Considering this case where we have 2 tables. The requirement is to implement a function to select the top 10 records (ordered by some rules) from TABLE_A and TABLE_B, where table_a.id == table_b.a_id == X. There are two options:
Using JOIN to query the SQL;
Making 2 selection queries from db: SELECT * FROM table_a WHERE id = X and SELECT * FROM table_b WHERE a_id = X, fetching 10 records from each query (let's assume the ordering is correct in this case) in memory, then join them in the code (using a for loop and a hashtable or sth like that).
I've heard that JOIN might lower the system performance (was "db performance" here but that was wrong)(see follow up below for reference). Besides, in this case we only queries for 10 results at maximum, which is acceptable to load them all in memory then join them there.
My question is, is there a general guideline in the industry, to say under what circumstances would we recommend using JOIN in database layer instead of doing it in memory, and when to do the opposite?
============
Follow up:
So here's some reason/scenario I've read for "moving JOIN from database layer to
service layer":
If we are joining multiple tables, they will all be locked at once. And if the operation take times and the service requires low response time, it might block other executions;
Hard to maintain in the big system. Changes of the tables that are involved in JOIN might make the query broken.
There might be some historical reason for those complicated systems, that data might be migrated/created in different db (or db systems, say one table in DynamoDB and the other one in Postgres), which makes JOIN in the database layer impossible.
To answer simply, it depends.
Generally, it is preferable to do data operations closer to data, instead of bringing them higher up in the layers and handle data operations. You can see many PL/SQL based implementations, where they do operations closer to data. Languages like PL/SQL(ORACLE) or TSQL(SQL Server) are designed to do complex data operations.
But, if you have an application, which brings data from disparate systems and have to join between them, you have to do them in memory.
If we are joining multiple tables, they will all be locked at once. And if the operation take times and the service requires low response time, it might block other executions;
Readers are not blocking other readers. They have something called sharedlock. Also, once the read operation is over, shared lock is released. As, #TimBiegeleisen, you can create indexes to speed up read operations, based on the need. It is always preferable to read only needed columns(projection), needed rows(filtering).
Hard to maintain in the big system. Changes of the tables that are involved in JOIN might make the query broken.
As long as you are selecting only the needed columns, instead of SELECT *, you should not be having issues. If many changes are coming, you can considering creating SCHEMA BINDING View, to avoid schema changes to the underlying tables.
There might be some historical reason for those complicated systems, that data might be migrated/created in different db (or db systems, say one table in DynamoDB and the other one in Postgres), which makes JOIN in the database layer impossible.
Design the application for current need. Dont assume that something like that will happen in future and design for that and compromise on current application performance. If there is a definite future need, go for in-memory operations. Otherwise, better go for database JOINs.
Related
I'm a database noobie when it comes to even moderately large data sets. I have a SQL database (multiple sql databases actually, a SQLite, Postgres, and MySQL database) all containing the same data, dumped from IMDB. I want to benchmark these different databases. The main table I want to query has about 15 million rows. I want a query that crosses two movies, right now my query looks like this
SELECT * from acted_in INNER JOIN actors
ON acted_in.idactors = actors.idactors WHERE
(acted_in.idmovies = %d OR acted_in.idmovies = %d)
the parameters are randomly generated ids. I want to test the relative speed of the databases by running this query multiple times for randomly generated movies and seeing the amount of time it takes on average. My question is, is there any better way to do the same query, I want to join who acted in what with their information from either of the two movies as this will be the core functionality for the project I am working on, right now the speed is abysmal currently the average speed for a single query is
sqlite: 7.160171360969543
postgres: 8.263306670188904
mysql: 13.27652293920517
This is average time per query (sample space of only 100 queries but it is significant enough for now). So can I do any better? The current run time is completely unacceptable for any practical use. I don't think the joining takes a lot of time, by removing it I get nearly the same results so I believe the lookup is what is taking a long time, as I don't gain a significant speed up when I don't join or look up using the OR conditional.
The thing you don't mention here is having any indexes in the databases. Generally, the way you speed up a query (except for terribly written ones, which this is not) is by adding indexes to the things which are used in join or where criteria. This would slow down updates since the indexes need to be updated any time the table is updated, but would speed up selections using those attributes quite substantially. You may wish to consider adding indexes to any attributes you use which are not already primary keys. Be sure to use the same index type in all databases to be fair.
