Enable Cache Fusion in Oracle RAC

Hi gurus,
I cannot find on google how to enable and test Cache Fusion feautre in Oracle RAC. Could you help me please?
Best.

I don't know. I cannot find parameter which enable or disable CFAs Aman has already stated the feature is already present
http://download.oracle.com/docs/cd/B28359_01/server.111/b28318/consist.htm#CNCPT1317
I even cannot find information how to test CF to ensure that it really works?!http://download.oracle.com/docs/cd/B28359_01/rac.111/b28254/monitor.htm#RACAD981
@Aman
Cache Fusion is the technology which makes the 10g RAC, 10g RAC. And 11g Database as well :)
Edited by: Amy De Caj on Jul 19, 2009 4:26 AM
Edited by: Amy De Caj on Jul 19, 2009 4:28 AM

Similar Messages

Row cache wait in Oracle RAC

Hi,
Our application is running on Oracle RAC. During certain time of the day, the applcation responds very slowly. At these times, it is observed that the row cache waits are very high. We have even tried altering the sys.AUDSES$ sequence and changing it cache size to 10000 from default 20, but this did not help.
Can anyone suggest a solution for this problem? And why this problem occurs?

Hi,
it looks like your problem is related to the fact that you do not cache sequences (this is a well know RAC tuning topic).
Oracle introduced sequences (wrong name, definitely) to generate unique numbers, not to actually support a time sequence of events, or to preserve an order or to have ascending sequences of numbers with no gaps.
Ordering a sequence of events is a serialization process that should not be implemented by a sequence.
Now, if you do not cash sequences, in RAC the lock (enqueue) on the sequence (that is required when you ask for the next set of values) is a global resource on which inter instance contention occurs.
Furthermore, in case the application has a high volume of inserts, having nocache sequences leads to inter instance index block contention.
Oracle says that the default cache value of 20 for sequences is inappropriate in most case of RAC implementations and it is frequent to have caches of 1000 values or more. You need to test what is your ideal value.
Now it is up to you to decide between:
- keep things as they are and have a non scalable RAC installation
- find a way to cache sequences without harming the application assumptions.
Hope it helps,
Regards,
Corrado

Disable RAC Cache Fusion in 10r2

Hi all,
is it possible to disable the Oracle Cache Fusion in Oracle Database 10.2.0.4 Database ?
Regards,

I think it is nonsense to think about disabling cache fusion. and it is almost not possible. On the other hand, how can RAC work without this feature ? How can you imagine the work of your RAC system without global synchronization between nodes ???
Regarding to singleton concept which is mentioned by guys, it means that the idea here is concentrating the database service which your application uses on one node, so mostly your application will be served by this node, which means the roundtrip between instances will be low. You can think about this state just like you are working with single instance database, as here cache synchronization overhead is almost eliminated. Basically RAC itself also does remastering of resources, based on workload dynamically, but remember that if your database service is served by several instances, your connections could be directed to both of them and in inpredictable fashion, based on load balansing goal to be exact.
Besides all of this, there are several wait event classes spesific for RAC and corresponding views for them. you can explore them and find the idea about which wait classes are bottleneck.

Oracle 9i rac cache fusion queries

Hi RAC experts,
Would like to understand more on the background of cache fusion.
Kindly advise on the following cache fusion operation as i'm having some confusion.
1. read/read - user on node 1 wants to read a block that user on node 2 has recently read.
in this case, the block is already on the cache of node 2 since it is recently read. so does node 1 read from disk to store in it's down cache since the info is not in it's cache? or it can request directly from the remote cache in node 2 to do shipping of the block to it's own cache for select?
2. read/write - user on node 1 want to read a block that user on node 2 has recently updated.
in this case, node one will get the recently updated info from remote cache(node 2) to it's local cache for update right?
3) write/read - user on node 1 want to update a block that user on node 2 has recently read.
does node 1 read from disk since the info is not in it's cache? or it can request directly from the remote cache in node 2 to do shipping of the block to it's own cache for update?
thanks
junior dba

