Hello.
Currently I'm using the FeatureWriter to update my file geodatabase. It's running at a rate of 18 features per minute, and is comparing 13,998 features against 1.2 million in the dataset being updated.
I'm running a test of this process on a group of features which is a good representation of what a typical update will be in terms of feature quantity.
Is there any way to speed this up?
Thanks.
Best answer by kimo
Does the EntityID have an index? This is critical for performance. Reindex it using ArcGIS indexing tool. It might be useful if the data type of the id is the same so that conversion between integer and string is not being done.
As a comparison, my last update took 10 min for an update of 120,780 updates on a table containing 23M records (200/sec)
Turning off transactions can improve the speed. In my case I see that I did not change the default count of 1000.
My dataset is local, not on a network.
+10Does the EntityID have an index? This is critical for performance. Reindex it using ArcGIS indexing tool. It might be useful if the data type of the id is the same so that conversion between integer and string is not being done.
As a comparison, my last update took 10 min for an update of 120,780 updates on a table containing 23M records (200/sec)
Turning off transactions can improve the speed. In my case I see that I did not change the default count of 1000.
My dataset is local, not on a network.
Does the EntityID have an index? This is critical for performance. Reindex it using ArcGIS indexing tool. It might be useful if the data type of the id is the same so that conversion between integer and string is not being done.
As a comparison, my last update took 10 min for an update of 120,780 updates on a table containing 23M records (200/sec)
Turning off transactions can improve the speed. In my case I see that I did not change the default count of 1000.
My dataset is local, not on a network.
So Transaction Type would be "None"? No problem trying that out, just verifying. I'm not really too familiar yet with how Transactions works, to be honest.
The EntityID field type in the dataset to be updated (File GDB) is written as DOUBLE, but I'm not sure about the updating features. The translations I work on are writing to a network drive.
Thanks!
Does the EntityID have an index? This is critical for performance. Reindex it using ArcGIS indexing tool. It might be useful if the data type of the id is the same so that conversion between integer and string is not being done.
As a comparison, my last update took 10 min for an update of 120,780 updates on a table containing 23M records (200/sec)
Turning off transactions can improve the speed. In my case I see that I did not change the default count of 1000.
My dataset is local, not on a network.
Thanks again.
Thanks @kimo. This is pretty delayed, but I've been using your answer and indexing the feature class works extremely well. According to my last test run, I went from four hours processing to five minutes. That's just crazy.
+10Thanks @kimo. This is pretty delayed, but I've been using your answer and indexing the feature class works extremely well. According to my last test run, I went from four hours processing to five minutes. That's just crazy.
I have just replaced my writer for version 2018.1 and used the fme_db_operation variable to merged the delete/update/add operations into one transformer and the time has been reduced to 1 minute from 20 min, a jump to 1700/sec. Maybe the CSV2 reader has made a difference too, but that is a big improvement. It seems that the changes are cached and then applied in one transaction instead of serially. I turned off transactions.
Title_Memorial 133920
2019-01-07 14:50:05| 74.3| 0.0|STATS |changeset 133920
2019-01-07 14:50:05| 74.3| 0.0|STATS |Total Features Written 267840
2019-01-07 14:50:05| 74.3| 0.0|INFORM|Translation was SUCCESSFUL with 909 warning(s) (267840 feature(s) output)
2019-01-07 14:50:05| 74.3| 0.0|INFORM|FME Session Duration: 1 minute 17.5 seconds. (CPU: 60.1s user, 14.1s system)
2019-01-07 14:50:05| 74.3| 0.0|INFORM|END - ProcessID: 16104, peak process memory usage: 375528 kB, current process memory usage: 117120 kBI have just replaced my writer for version 2018.1 and used the fme_db_operation variable to merged the delete/update/add operations into one transformer and the time has been reduced to 1 minute from 20 min, a jump to 1700/sec. Maybe the CSV2 reader has made a difference too, but that is a big improvement. It seems that the changes are cached and then applied in one transaction instead of serially. I turned off transactions.
Title_Memorial 133920
2019-01-07 14:50:05| 74.3| 0.0|STATS |changeset 133920
2019-01-07 14:50:05| 74.3| 0.0|STATS |Total Features Written 267840
2019-01-07 14:50:05| 74.3| 0.0|INFORM|Translation was SUCCESSFUL with 909 warning(s) (267840 feature(s) output)
2019-01-07 14:50:05| 74.3| 0.0|INFORM|FME Session Duration: 1 minute 17.5 seconds. (CPU: 60.1s user, 14.1s system)
2019-01-07 14:50:05| 74.3| 0.0|INFORM|END - ProcessID: 16104, peak process memory usage: 375528 kB, current process memory usage: 117120 kBWow, even better. Thanks so much for the info on this @kimo.
Enter your E-mail address. We'll send you an e-mail with instructions to reset your password.