First off, microbenchmarks on databases are pretty noninformative and it's not a good idea to base your decision on them. There are dozens of better criteria to select a db, like reliability, behavior under heavy loads, availability of certain features (eg an extensible language like the PostGIS extension for postgres, partitioning, ...), license (!!), and so on.
Second, if you want to tune your database, or database server, there's a number of things you have to consider. Some important ones:
db's like lots of memory and fast disks, so setup your server with ample quantities of both.
use the query analysis features offered by all major db's (eg the very visual explain feature in pgadmin for postgres) to analyze the behavior of queries that are important for your use case, and adapt the db based on what you learn from these analyses (eg extra or other indexes)
study to understand your db server well, these are pretty sophisticated programs with lots of settings that influence their behavior and performance
make sure you understand the workload your db is subjected to, eg by using a tool like pgfouine for postgres, others exist for other brands of databases.
Say tableA has 1 row to be returned but will have 100 columns returned while tableB has 100 rows to be returned but only one column from each. TableB has a foreign key for table A.
Will a left join of tableA to tableB return 100*100 cells of data while 2 separate queries return 100 + 100 cells of data or 50 times less data or is that a misunderstanding of how it works?
Is it ever more efficient to use many simple queries rather than fewer more complex ones?
First and foremost, I would question a table with 100 columns, and suggest that there is a possibly a better design for your schema. In the real world, this number of columns is less common, so typically the difference in the amount of data returned with one query vs. two becomes less significant. 100 columns in a table is not necessarily bad, just a flag that it shold be considered.
However, assuming your numbers are what they are to make clear the question, there are a few important variables to consider:
1 - What is the speed of the link between the db server and the application server? If it is very slow, then you are probably better off minimizing the amount data returned vs. the number of queries you run. If it is not slow, then you will likely expend more time in the execution of two queries than you would returning the increased payload. Which is better can only be determined by testing in your own environment.
2 - How efficient is the transport protocol itself? Perhaps there is some kind of compression of the data, or an even more clever algorithm that knows column 2 through 101 are duplicate for every row, so it only passes them once. Strategies like this in the transport protocol would mitigate any of your concerns. Again, this is why you need to test in your own envionment to know for sure.
As others have pointed out, you also need to consider what will be done with the data once you get it (e.g., JOINs, GROUPing, etc), but I am limiting my response to the specifics of your question around query count vs. payload size.
What is best at joining? A database engine or client code? Saying that, I use both techniques: it depends on the client and how data will be used.
Where the data requires some processing to, say, render on a web page I'd probably split header and details recordsets. We do use this because we have some business logic between DB and HTML
Where it's consumed simply and linearly, I'd join in the database to avoid unnecessary processing. For example, simple reports or exports
It depends, if you only take into account the SQL efficiency obviusly several simpler and smaller result queries will be more efficient.
But you need to take into account the whole process if the join will be made otherwise on the client or you need to filter results after the join, then probably the DBM will be more efficient that doing it on your code.
Coding is always a tradeoff between diferent systems, DB vs Client, RAM vs CPU... you need to be conscious about this and try to find the perfect solution.
In this case probably 2 queries outperform 1 but that is not a general solution.
I think that your question basically is about database normalization. In general, it is advisable to normalize a database into multiple tables (using primary and foreign keys) and to join them as needed upon queries. This is better for insert/update performance and for keeping the data consistent, and usually results in smaller database sizes as well.
As for the row numbers returned, only a cross join would actually return 100*100 rows; any inner or outer join will not create all combinations, but rather tie together rows on the given conditions, and for outer joins preserve rows which could not be matched. Wikipedia has some samples in its JOIN article.
For very query-intense applications, the performance may be better when using less normlized tables. However, as always with optimizations, I'd only consider going into that direction after seeing real measurable problems (e.g. with a profiling tool).
In general, try to keep the number of roundtrips to the database low; a large number of single simple queries will suffer from the overhead of talking to the DB engine (network etc.). If you need to execute complex series of statements, consider using stored procedures.
Generally fewer queries makes for better performance, as long as the queries return data that is actually related. There is no point in trying to put unrelated data into the same query just to reduce the number or queries.
