COLUMN STORE INDEX vs CLUSTERED INDEX..Which one to use? - sql

I’m trying to evaluate which type of indexes to use on our tables in the SQL Server 2014 data mart, which we are using to power our OLAP cube in SSAS. I have read the documentation on MSDN and still a bit unclear which is the right strategy for our use case with the ultimate goal of speeding up the queries on SQL Server that the cube issues when people browse the cube.
I have the tables related to each other as shown in the following snow flake dimensional model. The majority of the calculations that we are going to do in the cube, is COUNT DISTINCT of the users (UserInfoKey) based on different combination of dimensions (both filters and pivots). Keeping that in mind, what would the SQL experts suggest I do in terms of creating indexes on the tables?. I have the option of creating COLUMN STORE INDEXES on all my tables (partitioned by the HASH of primary keys) or create the regular primary keys (clustered indexes) on all my tables. Which one is better for my scenario? From my understanding the cube will be doing a lot of joins and groupby’s under the covers based on the dimensions selected by the user.
I tried both versions with some sample data and the performance isn’t that different in both cases. Now before I do the same experiment with real data (it’s going to take a lot of time to produce the real data and load it into our data mart), I wanted to check with the experts about their suggestions.
We are also evaluating if we should use PDW( Parallel Datawarehouse) as our data mart instead of vanilla SQL Server 2014.
Just to give an idea on the scale of data we are dealing with
The 2 largest tables are
ActivityData fact table : 784+ million rows
DimUserInfo dimension table: 30 + million rows
Any help or pointers are appreciated

Related

Star Schema Query - do you have to include all dimensions(joins) in the query?

Hi I have a question regarding star schema query in MS SQL datawarehouse.
I have a fact table and 8 dimensions. And I am confused, to get the metrics from Fact, do we have to join all dimensions with Fact, even though I am not getting data from them? Is this required for the right metrics?
My fact table is huge, so that's why I am wondering for performance purposes and the right way to query.
Thanks!
No you do not have to join all 8 dimensions. You only need to join the dimensions that contain data you need for analyzing the metrics in the fact table. Also to increase performance make sure to only include columns from the dimension table that are needed for the analysis. Including all columns from the dimensions you join will decrease performance.
It is not necessary to include all the dimensions. Indeed, while exploring fact tables, It is very important to have the possibility to select only some dimensions to join and drop the others. The performance issues must not be an excuse to give up on this capability.
You have a bunch of different techniques to solve performance issues depending on the database you are using. Some common ways :
aggregate tables : it is one of the best way to solve performance issues. If you have a huge fact table, you can create an aggregate version of it, using only the most frequently queried columns. This way, it should be much smaller. Then, users (or the adhoc query application) has to use the aggregrate table instead of the original fact table when this is possible. The good news is that most databases know how to automatically manage aggregate tables (materialized views for example). Queries that initially target the original fact table, are transparently redirected to the aggregate table whenever possible.
indexing : bitmap indexing for example can be an efficient way to increase performance in a star schema fact table.

