I am working on an application which needs to store financial transactions for an account.
This data will then need to be queried in a number of ways. I'll need to list individual transactions, monthly totals by category, for example. I'll also need to show a monthly summary with opening / closing balances.
As I see it, I could approach this in the following ways:
From the point of view of database consistency and normalisation, this could be modelled as a simple list of transactions. Balances may then be calculated in the application by adding up every transaction from the beginning of time to the date of the balance you wish to display.
A slight variation on this would be to model the data in the same way, but calculate the balances in a stored procedure on the database server. Appreciate that this isn't hugely different to #1 - both of these issues will perform slower as more data gets added to the system.
End of month balances could be calculated and stored in a separate table (possibly updated by triggers). I don't really like this approach from a data consistency point of view, but it should scale better.
I can't really decide which way to go with this. Should I start with the 'purest' data model and only worry about performance when it becomes an issue? Should I assume performance will become a problem and plan for it from day one? Is there another option which I haven't thought of which would solve the issue better?
I would look at it like this, the calculations are going to take longer and longer and that majority of the monthly numbers before the previous 2-3 months will not be changing. This is a performance problem that has a 100% chance of happening as financial date will grow every month. Therefore looking at a solution in the design phase is NOT premature optimization, it is smart design.
I personally am in favor of only calculating such totals when they need to be calculated rather than every time they are queried. Yes the totals should be updated based on triggers on the table which will add a slight overhead to inserts and deletes. They will make the queries for selects much faster. In my experience users tend to be more tolerant of a slightly longer action query than a much longer select query. Overall this is a better design for this kind of data than a purely normalized model as long as you do the triggers correctly. In the long run, only calculating numbers that have changed will take up far less server resources.
This model will maintain data integrity as long as all transactions go through the trigger. The biggest culprit in that is usually data imports which often bypass triggers. If you do those kinds of imports make sure they have code that mimics the trigger code. Also make sure the triggers are for insert/update and delete and that they are tested using multiple record transactions not just against single records.
The other model is to create a data warehouse that populates on a schedule such as nightly. This is fine if the data can be just slightly out of date. If the majority of queries of this consolidated data will be for reporting and will not involve the current month/day so much then this will work well and you can do it in an SSIS package.
Related
I think the question in the title speaks it all and is general.
I can give a concrete example as well:
I have tagged articles and want to find similar articles with the tags associated with them.
The score function will look at two articles and count the number of tags in common.
Since the score is not stored anywhere, I'll have to calculate the score everytime I need to find similar articles given an article.
But this is too expensive.
What is the common work-around to this kind of problem in general?
Is there a better approach for my specific tag problem? (e.g. solr's moreLikeThis)
edit
I'm using postgres, if that matters.
I'm looking for a general solution that people used successfully, such as you should batch calculate the score and save it somewhere and etc...
The answer will vary wildly by database product and version. For example, in some database products, it may be the case that a view or an indexed view might be faster than the more common solution...
Typically the way to handle a situation like this is by precalculating the result. You can do that in a handful of ways:
a. You can use something like triggers (added in the SQL 99 standard) that update the counts as rows are added, updated or removed from the source table. In this solution, you are making a (presumably) small sacrifice on inserts, updates and deletes of the source table in order to make significant gains in retrieving the information.
b. You can use a data warehouse where you accept some level of latency of live data to reported data. That means you accept that the data queried from the data warehouse will be stale by some accepted number of minutes, hours, days, or weeks. The data warehouse works by periodically querying the live OLTP (Online Transaction Processing) data and updates the OLAP (Online Analytical Processing) database which contains the precalculated results. You then run your reports off the OLAP data or a combination of OLTP and OLAP data. A formal database warehouse isn't required to achieve the equivalent results. You could write a procedure which is executed on a timer that updates a table periodically with updated results.
On sites like SO, I'm sure it's absolutely necessary to store as much aggregated data as possible to avoid performing all those complex queries/calculations on every page load. For instance, storing a running tally of the vote count for each question/answer, or storing the number of answers for each question, or the number of times a question has been viewed so that these queries don't need to be performed as often.
But does doing this go against db normalization, or any other standards/best-practices? And what is the best way to do this, e.g., should every table have another table for aggregated data, should it be stored in the same table it represents, when should the aggregated data be updated?
Thanks
Storing aggregated data is not itself a violation of any Normal Form. Normalization is concerned only with redundancies due to functional dependencies, multi-valued dependencies and join dependencies. It doesn't deal with any other kinds of redundancy.
The phrase to remember is "Normalize till it hurts, Denormalize till it works"
It means: normalise all your domain relationships (to at least Third Normal Form (3NF)). If you measure there is a lack of performance, then investigate (and measure) whether denormalisation will provide performance benefits.
So, Yes. Storing aggregated data 'goes against' normalisation.
There is no 'one best way' to denormalise; it depends what you are doing with the data.
