Forum Discussion
ODBC and Large Data Sets
I think that there is another perspective here around justifying Power BI versus Tableau. Power BI comes with Power Query, which is an incredibly powerful data mashup/ingest/ETL tool. To get anything even remotely equivalent in Tableau you have to purchase something like Alteryx, which every shop that owns Tableau buys (at least the people that I know that own Tableau).
So, the focus with Power Query is not to write SQL Queries for people but rather allow them to do much more powerful data manipulation. In order to do that, it needs a starting point in terms of the data. The mechanism for that working is not geared toward pre-filtering the data but rather working with the data as defined. It allows for the pre-filtering of data via SQL if you really want but is perfectly capable of working without it on large rowsets.
I understand what you are saying but I think it doesn't take into consideration the entire picture. When you say you wish Power BI wrote better queries, it can write queries that are incredibly more powerful and better than Tableau. Tableau out-of-the-box is really incredibly limited in it's query capabilities. It is like working with tinker toys compared to Power Query.
>>It is like working with tinker toys compared to Power Query.
It's funny you say that. As I mentioned I'm a big proponent of Power BI. Unfortunately for me, the tinker toy tool (Tableau) is actaully returning data. Because of the way Power BI writes queries, it's just blowing up. I don't need to look at the entire picture. One tool (Power BI) just flat doesn't work for large data sets via ODBC without hand feeding it data via custom SQL. Even this is pretty hard because if I use a limit clause, Power BI then proceeds to wrap this and aggregate only on my limited number of records. I can then fully develop my report, but I have to remove the limit clause at some point at which time Power BI proceeds to try to download 300M records again thereby choking. Final victory - "tinker toy" Tableau...
Thanks again for your time. I was already pretty sure that there was not a good solution I was just looking for confirmation.
Edit: grammar.
- jsperson8 years agoFrequent Visitor
Here is a bit of background as to what happens when I actually try to use Power BI. I also have a discussion going on the Power Query board. They don't seem to have an answer either. The example below explains what happens when I try to filter and group by in sequence. I've included the SQL generated as well as the M code. This is mainly for future reference. I don't expect anyone to actually diagnose this. At some point though I'd love to talk to the query engine development team...
It is sending the following query when coming into the Power Query editor in preview mode (10 min wait):
SELECT foo1, foo2,...foo74 FROM bar
Note: NO LIMIT CLAUSE! Tableau uses a limit clause and returns in 5 seconds.
Then I ask for only my two attributes and two measures (2-3 minute wait):
SELECT foo1, foo2, foo3, foo4 FROM bar
Again: no limit clause
Then I filter where one measure is not null (another 2-3 minute wait):
SELECT foo1, foo2, foo3, foo4 FROM bar WHERE foo3 IS NOT NULL
Then I do a group by both attributes with SUM on both measures. I can't provide this query from Power BI as it blows up my client before finishing. It's trying to return all 300M records to my client so it's pretty obvious that it did about the same thing as immediately above except without whatever limits it normally uses.
Tableau sends the following, which returns in about 5 seconds:
SELECT foo1, foo2, SUM(foo3) foo3, SUM(foo4) foo4 FROM bar WHERE FOO3 IS NOT NULL GROUP BY foo1, foo2 LIMIT 1000
Here is the M code from Power BI:
let
Source = Odbc.DataSource("dsn=datalake-poc-athena-64", [HierarchicalNavigation=true]),
AwsDataCatalog_Database = Source{[Name="AwsDataCatalog",Kind="Database"]}[Data],
hmda_Schema = AwsDataCatalog_Database{[Name="hmda",Kind="Schema"]}[Data],
hmda_lar_Table = hmda_Schema{[Name="hmda_lar",Kind="Table"]}[Data],
#"Removed Other Columns" = Table.SelectColumns(hmda_lar_Table,{"action_taken_name", "agency_name", "applicant_income_000s", "loan_amount_000s"}),
#"Filtered Rows" = Table.SelectRows(#"Removed Other Columns", each ([applicant_income_000s] <> null)),
#"Grouped Rows" = Table.Group(#"Filtered Rows", {"action_taken_name", "agency_name"}, {{"Income", each List.Sum([applicant_income_000s]), type number}, {"Loan", each List.Sum([loan_amount_000s]), type number}})
in
#"Grouped Rows"- jsperson8 years agoFrequent Visitor
We opened a support case and pursued it as far as possible. At this point there is no solution.
It seems that when working with ODBC sources Power BI just doesn't do much in the way of writing efficient SQL. The behavior when filtering then grouping is to filter at the source, but group on the client. Also the previews do not send limit clauses. Effectively this all means that Power BI can't work with datasets beyond a few thousand records with ODBC.
The development team did accept this as a potential improvement so there is hope that someday this capability will be added. Until then unfortunately I'll have to use a competing product to Power BI.
- SamTrexler7 years ago
Helper IV
I found this old post while trying to move my reports to use Athena/Parquet for a new client. I,too, and having performance issues with large datasets - and utilizing large datasetds is part of the point of going to Athena.
I've done some tracing with the ODBC administrator, and it seems the ODBC drivers retrieve one column at a time for each row - and inherently slow mechanism.
So if Power Query can't effectively fold the M statements into the query, as it does for SQL Server, then the combination of expensive queries with big rowsets and the inherently slow transmission via ODBC, it seems Power BI is not well suited to retrieving information from this big-data storage mechanism.
Has anyone found a solution to this?