Distributed process and scheduling

Contents
• Distributed process
• Process Management
– Processor allocation
– Process migration
– Threads
13/04/2012
Distributed Systems, B.Tech.IV,
SVNIT,Surat
2

References:
• Pradeep K. Sinha “Distributed Operation
Systems : Concepts and Design” PHI.
• “Process Migration” Dejan S. Milojicic, Fred
Douglis, Yves Paindaveine, Richard Wheeler
and Songnian Zhou
13/04/2012
SVNIT,Surat
3

Concept of Process
• Process: An operating system abstraction
representing an instance of a running
computer program
– Consists of data, stack, register contents, and the
state specific to the underlying OS
– Can have one or more threads of control
• Consists of their own stack and register contents, but
share a process’s address space and signals.
13/04/2012
SVNIT,Surat
4

Process management
• Conventional OS:
– deals with the mechanisms and policies for
sharing the processor of the system among all
processes
• Distributed operating system:
– To make best possible use of the processing
resources of the entire system by sharing them
among all processes
13/04/2012
SVNIT,Surat
5

Process management contd…
• Three concepts to achieve this goal:
– Processor allocation
• Deals with the process of deciding which process
should be assigned to which processor
– Process migration
• Deals with the movement of a process from its current
location to the processor to which it has been assigned
– Threads
• Deals with fine-grained parallelism for better utilization
of the processing capability of the system
13/04/2012
SVNIT,Surat
6

Process Migration
• Process Migration:
– The act of transferring a process between two machines
during its execution
– Relocation of a process from its current location (the
source node) to another node(the destination node)
• Goals of Process Migration:
– Dynamic load distribution
– Fault resilience
– Improved system administration
– Data access locality
13/04/2012
SVNIT,Surat
7

Process Migration contd…
• Problems with Process Migration:
– Complexity of adding transparent migration
– Lack of compelling commercial argument for OS
venders to support process migration
13/04/2012
SVNIT,Surat
8

• The flow of execution of a migrating process:
13/04/2012
SVNIT,Surat
9
Source node destination node
Process P1 in
execution
Execution
suspended
Transfer of
control
Execution resumed
Process P1 in
execution
Time
Freezing
Time

• Two types:
– Preemptive process migration
• Process may be migrated during the course of its
execution
– Non preemptive process migration
• Process may be migrated before it starts executing on
its source node
13/04/2012
SVNIT,Surat
10

• Involves three steps:
– Selection of a process that should be migrated
– Selection of the destination node to which the
selected process should be migrated
– Actual transfer of the selected process to the
destination node
13/04/2012
SVNIT,Surat
11

Desirable features
• Transperency
– Object class level
– System call and interprocess communication level
• Minimal interference
– Can be done by minimizing freezing time
– Freezing time: a time for which the execution o the
process is stopped for transferring its information to the
destination node
• Minimal residual dependencies
– Migrated process should not continue to depend on its
previous node once it has started executing on new node
13/04/2012
SVNIT,Surat
12

Desirable features contd…
• Effiency
– Time required of migrating a process
– The cost of locating an object
– The cost of supporting remote execution once the process
is migrated
• Robustness
– The failure of a node other than the one on which a
process is currently running should not affect the
execution of that process
• Communication between Coprocesses of a job
13/04/2012
SVNIT,Surat
13

Process migration mechanisms
• Four major subactivities
– Freezing and restarting the process
– Transfer of process’s address space
– Forwarding messages meant for the migrant
process
– Handling communication between cooperating
processes
13/04/2012
SVNIT,Surat
14

• Mechanisms for freezing and restarting a
process
– Immediate and Delayed blocking of the process
• may be blocked immediately or delayed
• if the process is not executing a system call
• if the process is executing a system call but is sleeping
at an interruptible priority waiting for a kernel event to
occur, it can be immediately blocked from further
execution
• -Do- and sleeping at noninterruptible priority waiting
for a kernel event to occur, it can not be blocked
13/04/2012
SVNIT,Surat
15

• Fast and Slow I/O operation
– frozen after the completion of all fast I/O operations
– What about slow I/O operations???
• Information about the open files
– No problem for network transparent execution
environment
– What about UNIX like systems???
• creation of link
• Reconstruction of file’s path when required
– What about frequently used files like commands???
– What about temporary files?
13/04/2012
SVNIT,Surat
16