Denormalisation should be treated the same way as premature optimisation: don't do it unless you have measured a performance problem.
Too much normalization will hurt performance so in the real world you have to find your balance.
I've handled a situation like this in two ways.
1) using DB2 I used a MQT (Materialized Query Table) that works like a view only it's driven by a query and you can schedule how often you want it to refresh; e.g. every 5 min. Then that table stored the count values.
2) in the software package itself I set information like that as a system variable. So in Apache you can set a system wide variable and refresh it every 5 minutes. Then it's somewhat accurate but your only running your "count(*)" query once every five minutes. You can have a daemon run it or have it driven by page requests.
I used a wrapper class to do it so it's been while but I think in PHP was was as simple as:
$_SERVER['report_page_count'] = array('timeout'=>1234569783, 'count'=>15);
Nonetheless, however you store that single value it saves you from running it with every request.
All,
Looking for some guidance on an Oracle design decision I am currently trying to evaluate:
The problem
I have data in three separate schemas on the same oracle db server. I am looking to build an application that will show data from all three schemas, however the data that is shown will be based on real time sorting and prioritisation rules that is applied to the data globally (i.e.: based on the priority weightings applied I may pull back data from any one of the three schemas).
Tentative Solution
Create a VIEW in the DB which maintains logical links to the relevant columns in the three schemas, write a stored procedure which accepts parameterised priority weightings. The application subsequently calls the stored procedure to select the ‘prioritised’ row from the view and then queries the associated schema directly for additional data based on the row returned.
I have concerns over performance where the data is being sorted/ prioritised upon each query being performed but cannot see a way around this as the prioritisation rules will change often. We are talking of data sets in the region of 2-3 million rows per schema.
Does anyone have alternative suggestions on how to provide an aggregated and sorted view over the data?
Querying from multiple schemas (or even multiple databases) is not really a big deal, even inside the same query. Just prepend the table name with the schema you are interested in, as in
SELECT SOMETHING
FROM
SCHEMA1.SOME_TABLE ST1, SCHEMA2.SOME_TABLE ST2
WHERE ST1.PK_FIELD = ST2.PK_FIELD
If performance becomes a problem, then that is a big topic... optimal query plans, indexes, and your method of database connection can all come into play. One thing that comes to mind is that if it does not have to be realtime, then you could use materialized views (aka "snapshots") to cache the data in a single place. Then you could query that with reasonable performance.
Just set the snapshots to refresh at an interval appropriate to your needs.
It doesn't matter that the data is from 3 schemas, really. What's important to know is how frequently the data will change, how often the criteria will change, and how frequently it will be queried.
If there is a finite set of criteria (that is, the data will be viewed in a limited number of ways) which only change every few days and it will be queried like crazy, you should probably look at materialized views.
If the criteria is nearly infinite, then there's no point making materialized views since they won't likely be reused. The same holds true if the criteria itself changes extremely frequently, the data in a materialized view wouldn't help in this case either.
The other question that's unanswered is how often the source data is updated, and how important is it to have the newest information. Frequently updated source day can either mean a materialized view will get "stale" for some duration or you may be spending a lot of time refreshing the materialized views unnecessarily to keep the data "fresh".
Honestly, 2-3 million records isn't a lot for Oracle anymore, given sufficient hardware. I would probably benchmark simple dynamic queries first before attempting fancy (materialized) view.
As others have said, querying a couple of million rows in Oracle is not really a problem, but then that depends on how often you are doing it - every tenth of a second may cause some load on the db server!
Without more details of your business requirements and a good model of your data its always difficult to provide good performance ideas. It usually comes down to coming up with a theory, then trying it against your database and accessing if it is "fast enough".
It may also be worth you taking a step back and asking yourself how accurate the results need to be. Does the business really need exact values for this query or are good estimates acceptable
Tom Kyte (of Ask Tom fame) always has some interesting ideas (and actual facts) in these areas. This article describes generating a proper dynamic search query - but Tom points out that when you query Google it never tries to get the exact number of hits for a query - it gives you a guess. If you can apply a good estimate then you can really improve query performance times
I work in a company that uses single table Access database for its outbound cms, which I moved to a SQL server based system. There's a data list table (not normalized) and a calls table. This has about one update per second currently. All call outcomes along with date, time, and agent id are stored in the calls table. Agents have a predefined set of records that they will call each day (this comprises records from various data lists sorted to give an even spread throughout their set). Note a data list record is called once per day.
In order to ensure speed, live updates to this system are stored in a duplicate of the calls table fields in the data list table. These are then copied to the calls table in a batch process at the end of the day.
The reason for this is not obviously the speed at which a new record could be added to the calls table live, but when the user app is closed/opened and loads the user's data set again I need to check which records have not been called today - I would need to run a stored proc on the server that picked the last most call from the calls table and check if its calldate didn't match today's date. I believe a more expensive query than checking if a field in the data list table is NULL.
With this setup I only run the expensive query at the end of each day.
