Title
#users-public
Manoj DeSilva

Manoj DeSilva

09/15/2022, 5:54 AM
Hi Team, We also have want to know about the same above, We are using 10 TCP ILP connections and QuestDB does not seems to process data more than 1MB per second so our data sends get blocked. We are using following configuration.
QDB_METRICS_ENABLED="TRUE"
QDB_PG_SELECT_CACHE_ENABLED="FALSE"
QDB_HTTP_QUERY_CACHE_ENABLED="FALSE"
QDB_CAIRO_COMMIT_LAG="30000"
QDB_LINE_TCP_IO_WORKER_COUNT="10"
So my concerns are,1. Is there any other configs that we can use to optimize ILP ingestion throughput. 2. What would be the optimal number of TCP connections to use. 3. Generally what amount of throughput Quest DB ILP can handle.
Jaromir Hamala

Jaromir Hamala

09/15/2022, 8:22 AM
hi Manoj, what is the latency between QuestDB client and QuestDB server? Are they on the same network? What are you using as a disk? Is it SSD/HDD/AWS EBS/something_else?
Manoj DeSilva

Manoj DeSilva

09/15/2022, 8:23 AM
Hi @javier ramirez, Thanks for the response. Both QuestDB and client runs on a Linode virtual machine. I think it is a HDD.
Jaromir Hamala

Jaromir Hamala

09/15/2022, 8:25 AM
do you have any monitoring in place? can you see CPU and I/O utilization?
Manoj DeSilva

Manoj DeSilva

09/15/2022, 8:27 AM
Sorry. we don't have it.
Jaromir Hamala

Jaromir Hamala

09/15/2022, 8:28 AM
what are the specs if your linode box? what do you use on a client side? ILP client? Python? are the 10 clients running on the same linode box as questdb?
Manoj DeSilva

Manoj DeSilva

09/15/2022, 8:35 AM
Specs are: • MEmory: 192 GB • CPUs: 32 • Storage: 3840 GB • Network In/Out : 40/12 Gbps ILP C++ Client written by us. yes. 10 clients running on the same machine.
Jaromir Hamala

Jaromir Hamala

09/15/2022, 4:15 PM
oh, that’s a pretty good box. do you have a designated timestamp defined? are data in-order or out-of-order?
Manoj DeSilva

Manoj DeSilva

09/15/2022, 5:49 PM
Yes. we have designated timestamps and there could be out of order data.
Jaromir Hamala

Jaromir Hamala

09/15/2022, 5:52 PM
Can you try to ingest test data which are ordered? That should give you performance baseline for in-order data. If that’s similar to 1MB/s you observed then something is wrong. If it’s >> 1MB/s then we’ll know we should tune out of order parameters.
Manoj DeSilva

Manoj DeSilva

09/16/2022, 11:56 AM
Thanks @Jaromir Hamala, In our setup we don't have a capability to filter the data. When I reduced number of ILP connections to QuestDB to 1, seems like issue was resolved. I am still to figure out whether this is the change that resolved the issue. If we insert data from multiple connections concurrently to one table will hinder the ILP ingest performance? Because we insert data to one table from 10 connections concurrently.
Jaromir Hamala

Jaromir Hamala

09/16/2022, 12:07 PM
hi Manoj, I’m glad to hear the issue is gone. using multiple connections helps to saturate your network. depending on network latency it can be tricky with just a single connection. but if just a single connection is enough then I would stick to that. with multiple connections you run into higher risk of out-of-order inserts.
Manoj DeSilva

Manoj DeSilva

09/16/2022, 12:08 PM
Okay. Thanks Jaromir.