100 Join SQL query

I'm looking to essentially have a centralized table with a number of lookup tables that surround it. The central table is going to be used to store 'Users' and the lookup tables will be user attributes, like 'Religion'. The central table will store an Id, like ReligionId, and the lookup table would contain a list of religions.
Now, I've done a lot of digging into this and I've seen many people comment saying that a UserAttribute table might be the best way to go, essentially using an EAV pattern. I'm not looking to do this. I realize that my strategy will be join-heavy and that's why I ask this question here. I'm looking for a way to optimize those joins.
If the table has 100 lookup tables, how could it be optimized to be faster than just doing a massive 100 table inner join? Some ideas come to mind like using many smaller joins, sub-selects and views. I'm open to anything, including a combination of these strategies. Again, just to note, I'm not looking to do anything that's EAV-related. I need the lookup tables for other reasons and I like normalized data.
All suggestions considered!
Here's a visual look:
Edit: Is this insane?
Optimization techniques will likely depend on the size of the center table and intended query patterns. This is very similar to what you get in data warehousing star schemas, so approaches from that paradigm may help.
For one, ensuring the size of each row is absolutely as small as possible. Disk space may be cheap, but disk throughput, memory, and CPU resources are potential bottle necks. You want small rows so that it can read them quickly and cache as much as possible in memory.
A materialized/indexed view with the joins already performed allows the joins to essentially be precomputed. This may not work well if you are dealing with a center table that is being written to alot or is very large.
Anything you can do to optimize a single join should be done for all 100. Appropriate indexes based on the selectivity of the column, etc.
Depending on what kind of queries you are performing, then other techniques from data warehousing or OLAP may apply. If you are doing lots of group by's then this is likely an area to look in to. Data warehousing techniques can be applied within SQL Server with no additional tooling.
Ask yourself why so many attributes are being queried and how they are being presented? For most analysis it is not necessary to join with lookup tables until the final step where you materialize a report, at which time you may only have grouped by on a subset of columns and thus only need some of the lookup tables.
Group By's generally should be able to group on the lookup Id's without needing the text/description from the lookup table so a join is not necessary. If your lookups have other information relevant to the query at hand then consider denormalizing it into the central table to eliminate the join and/or make that discreet value its own lookup, essentially splitting the existing lookup ID into another ID.
You could implement a master code table that combines the code tables into a single table with a CodeType column. This is not the same as EAV because you'd still have a column in the center table for each code type and a join for each, where as EAV is usually used to normalize out an arbitrary number of attributes. (Note: I personally hate master code tables.)
Lastly, consider normalization the center table if you are not doing data warehousing.
Are there lots of null values in certain lookupId columns? Is the table sparse? This is an indication that you can pull some columns out into a 1 to 1/0 relationships to reduce the size of the center table. For example, a Person table that includes address information can have a PersonAddress table pulled out of it.
Partitioning the table may improve performance if there's a large number of rows and you can determine that certain rows, perhaps with a certain old datetime from couple years in the past, would rarely be queried.
Update: See "Ask yourself why so many attributes are being queried and how they are being presented?" above. Consider a user wants to know number of sales grouped by year, department, and product. You should have id's for each of these so you can just group by those IDs on the center table and in an outer query join lookups for only what columns remain. This ensures the aggregation doesn't need to pull in unnecessary information from lookups that aren't needed anyway.
If you aren't doing aggregations, then you probably aren't querying large numbers of records at a time, so join performance is less of a concern and should be taken care of with appropriate indexes.
If you're querying large numbers of records at a time pulling in all information, I'd look hard at the business case for this. No one sits down at their desk and opens a report with a million rows and 100 columns in it and does anything meaningful with all of that data, that couldn't be accomplished in a better way.
The only case for such a query be a dump of all data intended for export to another system, in which case performance shouldn't be as much as a concern as it can be scheduled overnight.
Since you are set on your way. you can consider duplicating data in order to join less times in a similar way to what is done in olap database.
http://en.wikipedia.org/wiki/OLAP_cube
With that said I don't think this is the best way to do it if you have 100 properties.
Have you tried to export it to Microsoft Excel Power Pivot with Power Query? you can make fast data analysis with pretty awsome ways to show it with Power view video sample

Is it possible to column partitioning in SQL Server

The size of each record of table is a performance parameter. that means if size of record was small SQL Server fetch more records in each read from physical hard.
In most of our queries we not use all column of table, and may be some column use only in specific query. Is it possible for we partitioning columns of each table to have better performance.
I use SQL Server 2008 R2.
Thank you.
True column level partitioning comes with column oriented storage, see Inside the SQL Server 2012 Columnstore Indexes, but that is available only in SQL Server 2012 and addresses specific BI workloads, not general SQL Server apps.
In row oriented storage the vertical partitioning is actually another name for designing proper covering indexes. If the engine has an alternative narrow index it will use it instead of the base table, when possible.
The last alternative, manually splinting the table and joining the vertical 'shards' in queries (or defining joining views, same thing) is usually ill advised and seldom pays off.
At the moment with SQL Server 2008, you cannot partition tables horizontally. If you have a large number of columns, you would need to chop it into horizontal chunk tables that share a common key and then skin them with an update-able view to give the illusion of one very wide table.
If there are just a few large columns (e.g. VARCHAR(1000)), you can normalize your data into unique value tables.
The one exception to the no column partitioning rule are character columns declared as max (varchar(max), for example).
These are stored on a separate data page. I believe this page is not read in unless the column is referenced in the query. If I am wrong, I'am sure more knowledge people will correct me.

