The first thing that comes to most people’s mind when database table compression is mentioned is the savings it yields in terms of disk space. While reducing the footprint of data on disk is relevant, I would argue it is the lesser of the benefits for data warehouses. Disk capacity is very cheap and generally plentiful, however, disk bandwidth (scan speed) is proportional to the number of spindles, no mater what the disk capacity and thus is more expensive. Table compression reduces the footprint on the disk drives that a given data set occupies so the amount of physical data that must be read off the disk platters is reduced when compared to the uncompressed version. For example, if 4000 GB of raw data can compress to 1000 GB, it can be read off the same disk drives 4X as fast because it is reading and transferring 1/4 of the data off the spindles (relative to the uncompressed size). Likewise, table compression allows for the database buffer cache to contain more data without having to increase the memory allocation because more rows can be stored in a compressed block/page compared to an uncompressed block/page.
Row major table compression comes in two flavors with the Oracle database: BASIC and OLTP. In 11.1 these were also known by the key phrases COMPRESS or COMPRESS FOR DIRECT_LOAD OPERATIONS and COMPRESS FOR ALL OPERATIONS. The BASIC/DIRECT_LOAD compression has been part of the Oracle database since version 9 and ALL OPERATIONS/OLTP compression was introduced in 11.1 with the Advanced Compression option.
Oracle row major table compression works by storing the column values for a given block in a symbol table at the beginning of the block. The more repeated values per block, even across columns, the better the compression ratio. Sorting data can increase the compression ratio as ordering the data will generally allow more repeat values per block. Specific compression ratios and gains from sorting data are very data dependent but compression ratios are generally between 2x and 4x.
Compression does add some CPU overhead when direct path loading data, but there is no measurable performance overhead when reading data as the Oracle database can operate on compressed blocks directly without having to first uncompress the block. The additional CPU required when bulk loading data is generally well worth the down wind gains for data warehouses. This is because most data in a well designed data warehouse is write once, read many times. Insert only and infrequently modified tables are ideal candidates for BASIC compression. If the tables have significant DML performed against them, then OLTP compression would be advised (or no compression).
Given that most Oracle data warehouses that I have seen are constrained by I/O bandwidth (see Balanced Hardware Configuration) it is highly recommended to leverage compression so the logical table scan rate can increase proportionally to the compression ratio. This will result in faster table and partition scans on the same hardware.
Oracle Documentation References:
- Performance Tuning Guide: Table Compression
- VLDB and Partitioning Guide: Partitioning and Table Compression
- Guidelines for Managing Tables: Consider Using Table Compression
- Table Compression in Oracle9i Release 2: A Performance Analysis
- Table Compression in Oracle Database 10g Release 2
- Advanced Compression with Oracle Database 11g Release 2
[back to Introduction]
If you want to build a house that will stand the test of time, you need to build on a solid foundation. The same goes for architecting computer systems that run databases. If the underlying hardware is not sized appropriately it will likely lead to people blaming software. All too often I see data warehouse systems that are poorly architected for the given workload requirements. I frequently tell people, “you can’t squeeze blood from a turnip“, meaning if the hardware resources are not there for the software to use, how can you expect the software to scale?
Undersizing data warehouse systems has become an epidemic with open platforms – platforms that let you run on any brand and configuration of hardware. This problem has been magnified over time as the size of databases have grown significantly, and generally outpacing the experience of those managing them. This has caused the “big three” database vendors to come up with suggested or recommended hardware configurations for their database platforms:
- Oracle: Optimized Warehouse Initiative
- Microsoft: SQL Server Fast Track Data Warehouse
- IBM: Balanced Configuration Unit (BCU)
Simply put, the reasoning behind those initiatives was to help customers architect systems that are well balanced and sized appropriately for the size of their data warehouse.
Balanced Hardware Configurations
The foundation for a well performing data warehouse (or any system for that matter) is the hardware that it runs on. There are three main hardware resources to consider when sizing your data warehouse hardware. Those are:
- Number of CPUs
- Number of storage devices (HDDs or SSDs)
- I/O bandwidth between CPUs and storage devices
NB: I’ve purposely left off memory (RAM) as most systems are pretty well sized at 2GB or 4GB per CPU core these days.
A balanced system has the following characteristics:
As you can see, each of the three components are sized proportionally to each other. This allows for the max system throughput capacity as no single resource will become the bottleneck before any other. This was one of the critical design decisions that went into the Oracle Database Machine.
Most DBAs and System Admins know what the disk capacity numbers are for their systems, but when it comes to I/O bandwidth or scan rates, most are unaware of what the system is capable of in theory, let alone in practice. Perhaps I/O bandwidth utilization should be included in the system metrics that are collected for your databases. You do collect system metrics, right?
