DISTRIBUTED SYSTEM

Distributed System

Distributed System

Why computer clock synchronisation is necessary in a distributed system

One area where clock inaccuracies can be particularly troublesome is in distributed computing. The clocks of different processors need to be synchronized to limit errors. Synchronization can be a particularly difficult problem. For example, ifthe clocks of collaborative video discussion participants' are to be synchronized to a desirable level, e.g., up to 30 ms, it is not enough to adjust local clocks only at the beginning of a collaborative video discussion session. The clocks need to bemonitored and adjusted continuously. Otherwise, the clocks will drift apart because of, for example, temperature drift, aging and the mechanical accuracy at which the frequency can be set by cutting. The drift rate is the deviation from the CoordinatedUniversal Time (UTC), divided by the time T, the measuring period. (Mills, 2005, 37-87)

One approach to distributed clock synchronization was reported by Flaviu in 1989 in his article "Probabilistic Clock Synchronization", which is the basis for many synchronization algorithms. It is called probabilistic, because in the approach itis not guaranteed that a client can always read the clock of a server with an a-priori defined precision. By means of several successive reading attempts, the client is able to synchronize onto the server's clock with an adjustable precision, and with aprobability of success which is also adjustable via a maximum number of allowed reading attempts. The disadvantages of this method include: the method does not take any processing time into account, and no dynamic adjustment to the current trafficsituation in a TCP/IP network is implemented. Another disadvantage of the method include to achieve a good precision, a large load may be placed on the network, caused by the synchronization messages. (Mills, 2005, 37-87)

Another clock synchronization method is used in the Network Time Protocol (NTP), which is an Internet Standard Recommended Protocol, described in the RFC-1119, RFC-1305 and RFC-2030. For the basic approach to measure the offset of anothercomputer over TCP/IP, NTP is uses substantially the same algorithm as presented in David L. Mills, RFC-2030 Simple Network Time Protocol (SNTP) Version 4 for IPv4, IPv6 and OSI, Network Working Group, October 1996. To save messages, and therefore reducethe network load caused by the update messages, an additional "logical clock" is implemented on the client side. This logical clock controls the local hardware clock, so that high precision can be realized, without many synchronization messages. Although this protocol works well, and high precision can be normally achieved (up to a few tens of milliseconds in global WANs), a major disadvantage of NTP is its slow speed. Based on the experience of the Rutgers University with NTP, it may take upto 30 minutes for the client's clock to synchronize for the first time on a time server. If, in addition, the difference between the two clocks is more than a few minutes, it can take much more time for synchronization to occur for the first time. (Mills, 2005, 37-87)

Distinguish between the purposes and uses of external and internal ...