Wondering if what cluster config for aerospike to meet following SLA


#1

Hi,

We are evaluating aerospike for following use case:

  • UUID Primary Key - String Value
  • ~ 5 Billion records
  • ~ 15000 qps for read
  • ~ 20000 records/s for write
  • remove everything whose last_updated_at is before 1 year
  • across multiple data centers and cloud providers but sync latency can be at seconds
  • read latency 95percentile < 15 ms

Wondering if aerospike can do this, and if yes, how? What would be the cluster/db config look like? Can we deploy this on kubernetes (> 1.5) and achieve this performance?


#2

We have K8 templates https://github.com/aerospike/aerospike-kubernetes and the community has released a Helm Chart as well: https://github.com/kubernetes/charts/tree/master/stable/aerospike

Please consult with our Sales/Solutions Architects about specific sizing. You would need XDR, which is an Enterprise feature for your DC to DC replication. @rbotzer thoughts?


#3

Hi. As Richard pointed out, cross datacenter replication (XDR) is an enterprise feature. Please look at the product matrix, and there’s a ‘Contact us’ button at the top, if you’re interested to talk to a sales rep about Aerospike EE.

Which SSDs do you have? Are you on bare-metal, or in a cloud provider (Amazon EC2, Azure, GCP, etc)

If you read the SSD section of the capacity planning guide, and look at Aerospike’s open source tool for benchmarking SSDs (ACT), you’ll see that our standard apples-to-apples benchmarking of drives gives a passing grade as long as no more than 5% of operations exceed 1ms. You’re OK with 95% < 15ms. That typically requires a custom ACT test run against your SSDs. The sustainable read and write TPS discovered by ACT for your specific spec can then be used to size the cluster.

With some more information I can try to help you. Alternatively, reach out to sales about EE, and sizing and tuning of the cluster would be part of that process.