Publication performance: is 2+ minutes normal for level 2?

andreslopez

Hi, we are experiencing slow publication times and want to understand if this is expected. Our setup:

Two separate servers (work system and publication)
Network latency between servers: ₀.4ms
Publication log shows: 42 references per record, level 2, 2 GB memory, 2 min 20 sec total

{
"pid": 1565812,
"pfile": "process_-1_2026-05-20_17-04-47_5907638419358285",
"is_running": false,
"data": {
"result": true,
"msg": [
"OK. Request done successfully"
],
"errors": [],
"diffusion_data": [],
"memory": "2062.5 MB",
"last_update_record_response": {
"result": true,
"msg": [
"Record updated section_id: 1. Number of references: 57 in levels: 2"
],
"errors": [],
"class": "diffusion_mysql"
},
"time": "1.966 min"
},
"time": "2026-05-20 17:06:46",
"total_time": "1.986 min",
"update_rate": 1000,
"errors": [],
"fill_buffer": "3622

Alex

It depends on the section you're publishing in. A simple section with few relations and texts is different from a complex section with multiple relations and extensive texts.
If your case involves a simple section with few relations, it's not considered normal.
But, sections like tch1 could be dense, especially if they require extensive thesaurus resolution or contain numerous bibliographic references with full text of the publication (e.g., a 800-page book with millions of characters in one field). In such cases, it's normal.

Publication with a level two resolution, go in deep in this way:

1 main section ----> level0
2 direct relations ----> level 1
3 secondary relations ---> level 2

So, if you have a section with 57 direct relations, and each of these 57 relations has another one, the resolution will resolve 3249 sections. Remove the records that have been changed and add the records that it doesn't have. And if some of this section are dense... yes, it will normal.

In version 7, we are rebuilding the method for resolving diffusion (the model has more than 15 years of history)... to reduce memory usage during the resolution process (by flushing the memory in the resolution loop)... but it is still in the alpha state. Version 7 will be more efficient in this regard.

Best