It's quite interesting - this isn't ethernet as we know it. Instead of each NIC using its own free-running clock, all the physical layers are sync'ed to each other at layer 1. (note that gigabit ethernet, which is what it uses, sends data at all times - when idle it sends the idle symbol)
Haven't looked into this in depth but sub-nanosecond sync for systems up to 10km apart is interesting since 10km is about 33 light microseconds. There is some trickery going on.
nuccy 14 hours ago [-]
In our lab tests phase lock jitter between WR client and master is about 10ps (picoseconds) over 50km fiber (with temperature change of the fiber, so WR actively compensates elongations), so relative clock of one system can be transmitted with about that accuracy to another.
P.S. There is WR workshop this week with some talks being publicly available on CERN's indico website.
xxs 12 hours ago [-]
Even though you're commenting on While Rabbit post, it took some time to understand "WR" is white rabbit, esp. since describing the pico seconds in brackets.
elromulous 18 hours ago [-]
It's totally possible to achieve synchronization better than light transmission time. For the purposes of synchronization, the speed of light delay, and any other delay are indistinguishable, and need not be distinguished.
ooterness 18 hours ago [-]
Two-way time transfer measures the round-trip propagation time. As a result, it's not directly relevant to the accuracy.
catoc 16 hours ago [-]
So then you need to know distance / roundtrip-length within centimeter precision as well (below 29.98 cm for sub-nanosecond precision… to be precise).
Since cm precision is often not possible, is roundtrip-length an estimated average from prior roundtrips?
fsh 15 hours ago [-]
The roundtrip time is measured and compensated. Even NTP does this. Knowing the distance is not necessary for time synchronization.
Hmm one would expect heat expansion to change the length of fiber over tens of kilometers. Does it also affect light speed in the fiber? I think consumer fiber is not buried very deep on average, but maybe for these use cases you use something hefty anyway.
tempay 4 hours ago [-]
It doesn't matter if the length changes provided:
* you measure the round trip time often enough
* the shift affects light in both directions equally
netjiro 15 hours ago [-]
delay is easy
jitter kills
numpad0 11 hours ago [-]
... why would cm precision be often not possible?
dudinax 1 hours ago [-]
Because you're using fibres laid years ago. Often pulse velocity isn't known to better than one part in 1000.
colechristensen 18 hours ago [-]
The gravity well time dilation is about 3.5 nanoseconds per meter per year near the surface of the earth. (time changes rate with altitude in a gravity well)
Sub-nanosecond synchronization is getting into the relativity is measurable realm.
mike_hock 18 hours ago [-]
That means you get a free clock cycle every 2-3 hours on top of a mountain compared to sea level!
brookst 17 hours ago [-]
Datacenters in spaaaace!
Gravityloss 12 hours ago [-]
But they travel at 8 km/s so actually that cancels benefits? EDIT: checked, not enough to cancel them completely.
pwndByDeath 7 hours ago [-]
I wonder I'd that's the math for the ludicrous space data center ideas "floating" around...
OutOfHere 5 hours ago [-]
Just because you don't understand how something can work doesn't make it ludicrous. People like you are hell bent on destroying what's left of the Earth by turning it into a computer. If we left progress to those without an imagination, we wouldn't even have had a working calculator.
ted_dunning 29 minutes ago [-]
You're right.
It's the physics of cooling the beasts and the communication delays that make those plans ludicrous.
To turn your assertion on its head, the fact that the supporters don't seem to be able (or willing) to do the math to fact check these proposals is not an indicator that the plans will work.
As a starting point for comparison, the total power budget of the ISS is under 100kW and a single supercomputer rack dissipates about 4x that. What changes to the ISS can be made to get 100x more power and dissipate 100x more heat?
dboreham 43 minutes ago [-]
That might be true if the proponents had an argument of any sort in support of their plan, other than "we need idiots to give us money".
UltraSane 19 hours ago [-]
Yes, it uses phased locked loops and measures phase difference between the master clock and the local clock.
If this wasn't CERN tech I would think I was being taken for a ride. Conventional wisdom is that distributed consensus is not possible at this kind of performance, does anyone have a sense for how this is different and how my mental model is wrong?
tempay 4 hours ago [-]
> Conventional wisdom is that distributed consensus is not possible at this kind of performance
I'm not sure why you would think that? If you can assume the fiber is the same in both directions you know the round trip time is exactly double the latency of the connection. Then you know to phase shift your start time by that much when you get a start signal and you're in sync.
Obviously it's not trivial in practice, but it's not a fundamentally insurmountable problem.
davrosthedalek 19 minutes ago [-]
Twice the path delay + the time it takes to send the return packet. I assume WR does this in hardware to get a predictable time.
roughly 18 hours ago [-]
Haven't dug in on the technicals, but this is coming out of CERN, it looks like - and in that light, the links to "We're hiring" on that page almost feel like a flex...
andrewstuart 3 hours ago [-]
If your system depends on highly accurate time maybe you’re doing it wrong?
Of course no doubt there’s some requirements that cannot avoid that requirement.
But for the most part needing synchronized time is just going to be a forever problem isn’t it?
kikimora 13 hours ago [-]
What is significance of this?
perfmode 9 hours ago [-]
Distributed systems spend most of their effort on one problem: agreeing on the order of events across machines. Without synchronized physical clocks you have two options. Logical clocks (Lamport, vector) give you causal order but not wall-clock truth, so you can’t answer “did A really happen before B” for events that don’t have a happens-before relationship. Or you run consensus, which gives total order but costs round trips. At geographic scale that’s tens of milliseconds per decision, and the floor is set by the speed of light.
