In case you don’t have budget to buy Exadata you can still buy huge number of flash disks and put on them part of your database. But what should be stored on flash disks(very fast) and what on magnetic disks(very slow) ?
It’s not your businesses to know let decide database 🙂
Introduction
DB Smart Flash Cache is new extension for buffer cache area. This extra area should be defined on solid state disks (SSD) and has following features:
-
make performance improvements at moderate cost(cheaper than DRAM)
-
low latency compared to magnetic disks
-
higher throughput compared to magnetic disks
- easy to setup
- easy to control
- can be used for RAC cache fusion keeps consistency
- direct I/O bypasses buffer cache so as well bypasses DB smart flash cache
- can cache only clean blocks from buffer cache
- flash cache is not auto-tuned
- only blocks from standard buffer pool are cached in DB smart flash cache
Oracle recommends:
-
flash disks should have comparable read IOPs and IOPs write
-
this new layer should be at least 2-10 times bigger than buffer cache in DRAM
- mainly for OLTP systems
Architecture
if a oracle server process needs to read a block from database at first it must read it from magnetic disk(physical read). Then the block is stored in buffer cache memory and added to standard “LRU chain” list.
When “in memory buffers area” is getting full Oracle must decide which blocks needs to be removed from cache. If you have DB Smart Flash Cache enabled “clean” blocks are written to “Flash cache” by DBWR process so next time they can be read into memory from Flash Cache and improve your performance.
NOTE: “Dirty” blocks are never stored in Flash Cache
List of blocks cached in DB smart flash cache are stored in buffer cache area on two dedicated flash “LRU lists” depending on object attribute FLASH_CACHE:
-
DEFAULT – standard last recently used algorithm decides how long such blocks are cached in flash cache. It’s default value assigned to each object in database.
-
KEEP – such blocks are not removed from flash cache as long as the flash cache is large enough
alter|create table|index objectname storage ( buffer_pool { keep | recycle | default } flash_cache { keep | none | default } );
NONE value for FLASH_CACHE is blocking flash caching for a given object.
Statistics
All I/O operations from DB smart flash cache are counted as physical I/O however Oracle also collects such informations in new columns.
V$SQL - OPTIMIZED_PHY_READ_REQUESTS V$SQLAREA - OPTIMIZED_PHY_READ_REQUESTS V$FILESTAT - OPTIMIZED_PHYBLKRD
select name from v$statname where name like 'physical%optimized%'; NAME ---------------------------------------------------------------- physical read requests optimized physical read total bytes optimized
You can see such stats in V$SESSTAT and V$SYSSTAT
Setup
Two parameters must be set on database level to turn on DB smart flash cache:
- DB_FLASH_CACHE_FILE – defines (OS disk path or ASM disk group) and file name to store this data
- DB_FLASH_CACHE_SIZE – defines size of the flash cache
DB_FLASH_CACHE_FILE='/os path/flash_cache_file.dbf' DB_FLASH_CACHE_FILE='+FLASH_DISK_GROUP/flash_cache_file.dbf' DB_FLASH_CACHE_SIZE=200m
After setting both parameters you need to restart database.
DB_FLASH_CACHE_FILE
-
can’t be shared between many databases or instances
DB_FLASH_CACHE_SIZE
-
can’t be dynamically resized
-
can be set to 0 to disable DB smart flash cache
-
can be set to original size to re-enable DB smart flash cache
Performance improvements
Oracle conducted interesting test for a OLTP database 70GB size with 8GB SGA. From below picture you can see improvements for Transactions versus size of DB smart cache size.
Following picture shows improvement in transaction response time versus DB smart cache size
Example
I simulate SSD disk by creation ramdisk – disk based in memory using following steps:
1. create directory to mount ramdisk and change owner to oracle and group dba
[root@oel5 /]mkdir /ramdisk [root@oel5 /]chown oracle:dba -R /ramdisk
2. mount ramdisk and check it
[root@oel5 /]# mount -t tmpfs none /ramdisk -o size=256m [root@oel5 /]# mount | grep ramdisk none on /ramdisk type tmpfs (rw,size=256m)
3. set parameters for database and restart it as user oracle
SQL> alter system set db_flash_cache_file='/ramdisk/ram.dbf' SQL> scope=spfile; System altered. SQL> alter system set db_flash_cache_size=200M scope=spfile; System altered. SQL> startup force ORACLE instance started. Total System Global Area 835104768 bytes Fixed Size 2232960 bytes Variable Size 507514240 bytes Database Buffers 322961408 bytes Redo Buffers 2396160 bytes Database mounted. Database opened. SQL> show parameter flash_cache NAME TYPE VALUE ----------------------- ----------- ------------------------------ db_flash_cache_file string /ramdisk/ram.dbf db_flash_cache_size big integer 200M
4. Check new file exists in /ramdisk
[root@oel5 ramdisk]# ll total 8 -rw-r----- 1 oracle asmadmin 209715200 Feb 24 22:54 ram.dbf
5. Let’s create tables with flash_cache keep reference in storage clause so Oracle will try to keep the blocks in DB smart cache as long as possible.
create table test_tbl1 (id number, id1 number, id2 number) storage(flash_cache keep); begin for i in 1..1000000 loop insert into test_tbl1 values(i, i, i); end loop; commit; end; /
6. Eventually after some time you can see some data in flash cache – v$bh view.
select status, count(*) from v$bh group by status; STATUS COUNT(*) ---------- ---------- xcur 36915 flashcur 25583 cr 13
7. If you clean buffer cache as well db smart flash cache is purged
alter system flush buffer_cache; system FLUSH altered. STATUS COUNT(*) ---------- ---------- xcur 520 free 36411
Have a fun 🙂
Tomasz
Hi!
I do all steps of your manual, but after ‘startup force’ I have an error:
SQL> startup force
ORA-00439: feature not enabled: Server Flash Cache
What am I doing wrong?
My configuration:
Oracle Linux Server release 6.4
2.6.39-400.24.1.el6uek.x86_64
Oracle Database 11g 11.2.0.3.0
I found answer! 🙂
Patch 12949806: FLASH CACHE CHECK IS AGAINST ENTERPRISE-RELEASE
Now it works!
Pingback: [Oracle] 11g的SSD功能Smart Flash - MySQL, Oracle Abc - The Database Wiki
2. mount ramdisk and check it
[root@oel5 /]# mount -t tmpfs none /ramdisk -o size=256m
[root@oel5 /]# mount | grep ramdisk
none on /ramdisk type tmpfs (rw,size=256m)
I could not understand step 2. what does this command do????
please explain each command of step 2.
I simulate SSD disk by creation ramdisk. Ramdisk is created in physical memory and available in system as disk.
That’s all.
Regrads
Tomasz