There are several “exchanges” that data must flow through from storage devices to host CPUs, many of which could become bottlenecks. Those include:
- Back-end Fibre Channel loops (the fibre between the drive shelves and the storage array server processor)
- Front-end Fibre Channel ports
- Storage array server processors (SP)
- Host HBAs
One should understand the throughput capacity of each of these components to ensure that one (or more) of them do not restrict the flow of data to the CPUs prematurely.
Unbalanced Hardware Configurations
All too frequently systems are not architected as balanced systems and the system ends up being constrained in one of the following three scenarios:
From the production systems that I have seen, the main deficiency is in I/O bandwidth (both I/O Channel and HDD). I believe there are several reasons for this. First, too many companies capacity plan for their data warehouse based on the size the data occupies on disk alone. That is, they purchase the number of HDDs for the system based on the drive capacity, not on the I/O bandwidth requirement. Think of it like this: If you were to purchase 2 TB of mirrored disk capacity (4 TB total) would you rather purchase 28 x 146 GB drives or 14 x 300 GB drives (or even 4 x 1 TB drives)? You may ask: Well, what is the difference (other than price); in each case you have the same net capacity, correct? Indeed, both configurations do have the same capacity, but I/O bandwidth (how fast you can read data off the HDDs) is proportional to the number of HDDs, not the capacity. Thus it should be slightly obvious then that 28 HDDs can deliver 2X the disk I/O bandwidth that 14 HDDs can. This means that it will take 2X as long to read the same amount of data off of 14 HDDs as 28 HDDs.
Unfortunately what tends to happen is that the bean counter types will see only two things:
- The disk capacity (space) you want to purchase (or the capacity that is required)
- The price per MB/GB/TB
This is where someone worthy of the the title systems architect needs to stand up and explain the concept of I/O bandwidth and the impact it has on data warehouse performance (your systems architect does know this, correct?). This is generally a difficult discussion because I/O bandwidth is not a line item on a purchase order, it is a derived metric that requires both thought and engineering (which means someone had to do some thinking about the requirements for this system!).
When sizing the hardware for your data warehouse consider your workload and understand following (and calculate numbers for them!):
- What rate (in MB/GB per second) can the CPUs consume data?
- What rate can storage devices produce data (scan rate)?
- What rate can the data be delivered from the storage array(s) to the host HBAs?
If you are unable to answer these questions in theory then you need to sit down and do some calculations. Then you need to use some micro benchmarks (like Oracle ORION) and prove out those calculations. This will give you the “speed limit” and an metric by which you can measure your database workload against. All computer systems much obey the laws of physics! There is no way around that.
Kevin Closson has several good blog posts on related topics including:
- SAN Admins: Please Give Me As Much Capacity From As Few Spindles As Possible!
- Hard Drives Are Arcane Technology. So Why Can’t I Realize Their Full Bandwidth Potential?
as well as numerous others.
Oracle Documentation References:
At the 2009 Oracle OpenWorld Unconference back in October I lead a chalk and talk session entitled The Core Performance Fundamentals Of Oracle Data Warehousing. Since this was a chalk and talk I spared the audience any powerpoint slides but I had several people request that make it into a presentation so they could share it with others. After some thought, I decided that a series of blog posts would probably be a better way to share this information, especially since I tend to use slides as a speaking outline, not a condensed version of a white paper. This will be the first of a series of posts discussing what I consider to be the key features and technologies behind well performing Oracle data warehouses.
As an Oracle database performance engineer who has done numerous customer data warehouse benchmarks and POCs over the past 5+ years, I’ve seen many data warehouse systems that have been plagued with problems on nearly every DBMS commonly used in data warehousing. Interestingly enough, many of these systems were facing many of the same problems. I’ve compiled a list of topics that I consider to be key features and/or technologies for Oracle data warehouses:
Core Performance Fundamental Topics
- Balanced Hardware Configuration
- Table Compression
- Parallel Execution
- Data Loading
- Row vs. Set Processing
- Indexing and Materalized Views
In the upcoming posts, I’ll deep dive into each one of these topics discussing why these areas are key for a well performing Oracle data warehouse. Stay tuned…
Today, June 10th, marks the Yahoo! Hadoop Summit ’09 and the crew at Facebook have a writeup on the Facebook Engineering page entitled: Hive – A Petabyte Scale Data Warehouse Using Hadoop.