Tight clock sync collapses this. If clock uncertainty ε is small and bounded, you can timestamp a write, wait ε, and trust the global order without talking to anyone. Spanner’s external consistency works because TrueTime’s ε was a few milliseconds, so commit-wait was tolerable. The latency cost of planet-scale serializability stops depending on how far apart your replicas are and starts depending on how good your clocks are.
That’s the real significance. Time sync converts a coordination problem (bounded by physics) into a local computation (bounded by clock quality). Spanner proved this is possible but required GPS receivers and atomic clocks in every datacenter, which kept the capability inside Google for years. White Rabbit-class sync pushes ε from milliseconds toward sub-nanoseconds over commodity Ethernet hardware, and it’s now in IEEE 1588 as a standard PTP profile. If sub-nanosecond sync becomes baseline network infrastructure, the long-held assumption that strong consistency has to be slow at geographic scale stops holding, and a meaningful chunk of what databases currently work around (HLCs, weak isolation defaults, application-level reconciliation) becomes unnecessary.
hermitShell 6 hours ago [-]
Very good explanation and interesting take on the 'humanity scale' or internet scale significance. I work on a phased array system so significance of white rabbit for me was always sample alignment. Assumed CERN had a similar use case of needing to order (sensor data of) physical events happening far apart.
But if we imagine the vast majority of internet and telecom infrastructure is also implemented this way, we can reason about information over time in general. Makes me think of 'earth is a big computer' type of sci fi trope. Neat!
davrosthedalek 15 minutes ago [-]
Indeed, time synchronization across detectors is always tricky. Distributed clocks get messy at ATLAS dimensions. WR allows to distribute pretty good time sync over large detector systems.
Sometimes still not good enough though. Time-of-flight detectors try to get to single-digit ps level, and almost by definition, you have to synchronize two detectors that are some distance apart.
chrisweekly 1 hours ago [-]
Awesome explainer, thanks for that
Galanwe 9 hours ago [-]
Used quite a bit by stock exchanges to ensure consumers and publishers have a reasonably aligned time.
9 hours ago [-]
dkdcdev 12 hours ago [-]
it is useful e.g. to align the phase of signals being sent from different locations
9VMuzdNtfPK 5 hours ago [-]
[dead]
LowLevelKernel 17 hours ago [-]
Not on GitHub?
boguscoder 16 hours ago [-]
Its on gitlab but even there I failed to find sources, documentation/presentations are there though
Wololooo 14 hours ago [-]
If I remember everything is open hardware, CERN should have those repo accessible, last time I used it it was still very much in dev, especially their PCIe cards with custom kernel. This being said, I haven't touched it since ~6 years ago...
It's quite interesting - this isn't ethernet as we know it. Instead of each NIC using its own free-running clock, all the physical layers are sync'ed to each other at layer 1. (note that gigabit ethernet, which is what it uses, sends data at all times - when idle it sends the idle symbol)
P.S. There is WR workshop this week with some talks being publicly available on CERN's indico website.
Since cm precision is often not possible, is roundtrip-length an estimated average from prior roundtrips?
* you measure the round trip time often enough
* the shift affects light in both directions equally
jitter kills
Sub-nanosecond synchronization is getting into the relativity is measurable realm.
It's the physics of cooling the beasts and the communication delays that make those plans ludicrous.
To turn your assertion on its head, the fact that the supporters don't seem to be able (or willing) to do the math to fact check these proposals is not an indicator that the plans will work.
As a starting point for comparison, the total power budget of the ISS is under 100kW and a single supercomputer rack dissipates about 4x that. What changes to the ISS can be made to get 100x more power and dissipate 100x more heat?
In short, it's about giving PTP and SyncE some extra smarts.
I'm not sure why you would think that? If you can assume the fiber is the same in both directions you know the round trip time is exactly double the latency of the connection. Then you know to phase shift your start time by that much when you get a start signal and you're in sync.
Obviously it's not trivial in practice, but it's not a fundamentally insurmountable problem.
Of course no doubt there’s some requirements that cannot avoid that requirement.
But for the most part needing synchronized time is just going to be a forever problem isn’t it?
Tight clock sync collapses this. If clock uncertainty ε is small and bounded, you can timestamp a write, wait ε, and trust the global order without talking to anyone. Spanner’s external consistency works because TrueTime’s ε was a few milliseconds, so commit-wait was tolerable. The latency cost of planet-scale serializability stops depending on how far apart your replicas are and starts depending on how good your clocks are.
That’s the real significance. Time sync converts a coordination problem (bounded by physics) into a local computation (bounded by clock quality). Spanner proved this is possible but required GPS receivers and atomic clocks in every datacenter, which kept the capability inside Google for years. White Rabbit-class sync pushes ε from milliseconds toward sub-nanoseconds over commodity Ethernet hardware, and it’s now in IEEE 1588 as a standard PTP profile. If sub-nanosecond sync becomes baseline network infrastructure, the long-held assumption that strong consistency has to be slow at geographic scale stops holding, and a meaningful chunk of what databases currently work around (HLCs, weak isolation defaults, application-level reconciliation) becomes unnecessary.
But if we imagine the vast majority of internet and telecom infrastructure is also implemented this way, we can reason about information over time in general. Makes me think of 'earth is a big computer' type of sci fi trope. Neat!
Note that this is also for a large part a hardware-based technology