My company is currently testing a cassandra cluster with 7 nodes. When using thrift with the experimental CQL version 3.0.0 I am getting long latency times when I am retrieving large column families using token() call. Has anyone else seen this? This doesn't seem to happen with CQL version 2, I am retrieving in batches of 1000 records.
example below - CQL 3.
record count 1000
Retrieve number : 1
Retrieve for each call time : 2.199909
Total lat : 2.199909
record count 2000
Retrieve number : 2
Retrieve for each call time : 4.415638
Total lat : 6.615547
record count 3000
Retrieve number : 3
Retrieve for each call time : 7.191837
Total lat : 13.807384
record count 4000
Retrieve number : 4
Retrieve for each call time : 9.212177
Total lat : 23.019562
record count 5000
Retrieve number : 5
Retrieve for each call time : 11.386519
Total lat : 34.406081
…
example below - CQL 2.
record count 1000
Retrieve number : 1
Retrieve time : 0.167195
Total lat : 0.167195
record count 2000
Retrieve number : 2
Retrieve time : 0.182683
Total lat : 0.349879
record count 3000
Retrieve number : 3
Retrieve time : 0.180880
Total lat : 0.530758
record count 4000
Retrieve number : 4
Retrieve time : 0.182270
Total lat : 0.713028
record count 5000
Retrieve number : 5
Retrieve time : 0.195689
Total lat : 0.908717
I am using apache cassandra release 1.1.1, also on 1.1.2, and CQL version 3.0.0
