/ring)

• CSMA: carrier sense with multiple access (bus)

• site must listen for a free link

• if a collision occurs, try again after a random amount of time

• good for lightly-loaded systems

• token passing: a token continually circulates around a ring

• site can send messages only when it possesses the token

• uniform performance

• message slots: fixed-length message slots circulate in the ring

• each slot can hold a fixed-sized message and control

• site must wait until an empty slot arrives

• sites must check control info for possible messages

DS: ISO Network Model [228]

• use (7) layered approach to deal with complexity

• each layer ``talks'' with corresponding layer on another computer

DS: ISO Layers [229]

• physical: mechanical and electrial transmission of a bit stream

• data-link: transmission of frames or packets

• network: routing of packets

• transport: transmission of messages as packets, maintaining order

• session: implement sessions and protocols (say rlogin)

• presentation: resolve differences in format

• application: interact directly with the user

DS: TCP/IP [230]

• TCP/IP: Transmission Control Protocol/Internet Protocol

• fewer levels, more difficult, more efficient

• IP: transmission of datagrams, basic unit of information

• TCP: uses IP to transport a reliable stream between two processes

• establish and maintain a connection

• UDP: User Datagram Protocol

• uses IP to transfer packets, but adds error correction

• connection-less

DS: Distributed OS [231]

• users are unaware of the underlying structure or operations

• access remote resources in the same manner as local resources

• data migration: data is transferred to sites which require access

• transfer entire file or only those portions necessary

• partial computation migration:

• synchronous: remote procedure call (RPC) using a datagram

• asynchronous:

• message to start new process for designated task

• both return results

• process migration (full computation migration):

• load balancing to even the workload

• computation speedup via concurrency

• hardware or software preference

• data access

DS: Remote Procedure Call (RPC) [232]

result = multiply(7,2);                                              /*client*/  
             int multiply (x,y) {                                    /*stub  */
                 sprintf(str1,"%d %d",x,y);                           /*pack  */ 
                 general_transport(name_server("multiply"),str1,&str2);
                 sscanf(str2,"%d",&result);                          /*unpack*/
                 return(result);
             }
                          void general_transport (server,str1,str2) {       
                               send(server,str1);
                               receive(*str2,&server);               /*block */
                          }
                          /**************  NETWORK **************/
                          daemon multiply_transport () {            
                             while (1) {
                               receive(&str1,&client);               /*block */
                               multiply_server_stub(str1,&str2);  
                               send(client,str2);
                          }  }
             void multiply_server_stub (str1,str2) {                 /*stub  */
                  sscanf(str1,"%d %d",&x,&y);                        /*unpack*/
                  result = multiply(x,y);
                  sprintf(*str2,"%d",result);                        /*pack  */
             }
int multiply (x,y) { return(x*y); }                                  /*server*/

DS: Network Files System (NFS) [233]

• mount a remote directory

• tables of all mounts /etc/mtab also kept in kernel memory

DS: NFS - (Client and Server) [234]

• inode points to vnode