Netezza/PureData - Bad distribution key chosen in HASH JOIN - performance

I am using Netezza/Pure Data for a query. I have a INNER JOIN (which became a HASH JOIN) on two columns A and B. A is a column that has good distribution and B is a column that has bad distribution. For some reason, my query plan always uses B instead A as the distribution key for that JOIN, which causes immense performance issue.
GENERATE STATISTICS does help alleviate this issue, but due to performance constraints, it is not feasiable to GENERATE STATISTICS before every query. I do it before a batch run but not in between each query within a batch.
In a nutshell, the source tables have good distributions but when I join them, they choose a bad distribution key (which is actually never used as a distribution column at all in the sources).
So my question is, what are some good ways to influence the choice of distribution key in a JOIN without doing GENERATE STATISTICS. I've tried changing around the distribution columns of the source tables but that didn't do much even if I make sure all the skew's are less than 0.5.

You could create a temp table and force the distribution so that they both align, this should expedite the join

The workaround is to force exhaustive planner to be used.
set num_star_planner_rels = X; -- Set X to very high.
According to IBM Netezza team, queries with more than 7 entities (# of tables) will use a greedy query planner called "Snowflake". At 7 or less entities, it will use the brute force approach to find the best plan.
The trade off is that exhaustive search is very expensive for large number of entities.

Related

ORGANIZE / DISTRIBTE BY with CREATE TABLE in MPP systems

How is CREATE TABLE statement's performance influenced by ORGANIZE BY / DISTRIBUTE BY clause in MPP systems (Netezza / Teradata / Synapse)?
Also, what key should be picked for distribution in such MPP system?
The short answer is that as long as you choose a ‘distribution’ that enables a very even spread of rows across nodes, then it has little to no effect on insert performance.
On Netezza the ‘organize’ is not enforced while you create the table, not even while you insert/update data. The Groom operation does that on your request later. Side note: When doing CTAS or large inserts you should ‘order by’ in the insert statement if possible.
About choosing ‘distribution’ columns:
always ensure a very even spread (if not possible: RANDOM is better)
and never use more than one column
and choose one you plan to do a lot of ‘equal join’ on (a LOT)
About choosing ‘organization’ columns
only consider columns you plan to do a lot of ‘simple where’ clauses against (=,<,>, LIKE)
tend towards those with few distinct values
and ‘time’ columns are always a good guess
Hint: sometimes you get a ‘free’ organize effect on a column because it is closely related to another column that you already organize on.
Example: If the create_date on average is less than 30 days away from the end_date in a table with 5 years of data, then you will have that effect

Vertica query optimization

I want to optimize a query in vertica database. I have table like this
CREATE TABLE data (a INT, b INT, c INT);
and a lot of rows in it (billions)
I fetch some data using whis query
SELECT b, c FROM data WHERE a = 1 AND b IN ( 1,2,3, ...)
but it runs slow. The query plan shows something like this
[Cost: 3M, Rows: 3B (NO STATISTICS)]
The same is shown when I perform explain on
SELECT b, c FROM data WHERE a = 1 AND b = 1
It looks like scan on some part of table. In other databases I can create an index to make such query realy fast, but what can I do in vertica?
Vertica does not have a concept of indexes. You would want to create a query specific projection using the Database Designer if this is a query that you feel is run frequently enough. Each time you create a projection, the data is physically copied and stored on disk.
I would recommend reviewing projection concepts in the documentation.
If you see a NO STATISTICS message in the plan, you can run ANALYZE_STATISTICS on the object.
For further optimization, you might want to use a JOIN rather than IN. Consider using partitions if appropriate.
Creating good projections is the "secret-sauce" of how to make Vertica perform well. Projection design is a bit of an art-form, but there are 3 fundamental concepts that you need to keep in mind:
1) SEGMENTATION: For every row, this determines which node to store the data on, based on the segmentation key. This is important for two reasons: a) DATA SKEW -- if data is heavily skewed then one node will do too much work, slowing down the entire query. b) LOCAL JOINS - if you frequently join two large fact tables, then you want the data to be segmented the same way so that the joined records exist on the same nodes. This is extremely important.
2) ORDER BY: If you are performing frequent FILTER operations in the where clause, such as in your query WHERE a=1, then consider ordering the data by this key first. Ordering will also improve GROUP BY operations. In your case, you would order the projection by columns a then b. Ordering correctly allows Vertica to perform MERGE joins instead of HASH joins which will use less memory. If you are unsure how to order the columns, then generally aim for low to high cardinality which will also improve your compression ratio significantly.
3) PARTITIONING: By partitioning your data with a column which is frequently used in the queries, such as transaction_date, etc, you allow Vertica to perform partition pruning, which reads much less data. It also helps during insert operations, allowing to only affect one small ROS container, instead of the entire file.
Here is an image which can help illustrate how these concepts work together.

Oracle: Possible to replace a standard non-unique single column index with an unique combined index?

