MySQL Forums
Forum List  »  NDB clusters

Re: Cluster performance degradation
Posted by: Rytis Sileika
Date: October 23, 2004 01:45PM

It seems that I have found the answer by myself.
Here's what I have tested and the test results:
Test1:
Node1: NDB+API
Node2: NDB
Node3: Test
[Node4: Test, but the results came out the same]
Perf: ~2200/1700 (INSERT/SELECT)
Test2:
Node1: NDB+API
Node2: NDB+API
Node3: Test (towards N1)
Node4: Test (towards N2)
Perf: ~3600/2300 (same DB, just two different tables)
Thus increasing number of API's really helps a lot. This just once again proves, that MySQL has a serialized link towards DB engine (this was already mentioned somewhere).
Now I'm on my quest of finding why NDB becomes quite unstable at these speeds...

For those who are curious:
table is test_table (i int, s varchar(2040), primary key using hash(i)),
inserts are sequential (so, insert into ... (1, 'AAAA...'), then (2, 'AAAA...' and so on) within each writing thread, there were approx 14 writing threads.
selects were completely random, thus: select from ... where i = rnd(1,x), there were approx 14 reading threads as well.

Options: ReplyQuote


Subject
Views
Written By
Posted
4045
October 21, 2004 11:01AM
Re: Cluster performance degradation
2797
October 23, 2004 01:45PM


Sorry, you can't reply to this topic. It has been closed.

Content reproduced on this site is the property of the respective copyright holders. It is not reviewed in advance by Oracle and does not necessarily represent the opinion of Oracle or any other party.