There are of course exceptions, and your example may be one of them. However, it depends on more than the number of fields returnes, like what the fields actually return, i.e. the actual amount of data.
As an example of how the number of queries affects performance, I can mention a solution that I have (sadly enough) seen many times. In that solution the programmer would first get a number of records from one table, then loop through the records and run another query for each record to get the related records from another table. This clearly results in a lot of queries, and a solution having either one or two queries would be much more efficient.
“Is it ever more efficient to use many simple queries rather than fewer more complex ones?”
The query that requires the least amount of data to traverse, and gives you no more than what you need is the more efficient one. Beyond this, there can be RDBMS specific conditions that can be more efficient on one RDBMS system than another. At the very low level, when you deal with less data, then your results can be retrieved much quicker, so efficient queries are queries that only work with the least amount of data needed to get you the result you are looking for.
Why are joins bad or 'slow'. I know i heard this more then once. I found this quote
The problem is joins are relatively
slow, especially over very large data
sets, and if they are slow your
website is slow. It takes a long time
to get all those separate bits of
information off disk and put them all
together again.
source
I always thought they were fast especially when looking up a PK. Why are they 'slow'?
Scalability is all about pre-computing (caching), spreading out, or paring down the repeated work to the bare essentials, in order to minimize resource use per work unit. To scale well, you don't do anything you don't need to in volume, and the things you actually do you make sure are done as efficiently as possible.
In that context, of course joining two separate data sources is relatively slow, at least compared to not joining them, because it's work you need to do live at the point where the user requests it.
But remember the alternative is no longer having two separate pieces of data at all; you have to put the two disparate data points in the same record. You can't combine two different pieces of data without a consequence somewhere, so make sure you understand the trade-off.
The good news is modern relational databases are good at joins. You shouldn't really think of joins as slow with a good database used well. There are a number of scalability-friendly ways to take raw joins and make them much faster:
Join on a surrogate key (autonumer/identity column) rather than a natural key. This means smaller (and therefore faster) comparisons during the join operation
Indexes
Materialized/indexed views (think of this as a pre-computed join or managed de-normalization)
Computed columns. You can use this to hash or otherwise pre-compute the key columns of a join, such that what would be a complicated comparison for a join is now much smaller and potentially pre-indexed.
Table partitions (helps with large data sets by spreading the load out to multiple disks, or limiting what might have been a table scan down to a partition scan)
OLAP (pre-computes results of certain kinds of queries/joins. It's not quite true, but you can think of this as generic denormalization)
Replication, Availability Groups, Log shipping, or other mechanisms to let multiple servers answer read queries for the same database, and thus scale your workload out among several servers.
Use of a caching layer like Redis to avoid re-running queries which need complex joins.
I would go as far as saying the main reason relational databases exist at all is to allow you do joins efficiently*. It's certainly not just to store structured data (you could do that with flat file constructs like csv or xml). A few of the options I listed will even let you completely build your join in advance, so the results are already done before you issue the query — just as if you had denormalized the data (admittedly at the cost of slower write operations).
If you have a slow join, you're probably not using your database correctly.
De-normalization should be done only after these other techniques have failed. And the only way you can truly judge "failure" is to set meaningful performance goals and measure against those goals. If you haven't measured, it's too soon to even think about de-normalization.
* That is, exist as entities distinct from mere collections of tables. An additional reason for a real rdbms is safe concurrent access.
Joins can be slower than avoiding them through de-normalisation but if used correctly (joining on columns with appropriate indexes an so on) they are not inherently slow.
De-normalisation is one of many optimisation techniques you can consider if your well designed database schema exhibits performance problems.
article says that they are slow when compared to absence of joins. this can be achieved with denormalization. so there is a trade off between speed and normalization. don't forget about premature optimization also :)
First of all, a relational database's raison d'etre (reason for being) is to be able to model relationships between entities. Joins are simply the mechanisms by which we traverse those relationships. They certainly do come at a nominal cost, but without joins, there really is no reason to have a relational database.
In the academic world we learn of things like the various normal forms (1st, 2nd, 3rd, Boyce-Codd, etc.), and we learn about different types of keys (primary, foreign, alternate, unique, etc.) and how these things fit together to design a database. And we learn the rudiments of SQL as well as manipulating both structure and data (DDL & DML).
