Part 2 : Infrastructure Setting - Single vs Multiple data stream

Anirudha | Sat, 09/21/2019 - 15:18

Lets take look at Infrastructure and see what all things to check there.

Networking
  • If your application is capable of pushing data in multiple streams but you are still facing issues. Then you may want to check if network is a bottleneck between client and server.
  • You should use min 10GBPs link while deploying Nutanix Objects cluster , higher the network speed better throughput. But if you have lower network link between client to Objects cluster e.g - 1GBPs, then no matter how hard you push, you have only 1GB network pipe going to your Objects cluster.
Load balancers and nodes in your Nutanix Objects cluster
  • If you have deployed single node Nutanix Objects cluster. They you will have just one load balancer and one node Objects cluster deployed. You are again going to be bottleneck at network and small cluster size.
  • If you have more than one load balancer (Refer this for more details) then you will get higher throughput as number of endpoint IPs increases. You can have max 4 load balancer , so in this case look at this aspect as well.
  • You have 4 load balancer but your application is still not able to push data at higher speed :
    • In this case you should check your application setting. Is all the endpoints configured with the client?
    • If all the endpoints are configured then if your application is able to use all the endpoints at the same time. Verify if it has any active-active Or active-passive settings. You should keep all the endpoints as active.
    • Make sure the client is able to connect to all the endpoints. Try running test on each endpoint to see if you are getting even throughput from each endpoint. Double check your physical network settings in case you see uneven number. 
  • Are you using DNS load balancing (Refer this article for more details) for routing traffic to all Nutanix Objects endpoint IPs? - check if your DNS server is configured properly to forward all requests to all endpoints . If you keep higher TTL then you may or may not be able to use all the endpoints at the same time.
AOS cluster:
  • Number of Nodes in underlying AOS cluster
    • There is a possibility you can have larger Nutanix Objects cluster deployed on small AOS cluster. E.g 5 Node Nutanix Objects cluster on 3 Node AOS cluster. 
    • Deployment wizard doesn’t allow this configuration by default, you can deploy Nutanix Objects cluster as large as your AOS cluster. But if you remove a couple of AOS nodes post  deployment, then you may end up in having bigger Nutanix Objects cluster deployed on smaller AOS cluster.
    • In such a case you may get bottleneck by underlying AOS cluster throughput. So make sure underlying AOS cluster is equal or larger than Nutanix Objects cluster.
  • Mixed Workload
    • If you using underlying AOS cluster for other workload then you have more services running on the same cluster sharing all the underlying resources. 
    • In this case you will be again bottleneck by your physical cluster Or rather your cluster is not properly sized for the respective workload.
    • Make sure your Nutanix Objects deployment is properly sized. (Refer # this article for more details)
Platform you deployed (Refer#This)
  • If you deployed your Nutanix Objects cluster on lower platform such as NX3x or NX1x platform which comes with just 4HDD’s. Then your overall throughput is bottleneck by your physical cluster.
  • Consider using NX8X or HP380 or similar platform for your Nutanix Objects deployment. It will not only give you better performance but also reduce your cost per GB
Number of Nutanix Objects clusters deployed on single AOS cluster
  • Nutanix gives you flexibility to deploy multiple  Objects cluster on same AOS cluster. While its useful and flexible feature, but it's important to understand all these Nutanix Objects instances will share the same AOS resources.
  • If you have multiple applications actively using multiple Nutanix Objects instances then  AOS bandwidth will be shared by everyone , and you may not get max throughput from any single Nutanix Objects cluster.