How to speed up valueMap, values, properties query?

[vitaliishandra]

We have queries like this: g.V().has('AccountId', 1).out('Direct'). It is executed few milliseconds and returns ~13000 results. However when I try to query the values using one of the follwoing queries it took about 35-45 seconds to complete:
g.V().has('AccountId', 1).out('Direct').values('uid')
g.V().has('AccountId', 1).out('Direct').valueMap()
g.V().has('AccountId', 1).out('Direct').properties('uid')

Any ideas how it can be improved? query.batch is enabled.
Jnausgraph version 0.6, backend storage: Cassandra.

[porunov]

We have queries like this: g.V().has('AccountId', 1).out('Direct'). It is executed few milliseconds and returns ~13000 results. However when I try to query the values using one of the follwoing queries it took about 35-45 seconds to complete:
g.V().has('AccountId', 1).out('Direct').values('uid')
g.V().has('AccountId', 1).out('Direct').valueMap()
g.V().has('AccountId', 1).out('Direct').properties('uid')

Any ideas how it can be improved? query.batch is enabled.
Jnausgraph version 0.6, backend storage: Cassandra.

You need to enable multiquery (it will be enabled by default from JanusGraph version 1.0.0).
It will make values step faster.
That said, valueMap is not going to leverage that optimization until we add that optimization step. See opened issue: #2444

The same optimization should be implemented for "has" step which isn't leveraging an index (issue: #3244 )

[vitaliishandra]

Actually we have set storage.parallel-backend-executor-service.core-pool-size=1024 and storage.parallel-backend-ops=true. The result is the same. Regarding the valueMap and values duration differences - it can differ from run to run for us, and in average is the same.
We will check out resource consumption

[porunov]

Actually we have set storage.parallel-backend-executor-service.core-pool-size=1024 and storage.parallel-backend-ops=true. The result is the same. Regarding the valueMap and values duration differences - it can differ from run to run for us, and in average is the same. We will check out resource consumption

It's strange then. Do you know average network latency between JanusGraph and storage? In our case network latency is under 1ms usually but if you have higher latency then it could be the case.

[porunov]

Also, we are using JanusGraph 1.0.0-rc2 version, but we didn't have problems with previous versions, so I'm not sure it's the case. Nevertheless, you could try using 1.0.0-rc2 version just in case

[vitaliishandra]

@porunov thatnk you for avices. We have two separate JanusGrpah services. One communicate with Scylla, another - with Cassandra. The data is identical. And we don't see the issues with Scylla (the described queries took up to 3sec). So we will continue investigate what is wrong with Cassandra configuration

[vitaliishandra]

We found out that we used azurefile storage with StandardLRS type for Cassandra. We changed it to Premium and now queries wtih values/valueMap/properties took up to 1 sec.