What is the most scalable design for this table structure

DataColumn, DataColumn, DateColumn
Every so often we put data into the table via date.
So everything seems great at first, but then I thought: What happens when there are a million or billion rows in the table? Should I be breaking up the tables by date? This way the query performance will never degrade? How do people deal with this sort of thing?
You can use partitioned tables starting with SQL 2K5: Partitioned Tables
This way you gain the benefits of keeping the logical design pure while being able to move old data into a different file group.
You should not break your tables because of data. Instead, you should worry about your indexes, normalization and so on.
Update
A little deeper explanation. Let's suppose you have a table with a million records. If you have different dates on [DateColumn], your greatest ally will be the indexes that work with the [DateColumn]. Then you make sure your queries always filter by at least [DateColumn].
This way, you will be fine.
This easily qualifies as premature optimization, which is tough to achieve in db design IMHO, because optimization is/should be closer to the surface in data modeling.
But all you need to do is create an index on the DateColumn field. An index is actually a much better performance solution than any kind of table splitting/breaking up and keeps your design and therefore all of you programming much simpler. (And you can decide to use partitioning w/o affecting your design in the future if it helps.)
Sounds like you could use a history table. If you are mostly going to query the current date's data, then migrate the old data to the history table and your main table will not grow so much.
If I understand you question correctly, you have a table with some data and a date. Your question is -- will I see improved performance if I make a new table say, every year. This way the queries will never have to look at more than one years worth of data.
This is wrong. Instead what you should do is set the date field as an index. The server will be able to give you the performance gain you need if it is an index.
If you don't do this your program's logic will get crazy and ultimately slow down your system.
Keep it simple.
(NB - There are some advanced partitioning features you can make use of, but these can be layered in later if needed -- it is unlikely you will need these features but the simple design should be able to migrate to them if needed.)
When tables and indexes become very
large, partitioning can help by
partitioning the data into smaller,
more manageable sections.
Microsoft SQL Server 2005 allows you
to partition your tables based on
specific data usage patterns using
defined ranges or lists. SQL Server
2005 also offers numerous options for
the long-term management of
partitioned tables and indexes by the
addition of features designed around
the new table and index structure.
Furthermore, if a large table exists
on a system with multiple CPUs,
partitioning the table can lead to
better performance through parallel
operations.
You might need considering the
following too: In SQL Server 2005,
related tables (such as Order and
OrderDetails tables) that are
partitioned to the same partitioning
key and the same partitioning function
are said to be aligned. When the
optimizer detects that two partitioned
and aligned tables are joined, SQL
Server 2005 can join the data that
resides on the same partitions first
and then combine the results. This
allows SQL Server 2005 to more
effectively use multiple-CPU
computers.
Read about Partitioned Tables and Indexes in SQL Server 2005

To aggregate or not to aggregate, that is the database schema design question

If you're doing min/max/avg queries, do you prefer to use aggregation tables or simply query across a range of rows in the raw table?
This is obviously a very open-ended question and there's no one right answer, so I'm just looking for people's general suggestions. Assume that the raw data table consists of a timestamp, a numeric foreign key (say a user id), and a decimal value (say a purchase amount). Furthermore, assume that there are millions of rows in the table.
I have done both and am torn. On one hand aggregation tables have given me significantly faster queries but at the cost of a proliferation of additional tables. Displaying the current values for an aggregated range either requires dropping entirely back to the raw data table or combining more fine grained aggregations. I have found that keeping track in the application code of which aggregation table to query when is more work that you'd think and that schema changes will be required, as the original aggregation ranges will invariably not be enough ("But I wanted to see our sales over the last 3 pay periods!").
On the other hand, querying from the raw data can be punishingly slow but lets me be very flexible about the data ranges. When the range bounds change, I simply change a query rather than having to rebuild aggregation tables. Likewise the application code requires fewer updates. I suspect that if I was smarter about my indexing (i.e. always having good covering indexes), I would be able to reduce the penalty of selecting from the raw data but that's by no means a panacea.
Is there anyway I can have the best of both worlds?
We had that same problem and ran into the same issues you ran into. We ended up switching our reporting to Analysis Services. There is a learning curve with MDX and Analysis services itself, but it's been great. Some of the benefits we have found are:
You have a lot of flexibility for
querying any way you want. Before we
had to build specific aggregates,
but now one cube answers all our
questions.
Storage in a cube is far smaller
than the detailed data.
Building and processing the cubes
takes less time and produces less
load on the database servers than
the aggregates did.
Some CONS:
There is a learning curve around
building cubes and learning MDX.
We had to create some tools to
automate working with the cubes.
UPDATE:
Since you're using MySql, you could take a look at Pentaho Mondrian, which is an open source OLAP solution that supports MySql. I've never used it though, so I don't know if it will work for you or not. Would be interested in knowing if it works for you though.
It helps to pick a good primary key (ie [user_id, used_date, used_time]). For a constant user_id it's then very fast to do a range-condition on used_date.
But as the table grows, you can reduce your table-size by aggregating to a table like [user_id, used_date]. For every range where the time-of-day doesn't matter you can then use that table. An other way to reduce the table-size is archiving old data that you don't (allow) querying anymore.
I always lean towards raw data. Once aggregated, you can't go back.
Nothing to do with deletion - unless there's the simplest of aggregated data sets, you can't accurately revert/transpose the data back to raw.
Ideally, I'd use a materialized view (assuming that the data can fit within the constraints) because it is effectively a table. But MySQL doesn't support them, so the next consideration would be a view with the computed columns, or a trigger to update an actual table.
Long history question, for currently, I found this useful, answered by microstrategy engineer
BTW, another already have solutions like (cube.dev/dremio) you don't have to do by yourself.