Mechanisms for freezing and
restarting a process contd…
• Reinstating the process on its destination
node
– Creation of a new process
– Process identifier
– What about the process which was blocked while
executing a slow system call????
13/04/2012
SVNIT,Surat
17

UPTO THIS FOR MIDSEMESTER
EXAM
13/04/2012
Distributed Systems, B. Tech I, SVNIT,
Surat
18

contd…
• Address Space Transfer Mechanisms
– Information to be transferred from source node to
destination node:
• Process’s state information
• Process’s address space
– Difference between the size of process’s state information
and address space
– Possible to transfer the address space without stopping its
execution
– Not possible to resume execution until the state
information is fully transferred
13/04/2012
SVNIT,Surat
19

Address Space Transfer
Mechanisms
• Three methods for address space transfer
– Total Freezing
– Pretransferring
– Transfer on reference
13/04/2012
SVNIT,Surat
20

Mechanisms (cont.)
• Total Freezing:
– Execution is stopped while its address space is
being transferred
– Simple and easy to implement
– Process is suspended for a long time during
migration
– Not suitable for interactive process
13/04/2012
SVNIT,Surat
21

Mechanisms(cont.)
• Total Freezing:
13/04/2012
SVNIT,Surat
22
Migration decision
made
Execution
suspended
Transfer of
address space
Execution resumed
Time
Freezing
Time

Mechanisms (cont.)
• Pretransferring (precopying):
– Address space is transferred while the process is still
running on the source node
– Initial transfer of the complete address space followed by
repeated transfers of the pages modified during previous
transfer
– Remaining modified pages are retransferred after the
process is frozen for transferring its state information
– Pretransfer operation is executed at a higher priority than
all other programs on the source node
13/04/2012
SVNIT,Surat
23

Pretransferring (precopying) (cont.)
– freezing time is reduced
– Total time of migration is increased due to the
possibility of redundant page transfers
13/04/2012
SVNIT,Surat
24

Pretransferring(cont.)
13/04/2012
SVNIT,Surat
25
Migration decision
made
Execution
suspended
Transfer of
address space
Execution resumed
Time
Freezing
Time

Mechanisms (cont.)
• Transfer on reference
– Based on the assumption that the process tends to use only a
relatively small part of their address space while executing.
– A page of the address space is transferred from its source node to
destination node only when referenced
– Demand-driven copy-on-reference approach
– Switching time is very short and independent of the size of the address
space
– Not efficient in terms of cost
– Imposes continued load on process’s source node
– Results in failure if source node fails or reboots
13/04/2012
SVNIT,Surat
26

Transfer-on-reference
13/04/2012
SVNIT,Surat
27
Migration decision
made
Execution
suspended
On demand
Transfer of
address space
Execution resumed
Time
Freezing
Time

(cont.)
• Message forwarding mechanisms
– Ensures that all pending, en-route and future
messages arrive at the process’s new location
– Classification of the messages to be forwarded:
• Type 1: Messages received at the source node after the
process’s execution has been stopped on its source
node and process’s execution has not yet been started
on its destination node
• Type 2: Message received at the source node after the
process’s execution has started on its destination node
13/04/2012
SVNIT,Surat
28

Message forwarding mechanisms
contd…
• Type 3: Messages that are to be sent to the migrant
process from any other node after it has started
executing on the destination node
13/04/2012
SVNIT,Surat
29

(cont.)
• Mechanism of Resending the Message
– Messages of type 1 and 2 are returned to the
sender as not deliverable or are simply dropped
– Locating a process is required upon the receipt of
the nonnegative reply (messages of type 3)
– Drawback: nontransparent to the processes
interacting with the migrant process
13/04/2012
SVNIT,Surat
30

(cont.)
• Origin Site Mechanism
– Process identifier has the process’s origin site(or home
node) embedded in it
– Each site is responsible for keeping information about the
current location of all the processes created on it
– Messages are sent to the origin site first and from there
they are forwarded to the current location
– Drawbacks:
• not good from reliability point of view
• continuous load on migrant process’s original site
13/04/2012
SVNIT,Surat
31

