You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@cassandra.apache.org by Trevor Francis <tr...@tgrahamcapital.com> on 2012/04/19 17:24:42 UTC
Write Performance
Would there be any reason why I can't write more than 875 writes/sec to a cluster of 2 cassandra boxes? They are quad core machines with 8gb of ram running raid 10, so not huge servers….but certainly enough to handle a much larger load than that.
We are feeding data into it through a Flume sink. The disk io remains nearly nil along with load average across the cluster, but writes peak there. The data being written in is small (150-bytes). VNstat reveals ethernet bandwidth not exceeding 2mb/sec.
Not sure if this is a cassandra issue or a flume issue.
Thoughts?
Trevor Francis
Re: Write Performance
Posted by aaron morton <aa...@thelastpickle.com>.
You should be able to get more than that.
Run nodetool cfstats, look at the Write Latency (this is the recent latency, i.e. is reset each time you run it). This will give you an idea of how long an individual node is spending on a write.
Fire up JConsole, go to the StorageProxy MBean and look at the latency numbers there. This is how long the node takes to process a request for a client, it includes all network operations.
How many mutations are you sending in one batch ? You can get improvements by sending more, up to a point.
Are you using a single client thread / process ? Multiple clients is better.
Hope that helps.
-----------------
Aaron Morton
Freelance Developer
@aaronmorton
http://www.thelastpickle.com
On 20/04/2012, at 3:24 AM, Trevor Francis wrote:
> Would there be any reason why I can't write more than 875 writes/sec to a cluster of 2 cassandra boxes? They are quad core machines with 8gb of ram running raid 10, so not huge servers….but certainly enough to handle a much larger load than that.
>
> We are feeding data into it through a Flume sink. The disk io remains nearly nil along with load average across the cluster, but writes peak there. The data being written in is small (150-bytes). VNstat reveals ethernet bandwidth not exceeding 2mb/sec.
>
> Not sure if this is a cassandra issue or a flume issue.
>
> Thoughts?
>
>
> Trevor Francis
>
>