Not sure if this is allowed as it's not a question and it's not directly about SQLA, but I'm sure Breck will exercise his benevolent dictatorial powers appropiately if it's inappropiate :)
I noticed the announcements from Oracle about the Exadata technology that effectively means that some of the data selection (rows and columns) happens at the "data server" ie SAN etc, rather than at the database server, giving scope for drasticaly reducing the time need to process the data.
In the Oracle white paper on the Exadata thing, database awareness (pre-filtering of result sets) is only one of the bragging points; here's the full list:
And fine bragging points they are! Here are some excerpts...
(Note: In the following "Exadata Storage Server" is the same as "Exadata cell").
"Each Exadata cell comes preconfigured with: two Intel Xeon E5540 quad-core processors, 384 GB of Exadata Smart Flash Cache, twelve disks connected to a storage controller with 512MB battery-backed cache, 24 GB memory, dual port InfiniBand connectivity, management interface for remote access, dual-redundant hot-swappable power supplies, all the software preinstalled, and takes up 2U in a typical 19-inch rack."
(Note: The white paper refers to a three-cell setup as a "small environment")
"The Sun Oracle Exadata Storage Server comes with either twelve 600 GB Serial Attached SCSI (SAS) disks or twelve 2 TB Serial Advanced Technology Attachment (SATA) disks. SAS based Exadata Storage Servers provide up to 2 TB of uncompressed user data capacity, and up to 1.5 GB/second of raw data bandwidth. SATA based Exadata Storage Servers provide up to 7 TB of uncompressed user data capacity, and up to 0.85 GB/second of raw data bandwidth. When stored in compressed format, the amount of user data and the amount of data bandwidth delivered by each cell increases up to 10 times. User data capacity is computed after mirroring all the disk space, and setting aside space for database structures like logs, undo, and temp space. Actual user data varies by application."
"Queries that perform table scans can be processed within Exadata with only the required subset of data returned to the database server. Row filtering, column filtering and some join processing (among other functions) are performed within the Exadata storage cells."
"Often a ten-fold speed up in these operations is seen when using Exadata storage compared to storage products traditionally used with the Oracle Database ― but in many cases a 50-fold, or greater, speedup is achieved."
So here's a question: When will we see this setup hit the top of the charts, in the Top Ten TPC-C by Performance? The current Number One is from Sun - Oracle and it can be yours for Christmas for a mere $18M, but... it's not the Exadata thing, and it only does 7.2M transactions per minute :)
And here's a real question: Can the ideas of parallelism and storage-side database awareness be applied to SQL Anywhere? Say, perhaps a simple approach, maybe aiming low, a 5x improvement without adding administrative baggage? Maybe using hardware available here on Planet Earth? Like the stuff Google uses?
BTW, SQL Anywhere is still in the Top Ten TPC-C by Price/Performance... 20K transactions per minute for $17.5K.
Those are "K"s, by the way. As in $1,000s not $1,000,000s.
answered 24 Nov '09, 11:37