(cont.)
• Link Traversal mechanism:
– Uses message queue for storing messages of type 1
– Use of link (a forwarding address) for messages of type 2
and 3
– Link has two components: process identifier and last
known location of the process
– Migrated process is located by traversing a series of links
13/04/2012
SVNIT,Surat
32

(cont.)
• Link Traversal mechanism:
– Drawbacks:
• poor efficiency
• poor reliability
13/04/2012
SVNIT,Surat
33

(cont.)
• Link Update mechanism:
– Processes communicate via location independent
links
– During the transfer phase, the source node sends
link update message to all relevant kernels
13/04/2012
SVNIT,Surat
34

(cont.)
• Mechanisms for handling coprocesses
– Communication between a process and its
subprocesses
– Two different mechanisms
• Disallowing separation of Coprocesses
• home node or origin site concept
13/04/2012
SVNIT,Surat
35

Mechanisms for handling
coprocesses
• Disallowing separation of Coprocesses
– By disallowing the migration of processes that wait for
one or more of their children to complete.
– By ensuring that when a parent process migrates, its
children processes will be migrated along with it
• Concept of logical host
• Process id is structured as {logical host-id, local-index} pair
– Drawback :
• Does not allow parallelism within jobs
• Overhead is large when logical host contains several processes
13/04/2012
SVNIT,Surat
36

Mechanisms for handling
coprocesses
• home node or origin site concept
– Complete freedom of migrating a process or its
subprocesses independently and executing them
on different nodes
– Drawback:
• Message traffic and communication cost is significant
13/04/2012
SVNIT,Surat
37

Advantages of process migration
• Reducing average response time of processes
• Speeding up individual jobs
– Execute tasks of a job concurrently
– To migrate a job to a node having faster CPU
• Gaining higher throughput
– Using suitable load balancing policy
• Utilizing resources effectively
– Depending on the nature of the process, it can be
migrated to the most suitable node
13/04/2012
SVNIT,Surat
38

Advantages of process
migration(cont.)
• Reducing network traffic
– Migrate the process closer to the resources it is using most
heavily
– To migrate and cluster two or more processes which
frequently communicate with each other, on the same
node
• improving system reliability
– Migrating critical process to more reliable node
• Improving system security
– A sensitive process may be migrated and run on the secure
node
13/04/2012
SVNIT,Surat
39

Distributed Scheduling
13/04/2012
SVNIT,Surat
40

Contents
• Distributed Scheduler
• Motivation
• Issues in load distributing
• Load distributing algorithms
• Load sharing policies
13/04/2012
SVNIT,Surat
41

References:
• Mukesh Singhal, Niranjan Shivaratri.
“Advanced concepts in operating systems”.
Tata McGraw Hill publication
13/04/2012
SVNIT,Surat
42

Introduction
• Need for good resource allocation scheme for
DS
• Distributed scheduler:
A resource management component of a
distributed operating system that focuses on
judiciously and transparently redistributing the
load of the system among the computers such
that the overall performance of a system is
maximized.
• More suitable for LANs than WANs
13/04/2012
SVNIT,Surat
43

13/04/2012
SVNIT,Surat
44

Motivation
• Need for load distributing because of
– Random arrival of tasks
– Random CPU service requirements
• Needed both for heterogeneous and homogeneous systems
• E.g. system of N identical and independent servers
Let p be the utilization for each server
P0 = 1 – p; the probability that a server is idle
P be the probability that the system is in a state in which at least one
task is waiting for service and at least one server is idle
13/04/2012
SVNIT,Surat
45

Motivation contd…
13/04/2012
SVNIT,Surat
46

Motivation contd…
• Some Observations:
– For moderate system utilization the value of P is
high i.e. higher potential for load distribution
– At high system utilization the value of P is low i.e.
lower potential for load distribution
– At low system utilization the value of P is again
low
– As the number of the server in the system
increases, P remains high even at high system
utilization
13/04/2012
SVNIT,Surat
47

Issues in Load Distributing
• Some terminology
– Performance of a system
• one of the metric is average response time of task
which is the length of the time interval between its
origination and completion
– Defining proper load index
13/04/2012
SVNIT,Surat
48

