3 ms·

If you read the Spanner paper, there is a hierarchy of timers involved. GPS is one level, but every datacenter also has machines equipped with atomic clocks. I

If you read the Spanner paper, there is a hierarchy of timers involved. GPS is one level, but every datacenter also has machines equipped with atomic clocks. I suspect even with GPS failure they could run entirely on atomic clocks, it would just increase their uncertainty, which would increase the commit times (Spanner is based on the idea that a transaction is committed when the time at which it was committed is guaranteed to have passed). As far as I know, it could actually run really slowly without special timing equipment.

rwmj 14y ago
Please someone tell me where you can buy an atomic clock :-?
- teraflop 14y ago
  Here you go: http://www.thinksrs.com/products/PRS10.htm http://www.thinksrs.com/products/PRS10.htm
- DanBC 14y ago
  Agilent used to do them, but someone else does them now: (http://www.home.agilent.com/en/pd-1000001383%3Aepsg%3Apro/primary-frequency-standard?cc=GB&lc=eng http://www.home.agilent.com/en/pd-1000001383%3Aepsg%3Apro/pr...) (http://www.symmetricom.com/ http://www.symmetricom.com/)
- timdoug 14y ago
  eBay. :) http://www.ebay.com/sch/i.html?_nkw=FE-5680A http://www.ebay.com/sch/i.html?_nkw=FE-5680A
cynwoody 14y ago
>I suspect even with GPS failure they could run entirely on atomic clocks, it would just increase their uncertainty Using GPS is an operational convenience, not a necessity, I would think. If GPS didn't exist, they would have to have a master atomic clock, say, in Mountain View. Then they would have to bring each remote clock to Mountain View, synch it with the master, and then ship it (running continuously) to its final destination. I remember perusing an HP catalog in the late eighties. You could buy a portable "traveling clock" for around $40,000. It was intended to be used to synchronize remote clocks. Here is a 1965 article from the Hewlett-Packard Journal, which describes the process: http://www.hpl.hp.com/hpjournal/pdfs/IssuePDFs/1965-04.pdf http://www.hpl.hp.com/hpjournal/pdfs/IssuePDFs/1965-04.pdf

rwmj 14y ago

Please someone tell me where you can buy an atomic clock :-?

teraflop 14y ago
Here you go: http://www.thinksrs.com/products/PRS10.htm http://www.thinksrs.com/products/PRS10.htm
DanBC 14y ago
Agilent used to do them, but someone else does them now: (http://www.home.agilent.com/en/pd-1000001383%3Aepsg%3Apro/primary-frequency-standard?cc=GB&lc=eng http://www.home.agilent.com/en/pd-1000001383%3Aepsg%3Apro/pr...) (http://www.symmetricom.com/ http://www.symmetricom.com/)
timdoug 14y ago
eBay. :) http://www.ebay.com/sch/i.html?_nkw=FE-5680A http://www.ebay.com/sch/i.html?_nkw=FE-5680A

cynwoody 14y ago

>I suspect even with GPS failure they could run entirely on atomic clocks, it would just increase their uncertainty Using GPS is an operational convenience, not a necessity, I would think. If GPS didn't exist, they would have to have a master atomic clock, say, in Mountain View. Then they would have to bring each remote clock to Mountain View, synch it with the master, and then ship it (running continuously) to its final destination. I remember perusing an HP catalog in the late eighties. You could buy a portable "traveling clock" for around $40,000. It was intended to be used to synchronize remote clocks. Here is a 1965 article from the Hewlett-Packard Journal, which describes the process: http://www.hpl.hp.com/hpjournal/pdfs/IssuePDFs/1965-04.pdf http://www.hpl.hp.com/hpjournal/pdfs/IssuePDFs/1965-04.pdf