I´m currently working on optimzing my database schema in regards of index structures. As I´d like to increase my DDL performance I´m searching for potential drop candidates on my Oracle 12c system. Here´s the scenario in which I don´t know what the consequences for the query performance might be if I drop the index.
Given two indexes on the same table:
- non-unique, single column index IX_A (indexes column A)
- unique, combined index UQ_AB (indexes column A, then B)
Using index monitoring I found that the query optimizer didn´t choose UQ_AB, but only IX_A (probably because it´s smaller and thus faster to read). As UQ_AB contains column A and additionally column B I´d like to drop IX_A. Though I´m not sure if I get any performance penalties if I do so. Does the higher selectivity of the combined unique index have any influence on the execution plans?
It could do, though it's quite likely to be minor (usually). Of course it depends on various things, for example how large the values in column B are.
You can look at various columns in USER_INDEXES to compare the two indexes, such as:
BLEVEL: tells you the "height" of the index tree (well, height is BLEVEL+1)
LEAF_BLOCKS: how many data blocks are occupied by the index values
DISTINCT_KEYS: how "selective" the index is
(You need to have analyzed the table first for these to be accurate). That will give you an idea of how much work Oracle needs to do to find a row using the index.
Of course the only way to really be sure is to benchmark and compare timings or even trace output.

A join operation using Hadoop MapReduce

How to take a join of two record sets using Map Reduce ? Most of the solutions including those posted on SO suggest that I emit the records based on common key and in the reducer add them to say a HashMap and then take a cross product. (eg. Join of two datasets in Mapreduce/Hadoop)
This solution is very good and works for majority of the cases but in my case my issue is rather different. I am dealing with a data which has got billions of records and taking a cross product of two sets is impossible because in many cases the hashmap will end up having few million objects. So I encounter a Heap Space Error.
I need a much more efficient solution. The whole point of MR is to deal with very high amount of data I want to know if there is any solution that can help me avoid this issue.
Don't know if this is still relevant for anyone, but I facing a similar issue these days. My intention is to use a key-value store, most likely Cassandra, and use it for the cross product. This means:
When running on a line of type A, look for the key in Cassandra. If exists - merge A records into the existing value (B elements). If not - create a key, and add A elements as value.
When running on a line of type B, look for the key in Cassandra. If exists - merge B records into the existing value (A elements). If not - create a key, and add B elements as value.
This would require additional server for Cassandra, and probably some disk space, but since I'm running in the cloud (Google's bdutil Hadoop framework), don't think it should be much of a problem.
You should look into how Pig does skew joins. The idea is that if your data contains too many values with the same key (even if there is no data skew) , you can create artificial keys and spread the key distribution. This would make sure that each reducer gets less number of records than otherwise. For e.g. if you were to suffix "1" to 50% of your key "K1" and "2" the other 50% you will end with half the records on the reducer one (1K1) and the other half goes to 2K2.
If the distribution of the keys values are not known before hand you could some kind of sampling algorithm.

Oracle 'pseduo-fact' view

Assumptions:
I have a number of tables comprised of facts and foreign keys ('dimensional' and 'key-value' type). For example, ENCOUNTER:
ID - primary key
dimensions
LOCATION_ID
PATIENT_ID
key-value
TYPE_ID
STATUS_ID
PATIENT_CLASS_ID
DISPOSITION_ID
...
facts
ADMISSION_DATE
DISCHARGE_DATE
...
I don't have the option to create a data warehouse
I would like to simplify the data structure for reporting
My approach is to create a number of pseudo-dimensional views ('D_LOCATION' based on the DEPARTMENT and LOCATION tables) and pseudo-fact views ('F_ENCOUNTER' based on ENCOUNTER table). In the pseudo-fact view, I would JOIN the key-value tables (e.g. STATUS, PATIENT_CLASS) to the fact table to include the name fields (e.g. STATUS.NAME, PATIENT_CLASS.NAME).
Questions:
If a query selects a subset of all of the fields from F_ENCOUNTER (i.e. not all of the key-value.name fields), is the Oracle 10g optimizer smart enough to exclude some of the key-value table joins (i.e. the ones that aren't included in the query)?
Is there anything that I can do to optimize this architecture (other than indices)
Is there another approach?
** edit **
Goals (in order of importance):
reduce query complexity; increase query consistency; decrease report-development time
optimize query-processing
minimize administrator burden
decrease storage
One optimization suggestion is not to use key-value pair tables. The concept of a Dimension table is that each record should contain all information about that concept without needing to join to normalized tables - i.e. turning a star schema into a snowflake schema.
While values might be repeated across dimension table records, it has the advantage of fewer joins in your reporting queries. Denormalizing tables in this way might seem counter intuitive but where performance is paramount it is usually the best solution.
I don't believe Oracle would exclude any joins done in the view, because the joins can impact the number of rows returned. (As when an inner join fails to match any rows, making the whole result set empty.)
What are the goals of your optimization? Query speed? query simplicity? storage efficiency? If you can sacrifice storage efficiency for better query performance, then replace the key-value references with the values themselves (TYPE_NAME instead of TYPE_ID, PATIENT_CLASS_NAME instead of PATIENT_CLASS_ID, etc.).
[Edit:] If the original architecture cannot be modified, consider using a materialized view. It would essentially pre-compute the joins and store the result set, giving you speedy query time at the cost of extra storage space and possibly-not-fresh data. You can control the latter by specifying an appropriate refresh policy. See http://en.wikipedia.org/wiki/Materialized_view and http://download.oracle.com/docs/cd/B10500_01/server.920/a96520/mv.htm for further details.

Resources