Details
-
Bug
-
Status: Resolved
-
Normal
-
Resolution: Duplicate
-
None
-
None
-
None
-
RHEL 6.3, Cassandra 1:1.2.4-1, multi-dc environment
-
Normal
Description
Currently Cassandra seems to be sending data directly to all replicas in a multi-dc environment rather than to a datacenter specific coordinator node.
This should have been fixed in CASSANDRA-1530 (versions .7+). But with replication factor 1 for a keyspace in far DC we were sending 25Mbps over the WAN, then increasing the RF to 3 increased WAN traffic to 75Mbps. Also, after a 10 hour test run with RF=3 an estimate of the data sent over the WAN very closely correlated with the data residing on the local disks.
I'm unsure what other versions this effects. If this is fixed somewhere already I would be highly interested but I can't find a Jira ticket regarding this.
Attachments
Issue Links
- duplicates
-
CASSANDRA-5632 Cross-DC bandwidth-saving broken
- Resolved