Issues in Load Distributing contd…
• Load
– CPU queue length as a load indicator
– CPU utilization as a load indicator
13/04/2012
SVNIT,Surat
49

Classification of load distributing
algorithms
• Goal of Load distributing algorithm
– To transfer load from heavily loaded computers to
idle or lightly loaded computers
• broadly characterized as
– Static
• Decision is hard wired in the algorithm using a priory
knowledge of the system
– Dynamic
• Make use of system state information to make load
distributing decisions
13/04/2012
SVNIT,Surat
50

Classification of load distributing
algorithms contd…
– Adaptive
• Special class of dynamic algorithm
• they adapt their activities by dynamically changing the
parameters of the algorithm to suit the changing
system state
13/04/2012
SVNIT,Surat
51

Load balancing vs. Load sharing
• unshared state :
– A state in which one computer lies idle while at the same
time tasks contend for service at another computer
• to reduce the likelihood of unshared state
• Load balancing algo:
– Attempt to equalize the loads at all computers
– Higher overhead than load sharing algo
• anticipatory task transfer
– To reduce the duration of unshared state
13/04/2012
SVNIT,Surat
52

Preemptive vs. Nonpreemptive
transfers
• Preemptive:
– Transfer of a task that is partially executed
• Non preemptive:
– Transfer of a task that has not yet started
execution
13/04/2012
SVNIT,Surat
53

Components of Load Distributing
algorithm
• Four components
– Transfer policy
• Determines whether a node is in a suitable state to participate in a
task transfer
– Selection policy
• determines which task should be transferred
– Location policy
• determines to which node a task selected for transfer should be
sent
– Information policy
• responsible for triggering the collection of system state
information
13/04/2012
SVNIT,Surat
54

Transfer Policy
• Threshold policy
– Thresholds are expressed in terms of units of load
– Decided upon the origination of new task
– Concept of sender and receiver
• On detecting imbalance in load amongst
nodes in system
13/04/2012
SVNIT,Surat
55

Selection Policy
• Selects a task for transfer
• Simplest approach: to select the newly originated
task
• Overhead incurred in task transfer should be
compensated by the reduction in the response time
realized by the task
• Factors to consider:
– Overhead incurred by transfer should be minimal
– Number of location dependent system calls made by the
selected task should be minimal
13/04/2012
SVNIT,Surat
56

Location Policy
• To find suitable nodes to share load(sender or
receiver)
• Widely used method : polling
– Either serially or in parallel
– Either randomly or on a nearest-neighbor basis
• Alternative to polling
– Broadcast a query to find out if any node is
available for load sharing
13/04/2012
SVNIT,Surat
57

Information Policy
• To decide when, where and what information
about the states of other nodes on the system
should be collected
• One of three types:
– Demand driven
• Node collects the state of the other nodes only when it
becomes either a sender or a receiver
• dynamic policy
• can be sender-initiated, receiver-initiated or
symmetrically initiated
13/04/2012
SVNIT,Surat
58

Information Policy contd…
– Periodic
• Nodes exchange load information periodically
• Do not adapt their activity to the system state
• Benefits are minimal at high system loads ???
– State change driven
• Nodes disseminate state information whenever their
state changes by a certain degree
• Centralized and decentralized policy
13/04/2012
SVNIT,Surat
59

Load distributing algorithms
• Sender initiated algorithms
• Receiver initiated algorithms
• Symmetrically initiated algorithms
13/04/2012
SVNIT,Surat
60

Sender-Initiated algorithms
• Initiative by overloaded node (sender) to send a task
to an underloaded node(receiver)
• Transfer policy :
– Threshold policy based on CPU queue length
• Selection Policy:
– Consider only newly arrived tasks for transfer
• Location policy:
– Random :
• no remote state information
• task is transferred to a node selected at random
13/04/2012
SVNIT,Surat
61

– Random :
• Useless task transfer can occur
• Treating a transferred task
• Thrashing problem :
– Solution : limit the number of times a task can be transferred
• Substantial performance improvement over no load
sharing at all
13/04/2012
SVNIT,Surat
62

– Threshold
• Polling a node to determine whether it is receiver or
not
• PollLimit , limit on no. of nodes to poll
13/04/2012
SVNIT,Surat
63