In the corporate world, many of the academic constructs turn out to be substantially less viable than we had been led to believe. A perfect example is the notion of a primary key. Academically it is that attribute (or collection of attributes) that uniquely identifies one row in the table. So in many problem domains, the proper academic primary key is a composite of 3 or 4 attributes. However, almost everyone in the modern corporate world uses an auto-generated, sequential integer as a table's primary key. Why? Two reasons. The first is because it makes the model much cleaner when you're migrating FKs all over the place. The second, and most germane to this question, is that retrieving data through joins is faster and more efficient on a single integer than it is on 4 varchar columns (as already mentioned by a few folks).
Let's dig a little deeper now into two specific subtypes of real world databases. The first type is a transactional database. This is the basis for many e-commerce or content management applications driving modern sites. With a transaction DB, you're optimizing heavily toward "transaction throughput". Most commerce or content apps have to balance query performance (from certain tables) with insert performance (in other tables), though each app will have its own unique business driven issues to solve.
The second type of real world database is a reporting database. These are used almost exclusively to aggregate business data and to generate meaningful business reports. They are typically shaped differently than the transaction databases where the data is generated and they are highly optimized for speed of bulk data loading (ETLs) and query performance with large or complex data sets.
In each case, the developer or DBA needs to carefully balance both the functionality and performance curves, and there are lots of performance enhancing tricks on both sides of the equation. In Oracle you can do what's called an "explain plan" so you can see specifically how a query gets parsed and executed. You're looking to maximize the DB's proper use of indexes. One really nasty no-no is to put a function in the where clause of a query. Whenever you do that, you guarantee that Oracle will not use any indexes on that particular column and you'll likely see a full or partial table scan in the explain plan. That's just one specific example of how a query could be written that ends up being slow, and it doesn't have anything to do with joins.
And while we're talking about table scans, they obviously impact the query speed proportionally to the size of the table. A full table scan of 100 rows isn't even noticeable. Run that same query on a table with 100 million rows, and you'll need to come back next week for the return.
Let's talk about normalization for a minute. This is another largely positive academic topic that can get over-stressed. Most of the time when we talk about normalization we really mean the elimination of duplicate data by putting it into its own table and migrating an FK. Folks usually skip over the whole dependence thing described by 2NF and 3NF. And yet in an extreme case, it's certainly possible to have a perfect BCNF database that's enormous and a complete beast to write code against because it's so normalized.
So where do we balance? There is no single best answer. All of the better answers tend to be some compromise between ease of structure maintenance, ease of data maintenance and ease of code creation/maintenance. In general, the less duplication of data, the better.
So why are joins sometimes slow? Sometimes it's bad relational design. Sometimes it's ineffective indexing. Sometimes it's a data volume issue. Sometimes it's a horribly written query.
Sorry for such a long-winded answer, but I felt compelled to provide a meatier context around my comments rather than just rattle off a 4-bullet response.
People with terrabyte sized databases still use joins, if they can get them to work performance-wise then so can you.
There are many reasons not to denomalize. First, speed of select queries is not the only or even main concern with databases. Integrity of the data is the first concern. If you denormalize then you have to put into place techniques to keep the data denormalized as the parent data changes. So suppose you take to storing the client name in all tables instead of joining to the client table on the client_Id. Now when the name of the client changes (100% chance some of the names of clients will change over time), now you need to update all the child records to reflect that change. If you do this wil a cascade update and you have a million child records, how fast do you suppose that is going to be and how many users are going to suffer locking issues and delays in their work while it happens? Further most people who denormalize because "joins are slow" don't know enough about databases to properly make sure their data integrity is protected and often end up with databases that have unuseable data becasue the integrity is so bad.
Denormalization is a complex process that requires an thorough understanding of database performance and integrity if it is to be done correctly. Do not attempt to denormalize unless you have such expertise on staff.
Joins are quite fast enough if you do several things. First use a suggorgate key, an int join is almost alawys the fastest join. Second always index the foreign key. Use derived tables or join conditions to create a smaller dataset to filter on. If you have a large very complex database, then hire a professional database person with experience in partioning and managing huge databases. There are plenty of techniques to improve performance without getting rid of joins.
