Let’s take a moment to examine the under lying technologies to put in perspective why these each stand on their own merits. On the Flash side we have the F5100 which provides up to 1.6Million read IOPS in only 1 rack unit (RU) and has significant performance and power advantages over disk as you can see here. Later today you’ll learn more about flash’s tremendous performance advantages.
Now most of our customers already have Tier 1 storage deployed, but do you have the right disk storage? Our 6000 family of FC Arrays is already a multi-tier environment by integrating in SSDs and we have advantages over our leading competition in many ways including cost and performance.
Next, we have the ZFS Storage Appliance which seamlessly integrates DRAM and flash cache to transparently access to up to 6.7TB of fast-access media in Hybrid Storage Pools and does so at a better density and cost than all of our unified storage competitors.
Lastly, we round out the pyramid with our industry-leading tape portfolio scaling from 24 slots to over 100,000 slots (over 150PB in a single system!) to provide the highest scalability on the market, best products on the market and SIGNIFICANT advantages over disk even above and beyond the cost per GB savings. These include an impressive 290:1 energy savings over disk over a 5 year period (which is huge especially if you consider the fact some of this data needs to be around for 100 years!) and tape offers a much longer shelf life than disk so you have fewer Capital Expenditures over the long-haul and it’s even been found to be more cost effective than straight de-dup when you add all this up. And we are the highest scalability on the market by 4x along with the #1 rated in Quality and Reliability.
As you can see there are distinct advantages associated with each tier of storage and you have a lot of levers to pull to strike the right balance in your own environment between cost, capacity and performance as you look across each of these.
Source: 1) Clipper Group, 2009
Since CPUs and Memory are silicon based components they have been following Moore’s Law and getting faster and faster.
Traditional hard drives on the other hand are based on spinning platters and have not kept up the same pace.
What’s holding back you application performance? It could very well be your storage. Hard disk drives, even the fastest 15K RPM can not feed your servers fast enough. They are some 260 times slower than what today’s servers are capable of so they spend most of their time waiting for data after a request. They are starving. The traditional remedy of adding more expensive DRAM may no longer suffice as data sets double every 2 years.
Today your applications are being chocked by spinning disk drives which are causing storage latencies and I/O bottlenecks.
Flash storage technology can help bridge this gap by sitting in between the server and spinning hard disks. This allows applications to get the fast response time they require from flash while storing infrequently used data on slower HDD technology.
- Delivers Low Latency
- Solves the IO bottleneck
- Lower latency means applications respond more quickly
- Provides Higher Throughput
- More bandwidth means applications can send more data at the same time
- Requires Less Power
- Less power mean you can save on your energy bill while increasing performance
- Smaller Footprint
- More space for you to use for other projects
Just the hot data files were moved to flash, making the performance all the more impressive….
Read service time from media (db file sequential read) is the leading wait event influencing transaction response time for the OLTP test system. In the case of the hybrid flash/disk architecture, this average includes data from 2 different populations: reads from flash and reads from disk. In test system used to generate this illustration,40% of the read I/O comes from the indexes and 60% of the read I/O comes from the table data, and the I/O service time for reads from the data files.
Adding a Sun Storage F5100 to offload index I/O processing from an existing disk system to flash-based storage improves system performance in 2 ways: read service time from index files is reduced dramatically and read service times for data files is reduced noticeably. In the case of the index files, compared to a modestly loaded 15k RPM disk drive, service time drops from 15 ms to 1 ms or greater than 90%. In the case of the table data files, because index processing has been moved to the flash device there is less workload for the disk to support so the spinning disk can get the remaining work done more quickly.
Illustration 5 shows the Oracle-instance reported db file sequential read wait event verses the total front-end I/O rate. The front-end I/O rate is defined as the sum of the physical reads, physical writes, and transactions executed per second. The service time is defined as the average I/O service time over all table spaces, including the data and indexes. In the case of the test application, where about 50% of the I/O is executed against the indexes and 50% of the I/O is executed against the data, the average service time is approximately the average of the service time to each table space. In the case of migrating from spinning disk to Sun F5100 Flash Array technology, the nearly 10x reduction in service time to the index effectively halves the average service time for the system. In the lightly loaded case, average read service time drops from 6 ms to 3 ms, and as the disk begins to saturate, average read service time drops from 12 ms to 6 milliseconds.
