# Result differences in AWS and on-premise when running workflow

**URL:** <https://community.temporal.io/t/result-differences-in-aws-and-on-premise-when-running-workflow/5190>\
**Category:** Community Support\
**Created:** [July 2, 2022, 3:15pm UTC](https://community.temporal.io/t/result-differences-in-aws-and-on-premise-when-running-workflow/5190 "2022-07-02T15:15:18Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![ridwan.santoso](https://avatars.discourse-cdn.com/v4/letter/r/3ec8ea/32.png) [@ridwan.santoso](https://community.temporal.io/u/ridwan.santoso)\
**Post date:** [July 2, 2022, 3:15pm UTC](https://community.temporal.io/t/result-differences-in-aws-and-on-premise-when-running-workflow/5190/1 "2022-07-02T15:15:18Z")

</div>

Hi All,

Need help here, I am doing testing in 2 different environment:

- in AWS, the clients are in AWS EKS connecting to the Temporal instances and cassandra in the same EC2 VM (32 CPU 64GB RAM)
- on-premise, the clients are in Openshifts pods connecting to Temporal instances and cassandra in the same VM (also 32 CPU 64GB RAM)

Using the same settings I got different result where in AWS shows much better result (about 48 workflows per second) while on-premise only can generate 15 workflows per second.

I suspect the latency between client to temporal is the cause of this: in AWS the latency is less than 1 ms (between 0.877 ms to 0.984) while in on-premise varies between 3.90 ms to even 15.8 ms.

Can this latency cause the slower performance ? Or maybe any other things to check?

Please help.

Thanks & regards,

ridwan

---

<div class="post-metadata">

**Author:** ![tihomir](https://sea2.discourse-cdn.com/flex016/user_avatar/community.temporal.io/tihomir/32/6580_2.png) [@tihomir](https://community.temporal.io/u/tihomir)\
**Post date:** [July 2, 2022, 4:10pm UTC](https://community.temporal.io/t/result-differences-in-aws-and-on-premise-when-running-workflow/5190/2 "2022-07-02T16:10:21Z")

</div>

Yes network latencies can affect db latencies and can affect performance.  
Server emits `persistence_latency` metrics that you could check and compare in both scenarios for different operations:

`histogram_quantile(0.95, sum(rate(persistence_latency_bucket{}[1m])) by (operation, le))`

operations to maybe focus on are `CreateWorkflowExecution`, `UpdateWorkflowExecution`, `UpdateShard`