Berkeley’s Algorithm is a clock synchronization technique used in distributed systems. The algorithm assumes that each machine node in the network either doesn’t have an accurate time source or doesn’t possess an UTC server.
Algorithm
1) An individual node is chosen as the master node from a pool nodes in the network. This node is the main node in the network which acts as a master and rest of the nodes act as slaves. Master node is chosen using a election process/leader election algorithm.
2) Master node periodically pings slaves nodes and fetches clock time at them using Cristian’s algorithm.
Diagram below illustrates how the master sends request to slave nodes.
Diagram below illustrates how slave nodes send back time given by their system clock.
3) Master node calculates average time difference between all the clock times received and the clock time given by master’s system clock itself. This average time difference is added to the current time at master’s system clock and broadcasted over the network.
Psuedocode for above step:
# receiving time from all slave nodes
repeat_for_all_slaves:
time_at_slave_node = receive_time_at_slave()
# calculating time difference
time_difference = time_at_master_node - time_at_slave_node
# average time difference calculation
average_time_difference = sum(all_time_differences) / number_of_slaves
synchronized_time = current_master_time + average_time_difference
# broadcasting synchronized to whole network
broadcast_time_to_all_slaves(synchronized_time)
Diagram below illustrates the last step of Berkeley’s algorithm.
Scope of Improvement
- Improvision in accuracy of cristian’s algorithm.
- Ignoring significant outliers in calculation of average time difference
- In case master node fails/corrupts, a secondary leader must be ready/pre-chosen to take the place of the master node to reduce downtime caused due to master’s unavailability.
- Instead of sending the synchronized time, master broadcasts relative inverse time difference, which leads to decrease in latency induced by traversal time in the network while time of calculation at slave node.
The code below is a python script which can be used to trigger a master clock server.
# Python3 program imitating a clock server from functools import reduce from dateutil import parser import threading import datetime import socket import time # datastructure used to store client address and clock data client_data = {} ''' nested thread function used to receive clock time from a connected client ''' def startRecieveingClockTime(connector, address): while True : # recieve clock time clock_time_string = connector.recv( 1024 ).decode() clock_time = parser.parse(clock_time_string) clock_time_diff = datetime.datetime.now() - clock_time client_data[address] = { "clock_time" : clock_time, "time_difference" : clock_time_diff, "connector" : connector } print ( "Client Data updated with: " + str (address), end = "
" ) time.sleep( 5 ) ''' master thread function used to open portal for accepting clients over given port ''' def startConnecting(master_server): # fetch clock time at slaves / clients while True : # accepting a client / slave clock client master_slave_connector, addr = master_server.accept() slave_address = str (addr[ 0 ]) + ":" + str (addr[ 1 ]) print (slave_address + " got connected successfully" ) current_thread = threading.Thread( target = startRecieveingClockTime, args = (master_slave_connector, slave_address, )) current_thread.start() # subroutine function used to fetch average clock difference def getAverageClockDiff(): current_client_data = client_data.copy() time_difference_list = list (client[ 'time_difference' ] for client_addr, client in client_data.items()) sum_of_clock_difference = sum (time_difference_list, datetime.timedelta( 0 , 0 )) average_clock_difference = sum_of_clock_difference / len (client_data) return average_clock_difference ''' master sync thread function used to generate cycles of clock synchronization in the network ''' def synchronizeAllClocks(): while True : print ( "New synchroniztion cycle started." ) print ( "Number of clients to be synchronized: " + str ( len (client_data))) if len (client_data) > 0 : average_clock_difference = getAverageClockDiff() for client_addr, client in client_data.items(): try : synchronized_time = datetime.datetime.now() + average_clock_difference client[ 'connector' ].send( str ( synchronized_time).encode()) except Exception as e: print ( "Something went wrong while " + "sending synchronized time " + "through " + str (client_addr)) else : print ( "No client data." + " Synchronization not applicable." ) print ( "
" ) time.sleep( 5 ) # function used to initiate the Clock Server / Master Node def initiateClockServer(port = 8080 ): master_server = socket.socket() master_server.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1 ) print ( "Socket at master node created successfully
" ) master_server.bind(('', port)) # Start listening to requests master_server.listen( 10 ) print ( "Clock server started...
" ) # start making connections print ( "Starting to make connections...
" ) master_thread = threading.Thread( target = startConnecting, args = (master_server, )) master_thread.start() # start synchroniztion print ( "Starting synchronization parallely...
" ) sync_thread = threading.Thread( target = synchronizeAllClocks, args = ()) sync_thread.start() # Driver function if __name__ = = '__main__' : # Trigger the Clock Server initiateClockServer(port = 8080 ) |
Output:
New synchroniztion cycle started. Number of clients to be synchronized: 3 Client Data updated with: 127.0.0.1:57284 Client Data updated with: 127.0.0.1:57274 Client Data updated with: 127.0.0.1:57272
The code below is a python script which can be used to trigger a slave/client.
# Python3 program imitating a client process from timeit import default_timer as timer from dateutil import parser import threading import datetime import socket import time # client thread function used to send time at client side def startSendingTime(slave_client): while True : # provide server with clock time at the client slave_client.send( str ( datetime.datetime.now()).encode()) print ( "Recent time sent successfully" , end = "
" ) time.sleep( 5 ) # client thread function used to receive synchronized time def startReceivingTime(slave_client): while True : # receive data from the server Synchronized_time = parser.parse( slave_client.recv( 1024 ).decode()) print ( "Synchronized time at the client is: " + str (Synchronized_time), end = "
" ) # function used to Synchronize client process time def initiateSlaveClient(port = 8080 ): slave_client = socket.socket() # connect to the clock server on local computer slave_client.connect(( '127.0.0.1' , port)) # start sending time to server print ( "Starting to receive time from server
" ) send_time_thread = threading.Thread( target = startSendingTime, args = (slave_client, )) send_time_thread.start() # start recieving synchronized from server print ( "Starting to recieving " + "synchronized time from server
" ) receive_time_thread = threading.Thread( target = startReceivingTime, args = (slave_client, )) receive_time_thread.start() # Driver function if __name__ = = '__main__' : # initialize the Slave / Client initiateSlaveClient(port = 8080 ) |
Output:
Recent time sent successfully Synchronized time at the client is: 2018-11-23 18:49:31.166449
Below is a screenshot of runtime of above python scripts where top left console represents master thread while others represents slave threads.
Note: The scripts above closely depicts working of Berkley’s Algorithm but may differ from the actual implementation of the algorithm in production based distributed networking systems. Availability of port 8080 is machine dependent. In case port 8080 is not free, change the port number accordingly in both master and slave scripts.
References:
- https://en.wikipedia.org/wiki/Berkeley_algorithm
- https://tutorialspoint.dev/slugresolver/socket-programming-multi-threading-python/
- https://tutorialspoint.dev/slugresolver/cristians-algorithm/
leave a comment
0 Comments