We have Elastic Search installed on a Google Cloud Compute Engine. We query it via a Google Cloud App Engine. Now that should not be relevant....
The query execution time of Elastic is great! But our problem is the connection time.
At the moment about a significant of the time it takes to generate a page is the connection time sending the request to and getting the answer from Elastic.
Now significant is relative to the rest. The rest is fast! Super fast! A page (from request to sending html) for a specific search will take 80 to 200msec.
In these cases the connection time is between 75 and 160 mseconds. It is also strangely inconsistant. For the same query it can jump + or - 30%.
Now my questions are:
- Can it be that ES is overly optimisitic about its search performance and therefor I measure more time?
- Did someone already look into these kind optimizations?
- Does someone also measure the performance like this?