CODEWITHSUNDEEP
network-programmingconcurrencyparallel-processingdistributed-computing
ag07
ag07

Reputation: 409

Why is there no global clock in distributed systems?

I can't understand why is there no Global time in distributed systems.

Upvotes: 2

Views: 6811

Answers (4)

user11844224
user11844224

Reputation:

There is no global clock because to have one would require systems to communicate time. This communication itself takes time. Since there is no global clock this time cannot be measured, and so it cannot be compensated for. NTP guesses by assuming the delays are the same in each direction and the temporal distribution of the delays are symmetrical. NTP also uses "pop-corn" filters, clustering, and has requirements for local clocks' free-run accuracy. NTP does a great job, but still, even NTP has a theoretical limits.

The closest thing to global clocks are synchronized atomic clocks. NTP'd systems typically are rooted to atomic clocks across the internet. GPS'ing directly to atomic clocks is obviously more accurate.

Upvotes: 1

RJN
RJN

Reputation: 736

The clocks are coordinated to keep them somewhat consistent but no one clock has the exact time. Even if the clocks were some what in sync, the individual clocks on each component may run at a different rate or granularity leading to them being out of sync only after one local clock cycle.

No global clock possible for DC

Upvotes: 1

user3666197
user3666197

Reputation: 1

Where did you get there is no global clock in all distributed systems?

First of all, let me state, I feel to be too old to know everything, but what I do know pretty well is all what I have touched in person.

I was building and operating globally distributed systems, that were the first globe-wide applied solution to the problem of the global clock, that caused countless headaches between continents before this came into the arena.

For ages the TELCO used Plesiosynchronous Digital Hierarchy in transporting digitised-voice / data in PDH-transport networks.

APAC, Eurozone and Americas did not have a common way how to connect the respective PDH-network infarstructures. These were almost-fine inside their disjunct domains, but were principally impossible to mutually synchronise their traffic on their inter-continental "edges". Well, up until the global synchronous networks started to get aligned to a newly introduced global clock synchronous framework -- call it SDH ( an ETSI evolved standard ) or SONET ( an ANSI adopted set of standards ), the core of this agreement was right in using pretty stable and mutually well aligned global clock for the whole globally distributed-system.

This simple idea has made the world of distributed nodes being suddenly able to cooperate atop the global synchronous transport network ( if too young, forget about packets, L4/5+ packet-mapped protocols. The TELCO world has much higher-level and way stringent requirements, than any "just"-Best-Effort delivery ( incl. any ex-post optimistic beliefs alike the ToS-labeled messages, VLAN-sub-L3 arrangements et al ) -- there are very strict TELCO standards, starting for the PHY-signal alignment ( both in TimeDOMAIN, FrequencyDOMAIN and AmplitudeDOMAINs ), jitter + wander constraints and many more -- all of that right for the core-reason -- to the support of the global, synchronous, carrier-grade signal delivery ( and for all of the digitised-voice / data services, smart payload-mapped onto this Synchronous Transfer Mode-1 and higher ( STM-1+ ) vehicle ).

Until SDH / SONET, there were un-avoidable circuit bit-slippages on PDH-network edges, where signal traffic-flows were handed over from Japan TELCO networks into the US PDH-network hierarchy, that used different MUX-hierarchies and not having the exactly same sync-edges ( yep, remember, this was the pre-SDH/SONET-global clock synchronicity ) simply caused cascaded bit-slippages, that the PHD-equipment was not able to mitigate and the digital circuit equipment simply had to consider these to be network-/circuit-errors and the Quality Standards supervision started yelling on not meeting the international circuits' BER / ES / SES / ITU-T M.xxxx quality thresholds... If anyone's father or grandfather was active in TELCO or a MIL/SIGINT/OPS guy, ask 'em about maintaining the SES under set thresholds on these international PDH-carrier service transports ( the worse during Cold-War Epoch, when almost everything was labeled in french -- be it as an espionage or a sabotage ).

Who had a chance to feel the change, once SDH/SONET transports came in, can tell you the levels of comfort the global clock synchronous carrier-network elements have created for transported circuits ( incl. the wrapped / padded / STM-1 encapsulated old&good PDH-circuits, which now, suddenly benefited from the very global clock, that could now salvage the incoherences in both the digital MUX-hierarchies and in their end-points local-clock both static and dynamic mis-alignment and/or jitter/wander artifacts ).

Actually there were two networks built -- global clock + global transport

needless to tell that these two were independent and designed & operated with indeed the best of the knowledge and technology available and having paid all the due care, the global clock sources and re-distribution topologies were maintained as the best possible synchronisation framework for the rest of the hi-tech distributed toys.

Upvotes: 3

Ton Plooij
Ton Plooij

Reputation: 2651

Because it's hard to implement. Synchronising physical clocks (e.g using NTP) is difficult due to unpredictable message delay. There are logical global clock solutions for distributed system, see for example Lamport timestamps.

Upvotes: 4

Related Questions