Flash Cache is not Flash Disk. Right ?

Question

Hardware: X3 - Half Rack
# imageinfo
Kernel version: 2.6.32-400.11.1.el5uek #1 SMP Thu Nov 22 03:29:09 PST 2012 x86_64
Image version: 11.2.3.2.1.130109
Image activated: 2013-09-24 18:12:52 +0800
Image status: success
System partition on device: /dev/mapper/VGExaDb-LVDbSys1
In our Non-Exadata RAC DBs we use LUNs from Hitachi VSP frame. For very busy DBs like Siebel , we wanted to place our Online Redo logs on high speed disks.
So, our Hitachi SAN admins have allocated LUNs exclusively from Flash Disks within Hitachi VSP frame and we have created Redo Log diskgroup based on these 'Thick provisioned' disks.
This has increased the performance of our RAC DBs significantly.
As I understand, In exadata, you don't have the normal flash disk. Instead, the flash modules are mounted in a PCIexpress card.
In CellCLI> list physicaldisk detail you can see multiple entries like below belonging to the 'flash disk'
          name:                   FLASH_1_0
         diskType:               FlashDisk
         luns:                   1_0
         makeModel:              "Sun Flash Accelerator F40 PCIe Card"
         physicalFirmware:       UI39
         physicalInsertTime:     2013-09-13T17:56:10+08:00
         physicalSerial:         5L0066Y5
         physicalSize:           93.13225793838501G
         slotNumber:             "PCI Slot: 1; FDOM: 0"
         status:                 normal
Although these flash cards can store persistent data because it has battery in it in server goes down, is it actually used to store persistent data.
Smart flash cache is just a caching mechanism for frequently accessed data. Right ? We cannot create diskgroups based on disks from these 'Flash Disks' . Right ?

Answer 1

Hi GaryB,
The short answer to your question is:  yes, it's possible to use configure flash as permanent storage using "flash griddisks".  But in the case of redo logs, it would actually make them perform slower.
The longer explanation:  foreground redo writes (the "log file sync" wait event) are very latency sensitive, so it's important for write to be acknowledged as fast as possible.  Flash is considerably faster than disk, but not nearly as fast as battery-backed RAM.  And Exadata storage servers do have a RAM write cache on their disk controllers.  So you will get the best foreground redo write performance from RAM cache write, except when the write cache is full, at which point redo writes need to wait on disk. 
Since 11.2.2.4.0 was released in 2012, Exadata has the capability to write redo to _both_ RAM-cached disk and flash memory, and use whichever returns first.  So you get the lightning-fast latency of RAM cache unless the cache is full, and worst-case latency is the speed of flash.  This capability, smart flash logging, is enabled by default on new Exadata builds;  you can check its status by running "view flashlog detail" in CellCLI.
More details in the official whitepaper: http://www.oracle.com/technetwork/database/exadata/exadata-smart-flash-cache-366203.pdf
Hope this helps!
Marc