[ntp:questions] Proposed NTP solution for a network
bmwjason at bmwlt.com
Wed Mar 4 01:12:13 UTC 2009
> Jason <bmwjason at bmwlt.com> writes:
>> Terje Mathisen wrote:
>>> David J Taylor wrote:
>>>> Jason wrote:
>>>>> G. What have I missed, or gotten confused?
>>>> You forgot to say what the going rate is for the consultancy you want!
>>> Indeed. My standard rate is $200+ per hour.
>>> First of all: The 10s of us requirement within a site is _not_ feasible
>>> without direct PPS sync on every server, and since your blades cannot do
>>> this, the only alternative would be some other form of HW sync (like
>>> having blades that can slave their master clock to the host enclosure,
>>> and then using PPS to lock that master frequency).
>> This is what I was afraid of. When queiried, the manufacturer responded
>> with, well, nothing. Like they'd never heard of such a thing.
>>> OTOH 100-1000 us is in fact quite doable using standard NTP.
>> The critical time-stamp of the transactions must be tighter than 100us.
> Sorry, you simply cannot do it. the computers themselves have all sorts of
> internal processes which could delay things by 100us. Are you sure you are
> not confusing millisecods with microsecods (usec) a microsecond is
> 1/1000000 of a second.
Yes, but the sanity check is good.
EVen if you slave
> each computer to a GPS clock the internal workings can mess up the time by
> Who is setting this standard?
Software guys are driving the requirement. The systems and applications
are capable of < ~25 uSec (so I'm told) resolution between transactions.
(Very finely tuned application with little overhead on a tightly
controlled 8-core machine).
>>> I would as a bare minimum configure two S2 servers at each site, both of
>>> them as clients of all 3 GPS-based appliances, as well as 2-3
>>> internet-based sources (either pool.ntp or a private peering agreement
>>> with another nearby GPS-based company).
>> Good info, I'll start looking for both trusted (probably authenicated)
>> and accurate private sources -- some of our interconnected customers
>> also have sources.
>>> Finally I would use the same NTP configuration on every single of the
>>> remaining servers, syncing to all 6 of the secondary servers.
>> Without a doubt -- I'm pushing the sys admins to reconcile every server
>> NTP config.
>> Thank you.
More information about the questions