13/04/2012
SVNIT,Surat
64

– Shortest
• Choose best receiver for a task
• Make use of CPU queue length
• Information policy
– Demand driven
• Stability
– Instability at high system load
13/04/2012
SVNIT,Surat
65

Receiver-Initiated Algorithms
• Initiation by an underloaded node (receiver)
• Transfer policy
– Threshold policy based on CPU queue length.
– Triggered when the task departs
• Selection policy
– Any
• Location policy
– Threshold policy
13/04/2012
SVNIT,Surat
66

(cont.)
13/04/2012
SVNIT,Surat
67

(cont.)
– Demand-driven type.
• Stability
– Do not cause system instability at high load. Why????
– Do not cause system instability at low load. Why????
• Drawback
– Most transfers are preemptive.
– What about sender-initiated algorithms????
13/04/2012
SVNIT,Surat
68

Comparison of Sender-Initiated and
• Stability
• Robustness
– Has an edge over the sender-initiated policies.
• Performs acceptably with a single value of threshold
over entire load spectrum while sender-initiated
policies requires adaptive location policy
13/04/2012
SVNIT,Surat
69

Comparison of Sender-Initiated and
Receiver-Initiated Algorithms contd…
13/04/2012
SVNIT,Surat
70

Symmetrically Initiated Algorithms
• Both senders and receivers search for
receivers and senders respectively
• Advantages and disadvantages of both sender
and receiver initiated algorithms.
• Above average algorithm.
13/04/2012
SVNIT,Surat
71

The above average algorithm
• Proposed by Krueger and Finkel
• Tries to maintain the load at each node within
an acceptable range of system average
• Why not exact system average ????
13/04/2012
SVNIT,Surat
72

Transfer Policy
• Use two adaptive thresholds:
– Equidistant from the node’s estimate of the average load
across all nodes
– E.g. average load is 2  the lower threshold = 1 and the
upper threshold = 3
• A node whose load is greater than upper threshold
 a sender
• A node whose load is less than lower threshold  a
receiver.
• Nodes that have loads between these thresholds lie
within the acceptable range, so they are neither
senders nor receivers.
13/04/2012
SVNIT,Surat
73

Location Policy
• The location policy has the following two
components:
– Sender-initiated component
– Receiver-initiated component
13/04/2012
SVNIT,Surat
74

Sender-initiated component
• Sender node:
– TooHigh message
– TooHigh timeout alarm
• Receiver node
– TooLow timeout alarm
– accept message
– AwaitingTask timeout alarm
– Increases load before accepting a task. Why????
• What if sender receives TooLow message
while waiting for Accept message??
13/04/2012
SVNIT,Surat
75

Sender-initiated component
contd…
• On expiration of TooHigh timeout, if no Accept
message is received,
– Sender infers that its estimate of the average
system load is too low
– Hence, it broadcasts a ChangeAverage message to
increase the average load estimate at the other
nodes.
13/04/2012
SVNIT,Surat
76

Receiver-initiated component
• A node, on becoming a receiver, broadcasts a TooLow
message, set a TooLow timeout, and starts listening
for a TooHigh message.
• If a TooHigh message is received, the receiver
performs the same actions that it does under
sender-initiated negotiation
• If the TooLow timeout expires before receiving any
TooHigh message, the receiver broadcasts a
ChangeAverage message to decrease the average
load estimate at the other nodes
13/04/2012
SVNIT,Surat
77

Selection and Information Policy
• Selection policy
– This algorithm can make use of any of the
approaches discussed earlier.
– Demand-driven.
13/04/2012
SVNIT,Surat
78

Symmetrically initiated algorithm
• average system load is determined individually
at each node
• load balancing actions adapts to the state of
the communication network as well
13/04/2012
SVNIT,Surat
79

Adaptive Algorithms
• A stable symmetrically initiated algorithm
• A stable sender initiated algorithm
13/04/2012
SVNIT,Surat
80

A stable symmetrically initiated
algorithm
• Instability in previous algorithms is due to
indiscriminate polling by sender’s negotiation
component.
• Utilize the information gathered during polling to
classify the nodes in the system as either
Sender/overloaded, Receiver/underloaded, or OK.
• The knowledge concerning the state of node is
maintained by a data structure at each node: a
sender list, a receiver list, and an OK list.
• Initially, each node assumes that every other node is
a receiver.
13/04/2012
SVNIT,Surat
81