I found this an very interesting read given some of the Hadoop/MapReduce comments from David J. DeWitt and Michael Stonebraker as well as their SIGMOD 2009 paper, A Comparison of Approaches to Large-Scale Data Analysis. Now I’m not about to jump into this whole dbms-is-better-than-mapreduce argument but I found Facebook’s story line interesting:
When we started at Facebook in 2007 all of the data processing infrastructure was built around a data warehouse built using a commercial RDBMS. The data that we were generating was growing very fast – as an example we grew from a 15TB data set in 2007 to a 2PB data set today. The infrastructure at that time was so inadequate that some daily data processing jobs were taking more than a day to process and the situation was just getting worse with every passing day. We had an urgent need for infrastructure that could scale along with our data and it was at that time we then started exploring Hadoop as a way to address our scaling needs.
[The] Hive/Hadoop cluster at Facebook stores more than 2PB of uncompressed data and routinely loads 15 TB of data daily
Wow, 2PB of uncompressed data and growing at around 15TB daily. A part of me wonders how much value there is in 2PB of data or if companies are suffering from OCD when it comes to data. Either way it’s interesting to see how much data is being generated/collected and how engineers are dealing with it.
A few weeks ago I read Curt Monash’s report on interpreting the results of data warehouse proofs-of-concept (POCs) and I have to say, I’m quite surprised that this topic hasn’t been covered more by analysts in the data warehousing space. I understand that analysts are not database performance engineers, but where do they think that the performance claims of 10x to 100x or more come from? Do they actually investigate these claims or just report on them? I can not say that I have ever seen any database analyst offer any technical insight into these boasts of performance. If some exist be sure to leave a comment and point me to them.
Oracle Exadata Performance Architect Kevin Closson has blogged about a 485x performance increase of Oracle Exadata vs. Oracle Exadata and his follow-up post to explain exactly where the 485x performance gain comes from gave me the nudge to finish this post that had been sitting in my drafts folder since I first read Curt’s post.
Customer Bechmarketing Claims
I thought I would compile a list of what the marketing folks at other database vendors are saying about the performance of their products. Each of these statements have been taken from the given vendor’s website.
- Netezza: 10-100 times faster than traditional solutions…but it is not uncommon to see performance differences as large as 200x to even 400x or more when compared to existing Oracle systems
- Greenplum: often 10 to 100 times faster than traditional solutions
- DATAllegro: 10-100x performance over traditional platforms
- Vertica: Performs 30x-200x faster than other solutions
- ParAccel: 20X – 200X performance gains
- EXASolution: can perform up to 100 times faster than with traditional databases
- Kognitio WX2: Tests have shown to out-perform other database / data warehouse solutions by 10-60 times
Certainly seems these vendors are a positioning themselves against traditional database solutions, whatever that means. And differences as large as 400x against Oracle? What is it exactly they are comparing?
Investigative Research On Netezza’s Performance Claims
Using my favorite Internet search engine I came across this presentation by Netezza dated October 2007. On slide 21 Netezza is comparing an NPS 8150 (112 SPU, up to 4.5 TB of user data) server to IBM DB2 UDB on a p680 with 12 CPUs (the existing solution). Not being extremely familiar with the IBM hardware mentioned, I thought I’d research to see exactly what an IBM p680 server consists of. The first link in my search results took me to here where the web page states:
The IBM eServer pSeries 680 has been withdrawn from the market, effective March 28, 2003.
Searching a bit more I came across this page which states that the 12 CPUs in the pSeries 680 are RS64 IV microprocessors. According to Wikipedia the “RS64-IV or Sstar was introduced in 2000 at 600 MHz, later increased to 750 MHz”. Given that at best, the p680 had 12 CPUs running at 750 MHz and the NPS 8150 had 112 440GX PowerPC processors I would give the compute advantage to Netezza by a significant margin. I guess it is cool to brag how your most current hardware beat up on some old used and abused server who has already been served its end-of-life notice. I found it especially intriguing that Netezza is boasting about beating out an IBM p680 server that has been end-of-lifed more than four years prior to the presentation’s date. Perhaps they don’t have any more recent bragging to do?
Going back one slide to #20 you will notice a comparison of Netezza and Oracle. Netezza clearly states they used a NPS 8250 (224 SPUs, up to 9 TB of user data) against Oracle 10g RAC running on Sun/EMC. Well ok…Sun/EMC what??? Obviously there were at least 2 Sun servers, since Oracle 10g RAC is involved, but they don’t mention the server models at all, nor the storage, nor the storage connectivity to the hosts. Was this two or more Sun Netra X1s or what??? Netezza boasts a 449x improvement in a “direct comparison on one day’s worth of data”. What exactly is being compared is up to the imagination. I guess this could be one query or many queries, but the marketeers intentionally fail to mention. They don’t even mention the data set size being compared. Given that Netezza can read data off the 224 drives at 60-70 MB/s, the NPS 8250 has a total scan rate of over 13 GB/s. I can tell you first hand that there are very few Sun/EMC solutions that are configured to support 13 GB/s of I/O bandwidth. Most configurations of that vintage probably don’t support 1/10th of that I/O bandwidth (1.3 GB/s).
