incubator-cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From aaron morton <aa...@thelastpickle.com>
Subject Re: read latency in cassandra
Date Fri, 04 Feb 2011 20:02:43 GMT
What operation are you calling ? Are you trying to read the entire row back?

How many SSTables do you have for the CF? Does your data have a lot of overwrites ? Have you
modified the default compaction settings ?

Do you have row cache enabled ? 

How long does the second request take ?

Can you use JConsole to check the read latency for the CF?

Sorry for all the questions, the answer to your initial question is "mmm, that does not sound
right. It will depend on...."

Aaron

On 5 Feb 2011, at 08:13, Dan Kuebrich wrote:

> Hi all,
> 
> It often takes more than two seconds to load:
> 
> - one row of ~450 events comprising ~600k
> - cluster size of 1
> - client is pycassa 1.04
> - timeout on recv
> - cold read (I believe)
> - load generally < 0.5 on a 4-core machine, 2 EC2 instance store drives for cassandra
> - cpu wait generally < 1%
> 
> Often the following sequence occurs:
> 
> 1. First attempt times out after 2 sec
> 2. Second attempt loads fine on immediate retry
> 
> So, I assume it's an issue about cache miss and going to disk.  Is 2 seconds the normal
"I went to disk" latency for cassandra?  What should we look to tune, if anything? I don't
think keeping everything in-memory is an option for us given dataset size and access pattern
(hot set is stuff being currently written, stuff being accessed is likely to be older).
> 
> I didn't notice this problem with cassandra 0.6.8 and pycassa 0.3.
> 
> Thanks,
> dan


Mime
View raw message