[time-nuts] Google NTP Servers and smearing leap seconds...

[xaos at darksmile.net]

You are right.

To be more precise, I should have said that the time sync should be at  
least 1 order of magnitude less. In the case of <10us turnaround time,  
it is assumed that the timesync is <1us. This is the reason that  
everyone uses multiple stratum 1 NTP servers using GPS in their  
datacenters.

So the Forex transaction goes like this:

1. (Both parties) Are we in proper sync timewise?
2. (Party 1) I need transaction type x. My timestamp is: xxxx.xxxx. Go.
3. (PArty 2) Confirmed. My timestamp is: yyyy.yyyy. Go.

These timestamps are legal entities and bind both parties to the transaction.
That's why transactions have a data transfer entity in the middle  
(Reuters, Bloomberg) which guarantees proper timesync for all involved.

With Reuters in the middle, only the Reuters timestamp (arrival time  
and send time) can be trusted.

However, Many times you will see a Reuters machine lose sync and the  
UNIX SA's will restart NTP on it. Reuters puts more than one machine  
per site for redundancy.

Quoting Hal Murray <hmurray at megapathdsl.net>:

>
> xaos at darksmile.net said:
>> You can forget Wall St. firms and Banks for starters.
>
>> They need sub-microsecond accurate timing as some instruments (Forex)   are
>> moving to <10 microsecond latency from order entry to order ack.
>
> 10 microsecond latency doesn't say anything about how accurate the time has
> to be.
>
> Does anybody have a good URL on the accuracy requirements of banks and/or
> stock markets?  I expect there are both legal and technical issues.  I'd like
> to understand them separately but I won't be surprised if they are thoroughly
> tangled.
>
> --
> These are my opinions, not necessarily my employer's.  I hate spam.
>
>
>
>
> _______________________________________________
> time-nuts mailing list -- time-nuts at febo.com
> To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
> and follow the instructions there.
>