This slide illustrates how many HDDs you would need to deliver 1.6M read IOPS if you were to stack up individual HDDs by themselves without intervening disk controller technology.
This comparison is based on using 300GB 15K RPM FC disks and shows what happens to power and space (not to mention cost) when administrators are forced to “short stroke” their disk drives in order to sustain very low latency requirements of certain database driven applications.
* Calculation assumes 350 IOPS and 17 watts per high performance HDD
This internally developed model coincides with outside analyses. The $/GB projections are based upon 1) current component street prices divided by capacity; 2) HDD areal density improvements will continue at current rates (43% YOY); and, 3) Moore’s Law continues to hold for denser NAND chips used in SLCs (2x transistors every 2 years). Bottom line: HDDs to be history within 10 years (15K drives within 7 years; 7200 RPM drives within 10 years!). MLC is significant because it is rapidly approaching enterprise grade and may hasten 15K demise in IT shops before SLC does in 2017. There are several performance-based use cases that easily cost-justify flash today…short stroking for performance is an excellent one covered in a few moments.
Oracle has the broadest range of Flash Storage products in the industry. We offer standalone Flash Arrays as well as flash based components that are installed inside our server and other storage products.
Review specs on the slide
By adding these PCIe cards to your server you can dramatically increase IO response times. They are especially useful when deploying a grid architecture since they allow you to scale your performance linearly across a multitude of servers. Based on enterprise grade SLC flash technology they offer advanced wear leveling algorithms and use super capacitors to eliminate the need to rely on batteries to store data in the event of power loss.
Flash Characteristics
Better response time than disk (>10x faster for reads)
More I/O per second than disks (>100x for reads)
Cheaper and higher capacity than memory (less than $60K per TB)
Flash cache can also be configured to use flash disks on storage arrays. Retains advantage of automatic management for what is in flash, but potentially loses response time and throughput.
When data is written in the database the buffer cache is used first.
Then the dirty blocks are sent to the spinning disks.
Clean blocks are sent to the flash storage for quick retrieval.
When a block is requested for a read the software knows to check the flash storage when the data is not found in the buffer cache.
The Oracle Sun 6780 delivers best of breed performance in FC SAN environments.
The 6780 can support 170M email messages /hr, 92M documents/hr, 1.6M high-res (10MB) images/hr and 160,000 video files/hr.
(Note: waiting for a few more notes from Bill Leslie on the value prop around managing all different kinds of database data)
Exadata V2 from Oracle uses multiple F20 PCIe cards in each node in conjunction with Exadata Smart Flash Cache to manage data efficiently and deliver superior performance. The database is able to take the data that is most likely to be reused and place it in flash which dramatically improves database response times.
Oracle 11g Partitioning improves database performance and lowers storage cost by moving data to lower cost tiers of storage based on data access classification and data value.
Oracle storage is optimized for ILM Partitioning by offering best of breed storage in Tier 0 Flash, Tier 1 and 2 disk and Tier 2 tape.
Oracle Storage Archive Manager software (SAM) enables tiered storage of backups behind ILM Partitioning transparent to the database≥
© 2009 – 2011, www.oracledatabase12g.com. 版权所有.文章允许转载,但必须以链接方式注明源地址,否则追究法律责任.
相关文章 | Related posts:
- RMAN Performance Tuning Using Buffer Memory Parameters
- PeopleSoft Enterprise Performance on Oracle 10g Database
- QREF: Library Object Lock State Object
- TROUBLESHOOTING: Possible Causes of Poor SQL Performance
- Oracle AIX Performance and Tuning
- Oracle Streams Performance Advisor
- Yet Another Performance Profiling Method (Or YAPP-Method)
最新评论