On 24 Jun 2015, at 18:56, Kevin Liu <kevin...@fb.com<mailto:kevin...@fb.com>> wrote:
Continuing this thread beyond standalone - onto clusters, does anyone have experience successfully running any Spark cluster on IPv6 only (not dual stack) machines? More companies are moving to IPv6 and some such as Facebook are only allocating new clusters on IPv6 only network, so this is getting more relevant. YARN still doesn’t support IPv6 per http://wiki.apache.org/hadoop/HadoopIPv6 Facebook are doing work on IPv6, which they are clearly using in-house https://issues.apache.org/jira/browse/HADOOP-11890 Nobody has been reviewing that work, but we should. For most people you can just say "don't use ipv6", but FB have probably run out of IP addrs somewhere and are stuck unless they buy HP for the two Class-A subnets it holds. If people were to play with those patches on IPv6, it could make it into Hadoop 2.8
