Scylla vs Cassandra on a single laptop node using cassandra-stress

[pjsofts@gmail.com]

Hey guys,

Before doing a real benchmark on our company's servers, I tried to do a simple test by installing Centos 7.0 on my laptop. 

cassandra-stress with Cassandra reported 54000 ops/sec while with Scylla only 24000 op/s.

Cassandra could handle more than hundred threads while with Scylla timeouts started much earlier.

Is this because there's no network, partitioning overhead?

Sincerely, 

Pouria Jahandideh

[Eyal Gutkind]

Hi Pouria,

Thank you for giving Scylla a try.

While we want to make sure you see the benefits of Scylla, a Laptop environment is not something we recommend for benchmarking.

It is great for getting started. But, using a laptop, my guess you had to turn scylla-developer mode to 1 and had a setting of a single core.

Can you also elaborate on the file system you are using?

I'd recommend you try to replicate your staging/production environment with Scylla. I'm confident you'll be able to see Scylla's benefits.

If you provide information on the actual environment (# of CPUs, RAM and disks and operating system use) we can estimate performance for typical stress tests.

Data model will have its impact on all solutions.

Regards,  

--
You received this message because you are subscribed to the Google Groups "ScyllaDB users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to scylladb-users+unsubscribe@googlegroups.com.
To post to this group, send email to scylladb-users@googlegroups.com.
Visit this group at https://groups.google.com/group/scylladb-users.
To view this discussion on the web visit https://groups.google.com/d/msgid/scylladb-users/e9d273a6-508d-477c-b018-4a0ad5ba7c78%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

Eyal Gutkind

ScyllaDB

www.scylladb.com

[Pekka Enberg]

Hello,

On Wed, Dec 14, 2016 at 10:25 PM, <pjs...@gmail.com> wrote:
> Before doing a real benchmark on our company's servers, I tried to do a
> simple test by installing Centos 7.0 on my laptop.
> cassandra-stress with Cassandra reported 54000 ops/sec while with Scylla
> only 24000 op/s.
> Cassandra could handle more than hundred threads while with Scylla timeouts
> started much earlier.
>
> Is this because there's no network, partitioning overhead?

There are two things at play here: (1) Scylla is very sensitive to
other threads stealing the CPU's it is running its reactor loop on and
(2) Scylla takes over _all CPUs_ by default. So as you start the
"cassandra-stress" process on the same machine, you'll have less CPU
resources to run it and Scylla's performance is hurt because of CPU
sharing.

I don't really recommend running either Scylla or Cassandra and the
load generator on the same machine. But if you insist on doing that,
you probably should limit the number of CPUs Scylla is using with the
"--smp" command line argument.

- Pekka

[Pouria Jahandideh]

Thank you guys, I know this benchmark doesn't mean anything. 

I've decided to ask our infrastructure team to provide us with three Centos 7 nodes so that I can do a more reasonable benchmark on the deployment configuration. 

--

PJ