[Spread-users] WAN usage
John Robinson
jr at vertica.com
Thu Nov 15 17:14:42 EST 2007
I know this comes up on the list from time to time, and now it is my
turn to query.
I did a brief experiment (then lost the machines) where a real WAN
intervened between two local segments (same machine room) and one remote
segment (hosted machine elsewhere). Connectivity between segments used
a VPN arrangement so all ports could go through without need of firewalls.
The daemons all came up and found each other fine, but the remote one
would wink out every couple of minutes (membership change messages
etc.). The attached client processes were not sending any traffic
except a few brief messages to track their own group status in response
to spread's membership change messages.
I did not have a chance to try bandwidth checks with the set-up. Is
there any good rule-of-thumb for what bandwidth, latency and delivery
probability guarantees the spread daemons need to stay connected? Do I
need to spmonitor the setup? I am perfectly willing to believe that the
service I was receiving in my setup was atrocious.
Also, I had set broadcast addresses on the segments with 1 machine per
segment. I now know that this broadcasts needlessly on the segments but
aside from that should there be any ill effect on the inter-segment packets?
I hope to get to experiment more with the setup over the next few
days/weeks.
Many thanks,
/jr
More information about the Spread-users
mailing list