> -----Original Message-----
> From: Testbed Support for GridPP member institutes [mailto:TB-
> [log in to unmask]] On Behalf Of Jeremy Coles
>
> I think all the UK probes were registered as public and I can give you
> probe numbers… but site admins at those sites may want to comment on their
> own connectivity too!
>
This is us: https://atlas.ripe.net/probes/11551/ I would just slightly caveat the Oxford numbers if you go back a way, because we had an outage of the RIPE atlas probe itself as a knock-on effect of the Shellshock inspired PerfSonar downtime - the RIPE probe would have been fine, but I forgot that it was USB powered from one of the perfSonar boxes when I turned them off.
That aside, I think the connection as delivered by Janet to Oxford probably is 100%, but there are problems that occur closer to the edges. Off the top of my head, bits of the Oxford network have had outages caused by failing switch PSUs, failing mains power (and exhaustion of UPS batteries), and the classic digger through fibre. Our forthcoming University network replacement will improve the resilience further out to the edge, but the current/old one has a decently resilient core (so we're unlikely to lose the lot) but the path between that core and any given edge device will pass through many potential single points of failure.
My impression, not backed by stats, is that the RAL site network has relatively many fairly brief outages, and they have a widespread effect, whereas Oxford outages are less often, less widespread (arguably, they're less often for a given endpoint /because/ they're quite localised), but when they do happen they tend to be more serious and last longer. That may give an impression that one network is 'unreliable' whereas the other is reliable but subject to occasional natural disasters, even if the total percentage availability were similar.
Ewan
|