[ntp:questions] question regarding NTP configuration for clusters, and "cluster time" stability
E-Mail Sent to this address will be added to the BlackLists
Null at BlackList.Anitech-Systems.invalid
Sat Nov 21 04:19:00 UTC 2009
Richard B. Gilbert wrote:
> BlackList wrote:
>> rotordyn at yahoo.com wrote:
>>> Look at it this way: We produce a system (in the form
>>> of a cluster) that works today, but that can drift away
>>> from UTC since it currently doesn't accept any external
>>> time reference. That drift isn't a huge issue, but over
>>> the multi-year lifespan of the hardware, can be significant.
>>> If we allow the use of external NTP servers, we now have
>>> opened up what was a closed system, and must prevent that
>>> new input source from causing instability. ...
>>> An alternate approach would be to make our cluster
>>> software resilient to intra-node time variations, but it
>>> was deemed simpler to use NTP to have the cluster accept
>>> an external UTC source.
The article I referenced seemed relevant to what they
are attempting. <http://lopsa.org/node/1480>
When the external server(s) go away, the internal
servers take some number of minutes, but eventually
elect a leader and settle on a usable hierarchy.
They may drift in total from the absolute notion of NTP
time, but they remain synchronized with each other.
> There are solutions that do not require that you get time
> from the internet! ...
> A GPS timing receiver ...
> A VLF receiver ... WWVB ...
They already mentioned several times,
they aren't going to go that route:
>> A HW time reference isn't feasible, nor should it
>> really be necessary given our modest accuracy
>> Adding hardware isn't an option.
>> This product exists in the field.
>> With thousands of deployed systems, adding hardware
>> simply isn't an option.
>> Besides the issue of thousands of systems in use,
>> there's the added one that we're always deployed
>> under a roof, never on top of it. :)
>> Adding hardware is not viable. Besides which, it really
>> is a fairly modest requirement of time accuracy, which
>> shouldn't require special hardware.
>> And while a hardware time source would cleanly solve the
>> issue, it isn't feasible to retrofit that to thousands
>> of existing installations.
>> Even if it were $500, there is the expense of getting it
>> installed into thousands of commercial data centers.
>> Which would be a logistical nightmare, given issues of
>> leased spaces, being deep inside skyscrapers, etc.
>> It isn't always as simple as sticking a GPS antenna
>> to a window.
E-Mail Sent to this address <BlackList at Anitech-Systems.com>
will be added to the BlackLists.
More information about the questions