Here are a few more comparisons that I have seen in Netezza presentations:
- NPS 8100 (112 SPUs/4.5 TB max) vs. SAS on Sun E5500/6 CPUs/6GB RAM
- NPS 8100 (112 SPUs/4.5 TB max) vs. Oracle 8i on Sun E6500/12 CPUs/8 GB RAM
- NPS 8400 (448 SPUs/18 TB max) vs. Oracle on Sun (exact hardware not mentioned)
- NPS 8100 (112 SPUs/4.5 TB max) vs. IBM SP2 (database not mentioned)
- NPS 8150z (112 SPUs/5.5 TB max) vs. Oracle 9i on Sun/8 CPUs
- NPS 8250z (224 SPUs/11 TB max) vs. Oracle 9i on Sun/8 CPUs
As you can see, Netezza has a way of finding the oldest hardware around and then comparing it to its latest, greatest NPS. Just like Netezza slogan, [The Power to ]Question Everything™, I suggest you question these benchmarketing reports. Database software is only as capable as the hardware it runs on and when Netezza targets the worst performing and oldest systems out there, they are bound to get some good marketing numbers. If they compete against the latest, greatest database software running on the latest, greatest hardware, sized competitively for the NPS being used, the results are drastically different. I can vouch for that one first hand having done several POCs against Netezza.
One Benchmarketing Claim To Rule Them All
Now, one of my favorite benchmarketing reports is one from Vertica. Michael Stonebraker’s blog post on customer benchmarks contains the following table:
Take a good look at the Query 2 results. Vertica takes a query running in the current row store from running in 4.5 hours (16,200 seconds) to 1 second for a performance gain of 16,200x. Great googly moogly batman, that is reaching ludicrous speed. Heck, who needs 100x or 400x when you do 16,200x. That surely warrants an explanation of the techniques involved there. It’s much, much more than simply column store vs. row store. It does raise the question (at least to me): why Vertica doesn’t run every query in 1 second. I mean, come on, why doesn’t that 19 minute row store query score better than a 30x gain? Obviously there is a bit of the magic pixie dust going on here with, what I would refer to as “creative solutions” (in reality it is likely just a very well designed projection/materaizied view, but by showing the query and telling us how it was possible would make it less unimpressive [sic]).
What Is Really Going On Here
First of all, you will notice that not one of these benchmarketing claims is against a vendor run system. Each and every one of these claims are against existing customer systems. The main reason for this is that most vendors prohibit benchmark results being published with out prior consent from the vendor in the licensing agreement. Seems the creative types have found that taking the numbers from the existing, production system is not prohibited in the license agreement so they compare that to their latest, greatest hardware/software and execute or supervise the execution of a benchmark on their solution. Obviously this is a one sided apples to bicycles comparison, but quite favorable for bragging rights for the new guy.
I’ve been doing customer benchmarks and proof of concepts (POCs) for almost 5 years at Oracle. I can guarantee you that Netezza has never even come close to getting 10x-100x the performance over Oracle running on a competitive hardware platform. Now I can say that it is not uncommon for Oracle running on a balanced system to perform 10x to 1000x (ok, in extreme cases) over an existing poorly performing Oracle system. All it takes is to have a very unbalanced system with no I/O bandwidth, not be using parallel query, not use compression, poor or no use of partitioning and you have created a springboard for any vendor to look good.
One More Juicy Marketing Tidbit
While searching the Internet for creative marketing reports I have to admit that the crew at ParAccel probably takes the cake (and not in an impressive way). On one of their web pages they have these bullet points (plus a few more uninteresting ones):
- All operations are done in parallel (A non-parallel DBMS must scan all of the data sequentially)
- Adaptive compression makes disks faster…
Ok, so I can kinda, sorta see the point that a non-parallel DBMS must do something sequentially…not sure how else it would do it, but then again, I don’t know any enterprise database that is not capable of parallel operations. However, I’m going to need a bit of help on the second point there…how exactly does compression make disks faster? Disks are disks. Whether or not compression is involved has nothing to do with how fast a disk is. Perhaps they mean that compression can increase the logical read rate from a disk given that compression allows more data to be stored in the same “space” on the disk, but that clearly is not what they have written. Reminds me of DATAllegro’s faster-than-wirespeed claims on scan performance. Perhaps these marketing guys should have their numbers and wording validated by some engineers.
Do You Believe In Magic Or Word Games?
Creditable performance claims need to be accounted for and explained. Neil Raden from Hired Brains Research offers guidance for evaluating benchmarks and interpreting market messaging in his paper, Questions to Ask a Data Warehouse Appliance Vendor. I think Neil shares the same opinion of these silly benchmarketing claims. Give his paper a read.