There are many pitfalls in this design, the main limitation is my inexperience. This is my first SQL server system. It's pretty critical, and I had to ensure it would work and I could easily dump data back to access db during a live failure. It has worked for 11 months now (and no live failure, less downtime than the old system).
I have created pretty well normalized databases for other things (with far fewer users), but I'm hesitant to implement this for the calling database.
Specifically, I would like to know your thoughts on whether the duplication of the calls fields in the data list table is necessary in my current setup or whether I should be able to use the calls table. Please try and answer this from my perspective. I know you DBAs may be cringing!
Redesigning an already working Database may become the major flaw here. Rather try to optimize what you have got running currently instead if starting from scratch. Think of indices, referential integrity, key assigning methods, proper usage of joins and the like.
In fact, have a look here:
Database development mistakes made by application developers
This outlines some very useful pointers.
The thing the "Normalisation Nazis" out there forget is that database design typically has two stages, the "Logical Design" and the "Physical Design". The logical design is for normalisation, and the physical design is for "now lets get the thing working", considering among other things the benefits of normalisation vs. the benefits of breaking nomalisation.
The classic example is an Order table and an Order-Detail table and the Order header table has "total price" where that value was derived from the Order-Detail and related tables. Having total price on Order in this case still make sense, but it breaks normalisation.
A normalised database is meant to give your database high maintainability and flexibility. But optimising for performance is one of the considerations that physical design considers. Look at reporting databases for example. And don't get me started about storing time-series data.
Ask yourself, has my maintainability or flexibility been significantly hindered by this decision? Does it cause me lots of code changes or data redesign when I change something? If not, and you're happy that your design is working as required, then I wouldn't worry.
I think whether to normalize it depends on how much you can do, and what may be needed.
For example, as Ian mentioned, it has been working for so long, is there some features they want to add that will impact the database schema?
If not, then just leave it as it is, but, if you need to add new features that change the database, you may want to see about normalizing it at that point.
You wouldn't need to call a stored procedure, you should be able to use a select statement to get the max(id) by the user id, or the max(id) in the table, depending on what you need to do.
Before deciding to normalize, or to make any major architectural changes, first look at why you are doing it. If you are doing it just because you think it needs to be done, then stop, and see if there is anything else you can do, perhaps add unit tests, so you can get some times for how long operations take. Numbers are good before making major changes, to see if there is any real benefit.
I would ask you to be a little more clear about the specific dilemma you face. If your system has worked so well for 11 months, what makes you think it needs any change?
I'm not sure you are aware of the fact that "Database design fundamentals" might relate to "logical database design fundamentals" as well as "physical database design fundamentals", nor whether you are aware of the difference.
Logical database design fundamentals should not (and actually cannot) be "sacrificed" for speed precisely because speed is only determined by physical design choices, the prime desision factor in which is precisely speed and performance.
I want to maintain last ten years of stock market data in a single table. Certain analysis need only data of the last one month data. When I do this short term analysis it takes a long time to complete the operation.
To overcome this I created another table to hold current year data alone. When I perform the analysis from this table it 20 times faster than the previous one.
Now my question is:
Is this the right way to have a separate table for this kind of problem. (Or we use separate database instead of table)
If I have separate table Is there any way to update the secondary table automatically.
Or we can use anything like dematerialized view or something like that to gain performance.
Note: I'm using Postgresql database.
You want table partitioning. This will automatically split the data between multiple tables, and will in general work much better than doing it by hand.
I'm working on near the exact same issue.
Table partitioning is definitely the way to go here. I would segment by more than year though, it would give you a greater degree of control. Just set up your partitions and then constrain them by months (or some other date). In your postgresql.conf you'll need to turn constraint_exclusion=on to really get the benefit. The additional benefit here is that you can only index the exact tables you really want to pull information from. If you're batch importing large amounts of data into this table, you may get slightly better results a Rule vs a Trigger and for partitioning, I find rules easier to maintain. But for smaller transactions, triggers are much faster. The postgresql manual has a great section on partitioning via inheritance.
I'm not sure about PostgreSQL, but I can confirm that you are on the right track. When dealing with large data volumes partitioning data into multiple tables and then using some kind of query generator to build your queries is absolutely the right way to go. This approach is well established in Data Warehousing, and specifically in your case stock market data.
However, I'm curious why do you need to update your historical data? If you're dealing with stock splits, it's common to implement that using a seperate multiplier table that is used in conjunction with the raw historical data to give an accurate price/share.
it is perfectly sensible to use separate table for historical records. It's much more problematic with separate database, as it's not simple to write cross-database queries
automatic updates - it's a tool for cronjob
you can use partial indexes for such things - they do wonderful job
Frankly, you should check your execution plans and try fixing your queries or indexing before taking more radical steps.
Indexing comes at very little cost (unless you do a lot of insertions) and your existing code will be faster (if you index properly) without modifying it.
Other measures such as partioning come after that...