Distributed Systems Exam 1 UPDATED
ACTUAL Exam Questions and CORRECT
Answers
Bus-based multiprocessors - CORRECT ANSWER - On a bus-based multiprocessor
system, all the processors, system memory, and peripherals are connected to a shared system bus.
Only one device can access the bus at any time.
Cache coherency problems
Snoopy cache - CORRECT ANSWER - The problem with using cached data is that if a
processor modifies a memory location, it will only modify the local copy in the cache.
Other processors will read the previous value. This is unacceptable but can be remedied by
having write operations pass through to the system bus so that main memory can be updated.
This is known as write-through.
Even if a processor updates the data in main memory, another processor may still access its own
cached data, which is now obsolete. This can be remedied by having each processor's cache
controller monitor write operations on the system bus and detect whether others are modifying
any cached addresses (Snooping)
Snoopy cache
This can be remedied by having each processor's cache controller monitor write operations on
the system bus and detect whether others are modifying any cached addresses. This operation is
known as snooping. Together, write-through and snooping comprise a snoopy cache
Scalability beyond bus-based multiprocessors: switched multiprocessors - CORRECT
ANSWER - Crossbar switch connectivity
,It allows the switching of any processor to any chunk of memory without interfering with the
traffic from any other processor accessing any other chunk of memory.
-Expensive
Also NUMA
NUMA
Processor affinity
Directory
Home snoop vs. source snoop, home agent - CORRECT ANSWER - NUMA - non-
uniform memory architecture
- Two-tiered memory access (local memory on each processor and shared memory)
-
Processor affinity-
OS tries to reschedule proccesses to rerun on same CPU for optimal local memory accesses
Home node/Agent- the processor that is responsible for an assigned region of memory
Directory - Each processing node maintains a table that indetifies which remote nodes are using
chunks of its local memory
Home snoop vs Source snoop
Home snoop- Gets latest version of data from last referenced caching agent (CPU that last
referenced version of memory of home agent's memory)
, Source snoop-
Reduces hops, by making the the requesting cpu send the request message to all processors and
ones with a cached version sends back its contents
Single system image - CORRECT ANSWER - One goal in some distributed systems
software is providing a single-system image. This is software that makes a collection of
independent computers appear as a single system to the users.
Transparency goals (very high-level understanding) - CORRECT ANSWER - Location
The user should not be aware of where the software is actually running or where resources
reside.
Migration
The user should not be aware of the fact that the location of resources may have moved from one
place to another or that a process was restarted on another computer, possibly in another data
center.
Replication
The user should not be aware that data might be replicated for fault tolerance or for proximity in
order to provide faster access.
Concurrency
The user should not be aware that multiple processes might be accessing resources at
approximately the same time. Results should appear as if all the processes ran one after another
in some order. This means that some processes might be temporarily locked from being able to
access a set of resources while another process is using them.
Service models - CORRECT ANSWER - Centralized
Client-server
ACTUAL Exam Questions and CORRECT
Answers
Bus-based multiprocessors - CORRECT ANSWER - On a bus-based multiprocessor
system, all the processors, system memory, and peripherals are connected to a shared system bus.
Only one device can access the bus at any time.
Cache coherency problems
Snoopy cache - CORRECT ANSWER - The problem with using cached data is that if a
processor modifies a memory location, it will only modify the local copy in the cache.
Other processors will read the previous value. This is unacceptable but can be remedied by
having write operations pass through to the system bus so that main memory can be updated.
This is known as write-through.
Even if a processor updates the data in main memory, another processor may still access its own
cached data, which is now obsolete. This can be remedied by having each processor's cache
controller monitor write operations on the system bus and detect whether others are modifying
any cached addresses (Snooping)
Snoopy cache
This can be remedied by having each processor's cache controller monitor write operations on
the system bus and detect whether others are modifying any cached addresses. This operation is
known as snooping. Together, write-through and snooping comprise a snoopy cache
Scalability beyond bus-based multiprocessors: switched multiprocessors - CORRECT
ANSWER - Crossbar switch connectivity
,It allows the switching of any processor to any chunk of memory without interfering with the
traffic from any other processor accessing any other chunk of memory.
-Expensive
Also NUMA
NUMA
Processor affinity
Directory
Home snoop vs. source snoop, home agent - CORRECT ANSWER - NUMA - non-
uniform memory architecture
- Two-tiered memory access (local memory on each processor and shared memory)
-
Processor affinity-
OS tries to reschedule proccesses to rerun on same CPU for optimal local memory accesses
Home node/Agent- the processor that is responsible for an assigned region of memory
Directory - Each processing node maintains a table that indetifies which remote nodes are using
chunks of its local memory
Home snoop vs Source snoop
Home snoop- Gets latest version of data from last referenced caching agent (CPU that last
referenced version of memory of home agent's memory)
, Source snoop-
Reduces hops, by making the the requesting cpu send the request message to all processors and
ones with a cached version sends back its contents
Single system image - CORRECT ANSWER - One goal in some distributed systems
software is providing a single-system image. This is software that makes a collection of
independent computers appear as a single system to the users.
Transparency goals (very high-level understanding) - CORRECT ANSWER - Location
The user should not be aware of where the software is actually running or where resources
reside.
Migration
The user should not be aware of the fact that the location of resources may have moved from one
place to another or that a process was restarted on another computer, possibly in another data
center.
Replication
The user should not be aware that data might be replicated for fault tolerance or for proximity in
order to provide faster access.
Concurrency
The user should not be aware that multiple processes might be accessing resources at
approximately the same time. Results should appear as if all the processes ran one after another
in some order. This means that some processes might be temporarily locked from being able to
access a set of resources while another process is using them.
Service models - CORRECT ANSWER - Centralized
Client-server