Transfer policy
• A threshold policy where decisions are based on CPU
queue length.
• Trigger when a new task originates or when a task
departs.
• Two threshold values: a lower threshold (LT), an
upper threshold (UT).
• A node is said to be a sender if its queue length > UT,
a receiver if its queue length < LT, and OK if LT ≤
node’s queue length ≤ UT.
13/04/2012
SVNIT,Surat
82

Location policy
• Sender initiated component
• Receiver initiated component
13/04/2012
SVNIT,Surat
83

Sender initiated component
• Triggered when node becomes sender
• Sender polls a node at the head of the receiver lists to
determine whether it is receiver or not
• Processing at the polled node:
• Processing when the response arrives from the polled node:
• Polling stops if,
– A suitable receiver is found
– The number of polls reaches a PollLimit
– The receiver list at the sender node becomes empty
– And the task is processed locally
13/04/2012
SVNIT,Surat
84

Receiver initiated component
• Nodes polled are selected in following order,
– Head to tail in senders list
– Tail to head in OK list
– Tail to head in receivers list
• Receiver polls the selected node to determine whether it is
sender
• Processing if polled node is sender
• Processing if polled node is not a sender
• Polling process stops if,
– A sender is found
– If the receiver is no longer a receiver
– No. of polls reaches a PollLimit
13/04/2012
SVNIT,Surat
85

Selection and Information Policy
• Selection policy:
– The sender initiated component considers only
newly arrived tasks for transfer.
– The receiver initiated component can make use of
any of the approaches discussed earlier.
• Information policy: demand-driven.
13/04/2012
SVNIT,Surat
86

Discussion
• Future sender initiated polls at high system
loads are prevented. How???
• What about Receiver initiated component at
low system load ???
• Positive effect of updating the receiver list
13/04/2012
SVNIT,Surat
87

A stable sender initiated algorithm
• Two desirable properties:
– It does not cause instability
– Load sharing is due to non-preemptive transfer only.
• Uses the sender initiated load sharing component of
the stable symmetrically initiated algorithm
• Has a modified receiver initiated component to
attract the future non-preemptive task transfers from
sender nodes.
13/04/2012
SVNIT,Surat
88

• The data structure (at each node) of the stable
symmetrically initiated algorithm is augmented by a
array called statevector.
• The statevector is used by each node to keep track of
which list (senders, receivers, or OK) it belongs to at
all the other nodes in the system.
• When a sender polls a selected node, the sender’s
statevector is updated to reflect that the sender now
belongs the senders list at the selected node, the
polled node update its statevector based on the
reply it sent to the sender node to reflect which list it
will belong to at the sender
13/04/2012
SVNIT,Surat
89

• The receiver initiated component is replaced
by the following protocol:
– When a node becomes a receiver, it informs all the nodes that are are
misinformed about its current state. The misinformed node are those
nodes whose receivers lists do not contain the receiver’s ID.
– The statevector at the receiver is then updated to reflect that it now
belongs to the receivers list at all those nodes that were informed of
its current state.
– By this technique, this algorithm avoids the receivers sending
broadcast messages to inform other nodes that they are receivers.
• No preemptive transfers of partly executed
tasks here.
13/04/2012
SVNIT,Surat
90

Performance comparision
• Symmetrically initiated load sharing
• Stable load sharing algorithms
• Performance under heterogeneous workloads
13/04/2012
SVNIT,Surat
91

Symmetrically initiated load sharing
13/04/2012
SVNIT,Surat
92

Stable load sharing algorithms
13/04/2012
SVNIT,Surat
93

Selecting a suitable load sharing
algorithm
1. System under consideration never attains high loads
2. Systems that can reach high loads
3. Systems that experiences a wide range of load
fluctuations
4. Systems that experiences a wide range of
fluctuations in load and has a high cost of the
migration of partly executed tasks
5. Systems that experiences heterogeneous work
arrival
13/04/2012
SVNIT,Surat
94

Distributed process and scheduling

More Related Content

What's hot (20)

Viewers also liked (9)

Similar to Distributed process and scheduling (20)

Recently uploaded (20)

Distributed process and scheduling