If you just need query capability, then yes you can design a datawarehouse which can be denormalized and is populated through an ETL tool (optimized for speed) not user data entry.
Joins are slow if
the data is improperly indexed
results poorly filtered
joining query poorly written
data sets very large and complex
So, true, the bigger your data sets the the more processing you'll need for a query but checking and working on the first three options of the above will often yield great results.
Your source gives denormalization as an option. This is fine only as long as you've exhausted better alternatives.
The joins can be slow if large portions of records from each side need to be scanned.
Like this:
SELECT SUM(transaction)
FROM customers
JOIN accounts
ON account_customer = customer_id
Even if an index is defined on account_customer, all records from the latter still need to be scanned.
For the query list this, the decent optimizers won't probably even consider the index access path, doing a HASH JOIN or a MERGE JOIN instead.
Note that for a query like this:
SELECT SUM(transaction)
FROM customers
JOIN accounts
ON account_customer = customer_id
WHERE customer_last_name = 'Stellphlug'
the join will most probably will be fast: first, an index on customer_last_name will be used to filter all Stellphlug's (which are of course, not very numerous), then an index scan on account_customer will be issued for each Stellphlug to find his transactions.
Despite the fact that these can be billions of records in accounts and customers, only few will actually need to be scanned.
Joins are fast. Joins should be considered standard practice with a properly normalized database schema. Joins allow you to join disparate groups of data in a meaningful way. Don't fear the join.
The caveat is that you must understand normalization, joining, and the proper use of indexes.
Beware premature optimization, as the number one failing of all development projects is meeting the deadline. Once you've completed the project, and you understand the trade offs, you can break the rules if you can justify it.
It's true that join performance degrades non-linearly as the size of the data set increases. Therefore, it doesn't scale as nicely as single table queries, but it still does scale.
It's also true that a bird flies faster without any wings, but only straight down.
Joins do require extra processing since they have to look in more files and more indexes to "join" the data together. However, "very large data sets" is all relative. What is the definition of large? I the case of JOINs, I think its a reference to a large result set, not that overall dataset.
Most databases can very quickly process a query that selects 5 records from a primary table and joins 5 records from a related table for each record (assuming the correct indexes are in place). These tables can have hundreds of millions of records each, or even billions.
Once your result set starts growing, things are going to slow down. Using the same example, if the primary table results in 100K records, then there will be 500K "joined" records that need to be found. Just pulling that much data out of the database with add delays.
Don't avoid JOINs, just know you may need to optimize/denormalize when datasets get "very large".
Also from the article you cited:
Many mega-scale websites with billions
of records, petabytes of data, many
thousands of simultaneous users, and
millions of queries a day are doing is
using a sharding scheme and some are
even advocating denormalization as the
best strategy for architecting the
data tier.
and
And unless you are a really large
website you probably don't need to
worry about this level of complexity.
and
It's more error prone than having the
database do all this work, but you are
able to do scale past what even the
highest end databases can handle.
The article is discussing mega-sites like Ebay. At that level of usage you are likely going to have to consider something other than plain vanilla relational database management. But in the "normal" course of business (applications with thousands of users and millions of records) those more expensive, more error prone approaches are overkill.
Joins are considered an opposing force to scalability because they're typically the bottleneck and they cannot be easily distributed or paralleled.
Properly designed tables containing with the proper indicies and correctly written queries not always slow. Where ever you heard that:
Why are joins bad or 'slow'
has no idea what they are talking about!!! Most joins will be very fast. If you have to join many many rows at one time you might take a hit as compared to a denormalized table, but that goes back to Properly designed tables, know when to denormalize and when not to. in a heavy reporting system, break out the data in denormalized tables for reports, or even create a data warehouse. In a transactional heavy system normalize the tables.
The amount of temporary data that is generated could be huge based on the joins.
For an example, one database here at work had a generic search function where all of the fields were optional. The search routine did a join on every table before the search began. This worked well in the beginning. But, now that the main table has over 10 million rows... not so much. Searches now take 30 minutes or more.
I was tasked with optimizing the search stored procedure.
The first thing I did was if any of the fields of the main table were being searched, I did a select to a temp table on those fields only. THEN, I joined all the tables with that temp table before doing the rest of the search. Searches where one of the main table fields now take less than 10 seconds.
