Cassandra write performance test (Why cassandra's write is so slow)

You know, one of the biggest advantage of Cassandra is the high writing speed. So I wrote a test on a single Cassandra node to see what's happening. What I found is reaaaaaaaally out of my expect. Based on test result, the write speed of Cassandra is much slower than Oracle. I don't know why.

Oracle and Cassandra are both running on the same server and same disk.

In Cassandra, each row has 20 columns. The row data is like:
cassandra> get Keyspace1.Standard2['key1']
=> (column=name9, value=value19, timestamp=1279794828940)
=> (column=name8, value=value18, timestamp=1279794828940)
=> (column=name7, value=value17, timestamp=1279794828940)
=> (column=name6, value=value16, timestamp=1279794828940)
=> (column=name5, value=value15, timestamp=1279794828940)
=> (column=name4, value=value14, timestamp=1279794828940)
=> (column=name3, value=value13, timestamp=1279794828940)
=> (column=name20, value=value120, timestamp=1279794828941)
=> (column=name2, value=value12, timestamp=1279794828940)
=> (column=name19, value=value119, timestamp=1279794828941)
=> (column=name18, value=value118, timestamp=1279794828941)
=> (column=name17, value=value117, timestamp=1279794828941)
=> (column=name16, value=value116, timestamp=1279794828941)
=> (column=name15, value=value115, timestamp=1279794828941)
=> (column=name14, value=value114, timestamp=1279794828941)
=> (column=name13, value=value113, timestamp=1279794828941)
=> (column=name12, value=value112, timestamp=1279794828941)
=> (column=name11, value=value111, timestamp=1279794828940)
=> (column=name10, value=value110, timestamp=1279794828940)
=> (column=name1, value=value11, timestamp=1279794828903)

I use Java Thrift API [b]batch_insert [/b]to insert rows. You can download my code from the attachments. The Cassandra test result is:
[table]
|[b]Number of Thread[/b]| [b]rows per-thread[/b]| [b]Time esapsed[/b] | [b]Rows per-second[/b]|
|1 | 2000 | 8927 ms | 224 |
|1 | 5000 | 13854 ms | 360 |
|1 | 10000 | 20286 ms | 492 |
|1 | 20000 | 31974 ms | 625 |
|1 | 50000 | 62965 ms | 794 |
|5 | 1000 | 12401 | 403 |
|5 | 2000 | 17571 ms | 569 |
|5 | 4000 | 26202 ms | 763 |
|5 | 10000 | 50364 ms | 992 |
|10 | 1000 | 22200 ms | 450 |
|10 | 2000 | 32712 ms | 611 |
|10 | 5000 | 65625 ms | 761 |

[/table]


[b]The Oracle test result is:[/b]

[b]50000 [/b]rows are inserted
JDBC batch insert speed: [b]4000 [/b]rows persecond
JDBC　single isnert speed: [b]330 [/b]row　persecond

Compare the test result above, it's obviously that [b]the write speed of Oracle is much faster than Cassandra [/b]. But everybody is talking the Cassandra is very fast... I am really confused. Did I make some mistake about Cassandra setting? I only changed the path of data and commit log.

Can anybody help me?