benchmarking of thrift versus avro

[Phil Stanhope <st...@gmail.com>]

Has there been any benchmarking done on thrift vs avro bindings in 0.7
codebase?

If one was about to bring into alpha production in the Jan 2011 timeframe,
would going with avro be recommended?

If one was to benchmark ... I'd be more interested in benchmarks that
measure access to a node that actually has the data (i.e. ring awareness)
and that are at CL:1 on the read. The CFs and their mutations have extremely
low probability of concurrent writes (or overwrites) and are thus also
configured for CL:1.

I've got a pre-alpha (from my perspective) set of services that are using
thrift from both PHP and Java (not hector). All developers are isolated
behind scale agnostic tier ... so the changes to get onto 0.7/0.7.1 by end
of year and shift to avro would be completely isolated from the remainder of
the development team.

Thanks in advance.

Re: benchmarking of thrift versus avro

[Eric Evans <ee...@rackspace.com>]

On Mon, 2010-08-30 at 11:43 -0400, Phil Stanhope wrote:
> Has there been any benchmarking done on thrift vs avro bindings in 0.7
> codebase?

No, it's kinda early for that.  First priority IMO is to reach feature
parity with the thrift side, and then get comprehensive functional tests
to make sure that it's stable and stays that way.

That's not to say that I'd get in the way of anyone that wanted to work
on this (to the contrary in fact), just trying to convey why it has been
done (by me anyway) yet.

Some of us are meeting a co-working facility in Austin on Wednesday to
burn through as much as possible, hopefully we can get far enough to
start working on performance evaluation.

> If one was about to bring into alpha production in the Jan 2011
> timeframe, would going with avro be recommended? 

We're going to need people to test, so I wouldn't want to be too
discouraging.  But, from a don't-count-your-chickens standpoint, I
wouldn't recommend it, no.

-- 
Eric Evans
eevans@rackspace.com

Re: benchmarking of thrift versus avro

[Phil Stanhope <st...@gmail.com>]

Thanks. The development was started in March.

Have shifted through each of the 0.6 releases with no problems so far.

The jump to 0.7 creates many opportunities. For example I have use case for
dynamic keyspaces that could result in hundreds created per quarter once in
production. I'd like to focus on that and not the communication tier
changes. More interested in the API changes as well as standardized hadoop /
lucandra than RPC level changes.

On Mon, Aug 30, 2010 at 11:56 AM, Jonathan Ellis <jb...@gmail.com> wrote:

> On Mon, Aug 30, 2010 at 10:43 AM, Phil Stanhope <st...@gmail.com>
> wrote:
> > Has there been any benchmarking done on thrift vs avro bindings in 0.7
> > codebase?
> >
> > If one was about to bring into alpha production in the Jan 2011
> timeframe,
> > would going with avro be recommended?
>
> If you're starting dev now or soon, I wouldn't recommend it.
>
> --
> Jonathan Ellis
> Project Chair, Apache Cassandra
> co-founder of Riptano, the source for professional Cassandra support
> http://riptano.com
>

Re: benchmarking of thrift versus avro

[Jonathan Ellis <jb...@gmail.com>]

On Mon, Aug 30, 2010 at 10:43 AM, Phil Stanhope <st...@gmail.com> wrote:
> Has there been any benchmarking done on thrift vs avro bindings in 0.7
> codebase?
>
> If one was about to bring into alpha production in the Jan 2011 timeframe,
> would going with avro be recommended?

If you're starting dev now or soon, I wouldn't recommend it.

-- 
Jonathan Ellis
Project Chair, Apache Cassandra
co-founder of Riptano, the source for professional Cassandra support
http://riptano.com