- Breakthroughs of technology thanks to the realization of Moore's law has provided continued improvement in raw horsepower available under the covers. Oracle's acquisition of Sun's hardware platform gave them an inside track in to this fast moving domain of development. The Oracle Exadata machine is being specifically touted as being a Data Warehouse appliance. Certainly this can be true due to the robust configuration of even the entry level unit. However, I believe the advantages available of the same appliance device to OLTP environments cannot be minimized.
* CPU
Each Quarter Rack server contains 2 Quad-Core Intel® Xeon® E5540 Processors running at 2.53 GHz. There are faster clocked CPUs on the market both dual and quad core configurations. But the quad core configuration, even though under 3.0 Ghz, is still very good in parallelization activity. In comparison to other CPUs currently on the market, the E5540 lands in the top 20%.
In fact, that may be the good news that CPU upgrades are likely in the future boosting performance yet again at a machine level.
Each server carries 2 - 4 core CPUs.
* Memory
Historically increasing growth of memory structures in size due to changing price points in regards to speed and capacity.
The large memory structures provide performance improvement by providing an I/O free lookup to frequently accessed objects. The Exadata database server continues to provide ample head-room for most situations.
Each Exadata server contains 72 gigabytes of DRAM, 576 gigabytes for a full-rack configuration of 8 servers.
* I/O
Integrating flash memory (SRAM?) for read caching can eliminate some I/O entirely. Capable of using either SAS and SATA attached storage. The high-speed Infiniband interconnect between storage and database at 40 Gb/sec. The Automatic Storage Management feature (ASM) is integrated into the file system and volume management activities.
* Storage
Technological advances in speed and density of storage devices has increased dramatically allowing for increased performance and scalability. Sophisticated striping and mirroring strategies, along with parallelization and snapshoting create environments with great recoverability. The Storage Server role is expanded with the Smart Scan feature. This feature allows for data rows to be eliminated from the retrieval task at the storage level. This eliminates the retrieval of unneccsary data and likely should reduce the occurence of "throw away" rows identifiable in a 10046 level trace file.
The Storage Server continues to expand it's role through the use of Storage Indexes. This new structure basically contains max and min vaues for commonly queried columns. This new structure can basically be considered as a form of Histogram. Testing will be interesting to see how the impact of histograms changes, perhaps, in an Exadata world.
One of the largest contributors to the improved Exadata Storage Server performance is the use of solid-state disk (SSD) devices as non-volitale storage. These SSD devices are effectively placed between the memory and disk access for interim storage. Because this solution is devised as a hardware solution there are no complicated software requirements. The position of the SSD cells in front of the disk devices bypasses the disk controller. Because the frequently accessed elements are coming from the SSD, the disk activities are deferred. The overall strategies for maintaining the SSDs and disks is sophisiticated, and designed to reduce flushing and reload requirements. Tighter controlling cache management strategies can help reduce CPU load.
Each Storage Server contains an internal card with 384 gigabytes of "flash memory" SSDs. A complete rack of Storage Servers has over 5 Tb.
* Database
The optimization of database activity has traditionally been centered around the expectation of a workload of mostly writes and random reads I/O. In the case specifically of data warehousing the database patterns of activity might not be the same. In many cases query workload covers a much larger number of rows with fewer columns.
Oracle has introduced a new strategy that incorporates hybrid columnar compression. This option allows the system to be configured such that any table or partition may be marked for compression. In addition to the favorable use in DW environments, this feature is likely useful in regard to any archive areas. This is a point of integration between the database and the storage areas because of the Smart Scan's abilities to interpret the data requirements. DW tasks may only need to read the compressed data, based on columns referenced, sending the results without that column for subsequent processing. Certainly a significant workload with RT in many DW environments - nature of the query.
Tying the knot - benefits.
These sophisticated elements are integrated to together allowing a fabric the can flex to meet the processing requirements. Early benchmarks show Exadata as being dramatically faster than anything else on the market.
Nice Oracle made it simple: Small, Medium, Large. Large is commonly referred to as Full Rack. So it's measured as Quarter Rack, Half Rack and Full Rack.
One throat to choke. Oracle provides the Database Server, the Storage Server and rumors are that there is a mid-tier Server in the works. With Sun's history of hardware deployment this would seem to be simple compared to the work done with first HP and now internal Sun.
0 comments:
Post a Comment