Distributed systems: overview

Distributed systems: overview

1. Definition

   • Several computers connected over a network
   • AIMs
     • Resource sharing
       • Costly resources (printers) can be shared
     • Computational speed up
       • Concurrency
     • Reliablity
       • Failure on one site doesn't mean whole system fails

2. Design Issues

   • Several layers of distribution
     • Shared main memory
     • Shared FS
     • Bus systems
   • Problem areas
     • Transparency: pretend to be one computer
     • Reliability: want availability - fault tolerance
     • Performance
     • Scalibility

3. Communication

   • Simple client/server model
     • Group of processes (servers) used by clients
       • Advantage of simple communication
       • Diasdvantages
         • Addressing (which address to give)
           • Broadcasting: only works in local networks
           • Name servers - ask special host (DNS)
         • Blocking (synchronous) vs non-blocking (reading from disk)
         • Reliable vs unreliable primitives
           • Network error

4. Remote procedure call

   • Simple idea: execute procedure on different host
     • Goal of transparency
     • Classic client/server interaction
   • Issues: transfer parameters and dealing with failures
     • Problem with parameter passing
       • Different representation on different machines (LE/BE)
       • Call-by-reference parameters
         • Can copy arrays but no arbitrary data structures (hard to tell when data ends)
     • Failure issues more complicated
       • Client cannot locate server (transparency lost)
       • Lost request messages
         • Use timers
       • Lost reply messages
         • Can execute operation more than once (sequence numbers)
       • Server crashes
         • Sequence numbers inefficient - when did crash occur?
       • Client crash
         • Leves orphans that you cannot get rid of them

5. Timestamp generation

   • Unique timestamps needed for co-ordination
     • Trivial in monoprocessor; use system clock (more than 2 clocks in distributed system)
   • Solution
     • Each host maintains logical counter (event counter)
       • All messages from host contain logical clock
         • When message with higher logical clock received, increases own logical clock to this number
           • Earilier events are lower numbers
           • When there are two identical timestamps, hosts decides
           • Problem that fastest running clock determines everything else

Imagens e arquivos do mapa