Sometimes hbase do not handle any request at all

[Zhou, Yunqing]

I'm using IdentityTableReducer to insert about 1M records into hbase
on a 23 machines cluster.
but I found that sometimes it got junked.
everything suspended. all machine's work load are zeros.

then some reducer failed, new reducer begin to insert some records.
then the pheomenon appeared again.

I've set all machines nofile limit to 32768.
Do you know that's why?
Thanks.