Changing configuration not changing query time?

I am new to tigergraph, currently I am running a developer edition GSQL on a single machine.

I change some part of the setting:
GSE.RunningThreadNumber
GSE.LoadingThreadNumber
Perf.gpe.num.max_running_instances
Perf.gpe.num.post_threads
Perf.gpe.num.rebuild_threads
Perf.gpe.num.memory_server_threads

In default setting, they were integer around 2-4.
I assigned them all to 15 in my new setting.

I was expecting a decrease in query time, but the result is nothing different.

I use htop to monitor the CPU, I found out that in both the original setting and my new setting, when I run the query, almost all CPU(32 core) is used 100%, and running thread number is also around 32.

I was running connected_component from tigergraph algorithm library.

Anyone knows what’s going on here?

Many thanks in advance. :slight_smile:

Hi,

I am not sure if it is a good idea to tweak complicated system settings if you are new to TigerGraph. The default settings were established by after a long period of observations and experiments by the developers of the engine to ensure best performance. Changing them to a higher value without considering the side effects is not recommended, to say the least. And performance tuning of a developer edition on a single instance is a pointless exercise, since – as the name implies, – it’s a developer edition, not intended to be used in (production) environments where performance is crucial.

Using all cores/threads at the max is actually a sign of the database using all system resources efficiently, squeezing every drop of performance out the computer, so “tuning” system settings is probably not the best way to go to make it faster.

I would first (re)consider the schema design, and then the GSQL code to see it there are ways to improve performance, probably first by not doing things that do not need to be done, or doing things only once if the same thing is currently done multiple times. Some kind of pre-aggregation/pre-computing or creating some shortcuts in the graph might have profound impact on efficiency.

And if simply your data sat is too big, then you might want to have a small cluster, that you can easily get from TigerGraph Cloud (paid tier).