Originally from the User Slack
@Narayan_Ramamurthi: HI! I had a nice chat with @Lawrence_Wan today.
Basically here is our use case.
We are a telecom billing product development company with over 200+ installations across the globe.
Our billing product is primarily written in C/C++ on unix like platforms including RHEL linux, sun (, which is now oracle) solaris, aix, and hp-ux.
We also have parallel teams who write the surrounding software in various technologies including Java, Python and Kotilin.
We have traditionally been an Oracle-Database company
But over time in the recent years we started using Apache-/Datastacks-Cassandra both for ourselves and also our clients.
When Cassandra was being proposed, I also proposed using ScyllaDB as an alternative but our management did not consider my request at that time.
After happily using Cassandra for years now, we are facing troubles with Datastacks support and Apache open-source support to the cassandra-cpp-driver.
So, I brought up this topic of Scylla with the management. They seem to be more willing to listen this time.
I should also give some more background - our parallel Java/Go team has done a recent POC with the ScyllaDB and their results indicate that Cassandra performs 3x better than Scylla.
Our management is quoting this as an example to me now.
I am looking to prove them wrong, I am here seeking support to learn what can be done to prove them wrong.
Basically what settings can be done on the cluster and on the client side to make sure that Scylla beats Cassandra in fair competition.
I seek this vibrant community’s support to kindly suggest the needful.
Thanks,
@avi: Please specify the hardware on which you deployed.
To understand your performance problem, it’s best to set up scylla-monitoring and work with us to identify the problem.
@Narayan_Ramamurthi: intel xeon server family
Both Cassandra and ScyllaDB clusters were deployed using similar hardware
@avi: Please provide more information, number of cores, amount of memory, disk types
And set up scylla-monitoring
@Narayan_Ramamurthi: We have been using apache cassandra for quite a while now, and our experience says that cassandra is possibly good at doing writes but it is not really good at doing reads.
On the other hand, Oracle (the legacy SQL datastore) does it way better for either, but on a single node.
Can we expect something better with ScyllaDb? If so, can you provide me information where we can do both reads and writes much better, possibly unlike cassandra?
Regarding number of cores, hardware, etc: for the sake of comparison we shall setup identical hardware for cassandra and scylla: Haswell-based processors, something like Xeon E5-2683 v4 or the likes.
Note that this is being pursued in the POC (proof of concept) mode at the moment, so foresight is important.
@avi: Thank you in advance for your ever prompt response. I will be eagerly watching this space for help.
@avi: ScyllaDB should be faster than Cassandra for both reads and writes