We are in the final stages of our POC on Vertica,
We are running a single query using a projection on a large table,
When running the query using a single thread the response time is good between 9 to 13 milliseconds,
However when we run the same query concurrent the response time decrease to 130 milliseconds to 10 concurrent queries ant to 200 milliseconds for 20 concurrent users,
We defined a special resource pool for the user query the database with this parameters:
name | is_internal | memorysize | maxmemorysize | executionparallelism | priority | runtimepriority | runtimeprioritythreshold | queuetimeout | plannedconcurrency | maxconcurrency | runtimecap | singleinitiator
---------+-------------+------------+---------------+----------------------+----------+-----------------+--------------------------+--------------+--------------------+----------------+------------+-----------------
my_pool | f | 4G | 4G | 24 | 100 | HIGH | 2 | 300 | 100 | | | f
And assigned the user to this resource pool,
The machine on which we run has 24 Intel cores,
Can Vertica handle 20 concurrent requests on a single server without increasing latency?
Is there some configuration we can change for it to handle such case?
Concurrent Users
Moderator: NorbertKrupa
Re: Concurrent Users
Hi!
>>Can Vertica handle 20 concurrent requests on a single server without increasing latency?
It should (24 cores) but it depends also on storage/disks (amount, latency). Probably bottle neck will be in IO (Vertica disk-based database, so storage is important too). On a single server it a good practice to have a couple disks(best option - storage per table
)
>>Is there some configuration we can change for it to handle such case?
For what you described have no too much many options - POOL and STORAGE. Also you can separate TEMP storage location from DATA, it should help a little.
---
Daniel
>>Can Vertica handle 20 concurrent requests on a single server without increasing latency?
It should (24 cores) but it depends also on storage/disks (amount, latency). Probably bottle neck will be in IO (Vertica disk-based database, so storage is important too). On a single server it a good practice to have a couple disks(best option - storage per table
)>>Is there some configuration we can change for it to handle such case?
For what you described have no too much many options - POOL and STORAGE. Also you can separate TEMP storage location from DATA, it should help a little.
---
Daniel