If none of the main table fields are begin searched, I do similar optimizations for other tables. When I was done, no search takes longer than 30 seconds with most under 10.
CPU utilization of the SQL server also went WAY DOWN.
While joins (presumably due to a normalized design) can obviously be slower for data retrieval than a read from a single table, a denormalized database can be slow for data creation/update operations since the footprint of the overall transaction will not be minimal.
In a normalized database, a piece of data will live in only one place, so the footprint for an update will be as minimal as possible. In a denormalized database, it's possible that the same column in multiple rows or across tables will have to be updated, meaning the footprint would be larger and chance of locks and deadlocks can increase.
Well, yeah, selecting rows from one denormalized table (assuming decent indexes for your query) might be faster that selecting rows constructed from joining several tables, particularly if the joins don't have efficient indexes available.
The examples cited in the article - Flickr and eBay - are exceptional cases IMO, so have (and deserve) exceptional responses. The author specifically calls out the lack of RI and the extent of data duplication in the article.
Most applications - again, IMO - benefit from the validation & reduced duplication provided by RDBMSs.
They can be slow if done sloppily. For example, if you do a 'select *' on a join you will probaby take a while to get stuff back. However, if you carefully choose what columns to return from each table, and with the proper indexes in place, there should be no problem.
Suddenly (but unfortunately I don't know when "suddenly" was; I know it ran fine at some point in the past) one of my queries started taking 7+ seconds instead of milliseconds to execute. I have 1 local table and 3 tables being accessed via a DB link. The 3 remote tables are joined together, and one of them is joined with my local table.
The local table's where clause only takes a few millis to execute on its own, and only returns a few (10's or 100's at the most) records. The 3 remote tables have many hundreds of thousands, possibly millions, of records between them, and if I join them appropriately I get tens or hundreds of thousands of records.
I am only joining with the remote tables so that I can pull out a few pieces of data related to each record in my local table.
What appears to be happening, however, is that Oracle joins the remote tables together first and then my local table to that mess at the end. This is always going to be a bad idea, especially given the data set that exists right now, so I added a /*+ LEADING(local_tab remote_tab_1) */ hint to my query and it now returns in milliseconds.
I compared the explain plans and they are almost identical, save for a single BUFFER SORT on one of the remote tables.
I'm wondering what might cause Oracle to approach this the wrong way? Is it an index issue? What should I be looking for?
When choosing an execution plan, oracle estimates costs for the different plans. One crucial information for that estimate is the amount of rows will get returned from a step of the execution plan. Oracle tries to estimate those using 'statistics', i.e. information about how many rows a table contains, how many different values a column contains; How evenly these values are distributed.
These statistics are just that statistics, and they might be wrong, which is one of the most important reasons for misjudgments of the oracle optimizer.
So gathering new statistics as described in a comment might help. Have a look at the documentation on that dbms_stats package. There are many different ways to call that package.
A common problem I've come across is a query that joins many tables, where the joins form a chain from one end to another, e.g.:
SELECT *
FROM tableA, tableB, tableC, tableD, tableE
WHERE tableA.ID0 = :bind1
AND tableA.ID1 = tableB.ID1
AND tableB.ID2 = tableC.ID2
AND tableC.ID3 = tableD.ID3
AND tableD.ID4 = tableE.ID4
AND tableE.ID5 = :bind2;
Notice how the optimiser might choose to drive the query from tableA (e.g. if the index on ID0 is nicely selective) or from tableE (if the index on tableE.ID5 is more selective).
The statistics on the tables might cause the choice between these two plans to balance on a knife-edge; one day it's working fine (driving from tableA), next day new stats are gathered and all of a sudden the alternative plan driving from tableE has a lower cost and is chosen.
In this circumstance, adding a LEADING hint is one way to nudge it back to the original plan (i.e. drive from tableA) without dictating too much to the optimiser (i.e. it doesn't force the optimiser to choose any particular join methods).
You're doing distributed query optimization, and that's a tricky beast. It could be that the your table's statistics are current, but now the tables at the remote system are out-of-whack or have changed. Or the remote system added/removed/modified indexes, and that broke your plan. (This is an excellent reason to consider replication -- so you can control indexes and statistics against it.)
