How to use correctly cassandra-stress tool in user profile mode?

I'm trying to test my 3 Cassandra(3.11.3.5) cluster nodes using cassandra-stress. Currently I'm running 3 nodes and 1 machine where cassandra-tool is running, everything is on an openvpn network.

I have created my .yaml user profile test file, here:

### DML ###

# Keyspace Name
keyspace: mykeyspace

# The CQL for creating a keyspace (optional if it already exists)
keyspace_definition: |
  CREATE KEYSPACE mykeyspace WITH replication = {'class': 'NetworkTopologyStrategy', 'DC1': '2'}  AND durable_writes = false;

# Table name
table: mytable

# The CQL for creating a table you wish to stress (optional if it already exists)
table_definition: |
  CREATE TABLE mytable (
    id bigint, 
    type int,
    txt text,
    event_datetime timestamp,
    bigtxt text,    
    page int,
    PRIMARY KEY ((id, type), page, event_datetime)
  ) WITH CLUSTERING ORDER BY (page DESC, event_datetime DESC)
    AND bloom_filter_fp_chance = 0.01
    AND caching = {'keys': 'ALL', 'rows_per_partition': 'NONE'}
    AND comment = 'ciao'
    AND compaction = {'class': 'org.apache.cassandra.db.compaction.SizeTieredCompactionStrategy', 'max_threshold': '32', 'min_threshold': '4'}
    AND compression = {'chunk_length_in_kb': '64', 'class': 'org.apache.cassandra.io.compress.LZ4Compressor'}
    AND crc_check_chance = 1.0
    AND dclocal_read_repair_chance = 0.1
    AND default_time_to_live = 0
    AND gc_grace_seconds = 90000
    AND max_index_interval = 2048
    AND memtable_flush_period_in_ms = 0
    AND min_index_interval = 128
    AND read_repair_chance = 0.0
    AND speculative_retry = '99PERCENTILE';

### Column Distribution Specifications ###

columnspec:
  - name: id
    size: gaussian(1..1000)       
    population: gaussian(1..500k)  

  - name: type
    size: gaussian(0..5)
    population: gaussian(1..5)

  - name: event_datetime
    cluster: fixed(1)         

  - name: page                  
    size: ~exp(1..20)
    population: ~exp(1..20)
    cluster: fixed(1)

  - name: txt
    size: exp(30..1k)

  - name: bigtxt
    size: gaussian(10k..30M)    


### Batch Ratio Distribution Specifications ###

insert:
  partitions: fixed(1)            # Our partition key is the domain so only insert one per batch

  select:    fixed(1)/1        

  batchtype: UNLOGGED             # Unlogged batches

queries:
   pages:
      cql: select id, page, type, txt, event_datetime, bigtxt from mytable where id = ? and type = ? and page=? limit 10;
      fields: multirow

Commands I run so far are modification of this one:

cassandra-stress user n=30 profile=./myprofile.yml ops\(insert=1\) -rate threads=10 -node 10.5.0.1,10.5.0.6,10.5.0.8

what I have changed:

• n
• thread count
• used throttle
• used fixed

and I always got some errors, like:

com.datastax.driver.core.exceptions.NoHostAvailableException: All host(s) tried for query failed (tried: 10.5.0.1/10.5.0.1:9042 (com.datastax.driver.core.exceptions.OperationTimedOutException: [10.5.0.1/10.5.0.1] Timed out waiting for server response), 10.5.0.6/10.5.0.6:9042 (com.datastax.driver.core.exceptions.OperationTimedOutException: [10.5.0.6/10.5.0.6] Timed out waiting for server response), 10.5.0.8/10.5.0.8:9042 (com.datastax.driver.core.exceptions.OperationTimedOutException: [10.5.0.8/10.5.0.8] Timed out waiting for server response))

I get those errors also in the warming phase.

And the run results is something like:

Results:
Op rate                   :        0 op/s  [insert: 1 op/s]
Partition rate            :        0 pk/s  [insert: 1 pk/s]
Row rate                  :        2 row/s [insert: 3 row/s]
Latency mean              : 38172.3 ms [insert: 38,172.3 ms]
Latency median            : 37279.0 ms [insert: 37,279.0 ms]
Latency 95th percentile   : 59190.0 ms [insert: 59,190.0 ms]
Latency 99th percentile   : 59458.5 ms [insert: 59,458.5 ms]
Latency 99.9th percentile : 59458.5 ms [insert: 59,458.5 ms]
Latency max               : 59458.5 ms [insert: 59,458.5 ms]
Total partitions          :         30 [insert: 30]
Total errors              :          0 [insert: 0]
Total GC count            : 2
Total GC memory           : 3.396 GiB
Total GC time             :    0.2 seconds
Avg GC time               :  117.0 ms
StdDev GC time            :    0.0 ms
Total operation time      : 00:01:04

What I cannot understand is:

1. why does latency time is completely different from what I get if I run nodetool cfhistograms mykeyspace mytable on nodes? on nodes I get something like 200ms in latency 99%ile for writes. Could be related to the time that cassandra-tool spends to send MBs of data to the coordinator?

2. why don't I see any effects in changing throttle/fixed rate switch in the cassandra-tool command? I always get 1 op/s

3. why do I get the previous errors so often if I have a production cluster, like the testing one, that doesn't have those errors even if it receives a much heavier mixed workload (600k reads and 1M writes in about 3 hours)?