Is it reasonable to share CQL workload and DDB workload on the same cluster ?

[Cong Guo]

Hi experts,

I want to share CQL workload an DDB workload on the same cluster to reduce the effort for maintaining too many clusters. Do you know is it any side-effect for such usage?

Thanks, 

Cong

[bdenes@scylladb.com]

What is DDB? If this is another software then this is not recommended, Scylla doesn't like to share the hardware with others. That said it is possible, if you pin them to different CPU sets, carefully partition memory and have them use different disks. But still, scylla works best if left alone, undisturbed by neighbours.

--
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-user...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/scylladb-users/48024928-9baa-4424-a762-a03ffc0f1d65n%40googlegroups.com.

[Nadav Har'El]

On Fri, May 13, 2022 at 5:17 AM 'Cong Guo' via ScyllaDB users <scyllad...@googlegroups.com> wrote:

Hi experts,

I want to share CQL workload an DDB workload on the same cluster to reduce the effort for maintaining too many clusters. Do you know is it any side-effect for such usage?

Hi Cong,

I assume by "DDB" you mean the DynamoDB API. I.e., Alternator.

Yes, you can definitely share the same cluster for both CQL and DynamoDB APIs. Scylla can have both listening ports open at the same time,

and all the background work (compactions, repair, etc.) and underlying implementation (memory and disk use, etc.) are more-or-less the

same for both anyway.

What you should *not* do is attempt to access or modify the same table through the two APIs - e.g., create a table with the DynamoDB API and

then attempt to read it with the CQL API. Although technically this might be possible, we recommend against it because it may stop working if the

Alternator implementation changes in the future, and the low-level format of how Alternator stores data changes.

But if you want to use the same cluster for different DynamoDB and CQL tables - that's fine.

Of course, the usual caveats of running multiple workloads in the same cluster apply. For example, if one workload is a heavy

throughput-oriented workload (e.g., data ingestion or OLAP) and the other workload is latency-critical (e.g., OLTP), one may

hurt the performance of the other. The enterprise version of Scylla has some additional features to improve the performance

isolation between completely different workloads with different goals which happen to run on the same cluster.

[Cong Guo]

Hi Nadav,

Yes I mean DDB as Dynamodb API. I see the points to pay attention to, we could shared the cluster, but we'd better not to share the keyspace/table. Thanks a lot for your reply!

Thanks,
Cong

[Cong Guo]

Hi Botond,

I mean ddb as Dynamodb API access via Alternator. Sorry for the misleading expression.

Thanks,
Cong