That said, Oracle's estimate of cardinality is a primary driver in execution plan. A 10053 trace analysis (Jonathan Lewis' Cost-Based Oracle Fundamentals book has wonderful examples from 8i to 10.1) can help shed light on why your statement's now broken and how the LEADING hint fixes it.
The DRIVING_SITE hint might be a better choice if you know you always want the local tables to be joined first before going after the remote site; it clarifies your intention without driving the plan the way a LEADING hint would.
Might not be relevant but I had a similar situation once where the remote table had been replaced by a single-table view. When it was a table the distributed query optimizer 'saw' that it had an index. When it became a view it couldn't see the index anymore and couldn't cost a plan that used an index on the remote object.
That was a few years ago. I documented my analysis at the time here.
RI,
It's hard to be sure about the cause of the performance problems without seeing the SQL.
When an Oracle query was performing well before, and suddenly starts performing badly, it is usually related to one of two issues:
A) Statistics are out of date. This is the easiest and quickest thing to check, even if you have a housekeeping batch process that's supposed to take care of it ... always double-check.
B) Data volume / data pattern change.
In your case, running a distributed query across multiple databases makes it 10x harder for Oracle to manage performance between them. Is it possible to put these tables in one database, perhaps separate schema owners in one database?
Hints are notoriously fragile, as Oracle is under no obligations to follow the hint. When the data volume or pattern changes some more, Oracle may just ignore the hint and do what it thinks is best (ie. worst ;-).
If you cannot put these tables all in one database, then I recommend you look to break your query up into two statements:
INSERT on sub-SELECT to copy external data to a global temporary table in your current database.
SELECT from the global temporary table to join with your other table.
You will have complete control over performance of step 1 above without resorting to hints. This approach typically scales well, providing you take time to do the performance tuning. I've seen this approach solve many complex performance problems.
The overhead for Oracle to create a whole new table, or insert a heap of records, is much smaller than most people expect. Defining a global temporary table further reduces that overhead.
Matthew
I would like to know if there's a really performance gain between those two options :
Option 1 :
I do a SQL Query with a join to select all User and their Ranks.
Option 2 :
I do one SQL Query to select all User
I fetch all user and do another SQL Query to get the Ranks of this User.
In code, option two is easier to realize for me. That's only because the way I design my Persistence layer.
So, I would like to know what's the impact on performance. After what limit I should consider to take Option 1 instead of Option 2 ?
Generally speaking, the DB server is always faster at joining than application code. Remember you will have to do an extra query with a network round trip for each join. However, if your first result set is small and your indexes are well tuned, this model can work fine.
If you are only doing this to re-use your ORM solution, then you may be fighting a losing battle. I have invariably found that I need read-only datasets that can only be produced with SQL, so I now use ORM for per-object CRUD operations and regular SQL for searches, reports, aggregates etc.
If ranks are static values, consider caching them in your application.
If you need users frequently and ranks only rarely, consider lazy-loading of ranks. (e.g., separate queries, but the second query gets used only occasionally).
Use the join if you're always going to need both sets of data, and they have to be current copies of the database.
Prototype any likely choices, and run performance tests.
EDIT: Further thoughts on your persistence layer, because I'm facing this one myself. Consider adding "persistence-like" classes that handle joins as their basic query, and are read-only. Whether this fits your particular scenario is for you to decide, but a lot of database access for many apps is based on joins, which can be rather large and complex. If you can handle these in a consistent manner with your persistent, updatable objects, it can be a big win for your overall architecture. Conceptually, it's a lot like having a view in the database, and querying the view instead of writing a join, but you're doing it all in code.
It depends upon how many users you anticipate. Option one will definitely be faster, but with a reasonable amount of data, the difference will be negligible.
In 99% situations join will be faster.
However there is one rare situations when it can be slower. If your are doing one to many join on table with large row size and you are hitting network bandwidth limit.
For example there is a blob column in T1 of 1MB size, you are joining T2 which consist 100 rows for each T1 row. The result set would be T1 row count multiple 100.
So if you are querying one T1 row with join it would be 100MB result set, if you fetch T1 row (1MB) and then do separate select to fetch 100 T2 for this T1 the result set will be 1MB.