Will virtual nodes have worse performance?

[Benyi Wang <be...@gmail.com>]

I have a small cluster with 3 nodes and installed Cassandra 2.1.2 from
DataStax YUM repository. I knew 2.1.2 is not recommended for production.

The problem I observed is:

   - When I use vnode with num_token=256, the read latency is about 20ms
   for 50 percentile.
   - If I disable vnode, the read latency is about 1ms for 50 percentile.

I'm wondering what is the root cause of the worse performance for vnode:

   - Is ver 2.1.2 the root cause?
   - num_token is too high for 3-node cluster?

Thanks.

Re: Will virtual nodes have worse performance?

[Robert Coli <rc...@eventbrite.com>]

On Wed, Jul 15, 2015 at 9:28 AM, Benyi Wang <be...@gmail.com> wrote:

> I have a small cluster with 3 nodes and installed Cassandra 2.1.2 from
> DataStax YUM repository. I knew 2.1.2 is not recommended for production.
> I'm wondering what is the root cause of the worse performance for vnode:
>
>    - num_token is too high for 3-node cluster?
>
> Wish I had a blog post to link to, but... briefly..

If you have RF=3 and under probably about 9 nodes, you probably "just lose"
from having vnodes.

This is because the overhead of dealing with so many ranges is not
sufficiently counterbalanced by the advantages of for example significantly
faster bootstrap.

Vnode performance has been increasing all the time, because the other
advantages are significant and so people are motivated to improve it.

If you're interested in observing the significance of number of vnodes,
just recreate your cluster with half, then half again tokens per node,
testing each time?

=Rob