Sizing of Flume and resiliency

[Rani Yaroshinski <ra...@gmail.com>]

Hi all,

suppose I am using flume for the sake of transferring events from different
and complex sources
into HDFS sink mainly.

I am looking for some baseline paper for sizing flume installation, since
the only relevant material
for this, is dealing with event aggregations which are not needed by me.

I am looking for sizing from the point of view of memory, core and I/O
according
to the event's characteristicsm the number of sources and the number of
channels.

Another issue, which still falls into this section of architectural and
non-functional considerations
is the issue of fail over and resiliency.
I can see failover only for sinks, but not for channel, not for sources,
and not for agents, and machines running the flume-agent instance.

Regards,
Rani