1. read/read - user on node 1 wants to read a block that user on node 2 has recently read.Since the block is in another instance's buffer cache, the questing instance(1 in your case) will request the block to be shipped from node 2. It will avoid going to disk.
.2. read/write - user on node 1 want to read a block that user on node 2 has recently updated.in this case, node one will get the recently updated info from remote cache(node 2) to it's local cache for update right?
That's correct.
oes node 1 read from disk since the info is not in it's cache? or it can request directly from the remote cache in node 2 to do shipping of the block to it's own cache for update?Same as in 1) above, it gets from the other's instance.
These concepts have been explained well in RAC Handbook by K Gopalakrishnan. Here is the additional information on the book. http://www.amazon.com/Database-Application-Clusters-Handbook-Osborne/dp/007146509X/ref=pd_bbs_sr_1?ie=UTF8&s=books&qid=1211888392&sr=8-1
HTH
Thanks
-Chandra

Oracle RAC Cache Fusion , Insert intensive tables.

Hi,
I have one general questions in RAC cache fusion. It was observed that there are too much interinstance contention as the insert intensive queries are executed. What is the best possible solution to solve this. I mean how to avoid the contention. Is there any general approach.
Thanking you
Rocky

As others have pointed out, the first step would be to determine what object or objects was actually causing contention.
If we're talking about a hypothetical OLTP application with 400 concurrent users on 2 nodes of a RAC cluster, inserting into a heap-organized table with no bitmap indexes I would wager that the most likely cause of contention would be contention for index blocks on monotonically increasing columns (i.e. sequence-generated primary keys). Since indexes are sorted structures, the next row will almost always be going into the right-most block of the index. If both nodes are fighting over who has that right-most block at any instant, that can cause lots of contention as the block gets moved back and forth between the nodes. If you don't need to do range scans on the index, the solution would be to change the primary key index to be a reverse key index so that inserts are spread across a larger number of index blocks at any instant. Of course, this is just one possible scenario-- there are dozens if not hundreds or thousands of possible permutations. It comes down to figuring out what resource the various sessions are contending on and then figuring out how to reduce that contention without screwing up anything else too badly.
Justin
Edited by: Justin Cave on Jun 19, 2009 12:50 AM

[RAC] 9I REAL APPLICATION CLUSTERS의 CACHE FUSION 에 대한 이해

제품 : ORACLE SERVER
작성날짜 : 2004-08-13
[RAC] 9I REAL APPLICATION CLUSTERS의 CACHE FUSION 에 대한 이해
==========================================================
PURPOSE
이 문서는, 오라클 리얼 애플리케이션 클러스터 환경의 캐쉬 퓨전의 기능과 장점을
설명하는 것을 목적을 한다.
SCOPE
Real Application Clusters(RAC) Option은 9i Standard Edition에서는
지원하지 않는다.
Explanation
Oracle 8i OPS에서 캐쉬 퓨전의 개념이 소개 되었다. 이 기능은, 원격 인스턴스가 관장하는
블럭에 대한 읽기 일관성 있는 뷰를 위해 디스크를 통한 블럭의 pinging을 감소 시키는 목적으로
추가되었다. 이 기능은, 다른 인스턴스에 의해 lock이 걸린 데이터를 select하는데 필요한 시간을
현격하게 감소 시켜 주었다.
lock을 건 인스턴스가 변경 사항을 기록하도록 강제 하여 (I/O 발생을 강제 발생) 원격 인스턴스가
읽을 수 있게 하는 대신, 캐쉬 퓨전은 버퍼의 사본을 생성하여, 노드간 고속 연결 라인을 통해
데이터를 조회하는 측에 전달 시켜준다. 이 개념은, 읽기/쓰기 경합에 따른 성능상의 문제를
경감시켜 주기는 하나, 블럭의 내용을 변경시켜야 한다면, 원격 인스턴스에서 해당 블럭에 대해 쓰기를
하도록 한 후 읽어 들여 사용하는 기존 방식과 동일한 ping 메카니즘을 사용 해야 했다.
9i 리얼 애플리케이션 클러스터에서는, 쓰기/쓰기 경합 문제를 해결할 수 있도록 캐쉬 퓨전이
설계되었으므로, 성능 향상 및 확장성이 개선 되었다. RAC cache fusion을 통해 Oracle 에서는
다른 인스턴스에서 lock이 걸린 블럭에 대해 디스크를 통한 'ping'이 개념상으로는 더이상
발생하지 않도록 함으로써, I/O의 양을 더 줄여 주게 되었다. 대신 RAC 인스턴스는, 더티
버퍼의 사본에 대해 원격 인스턴스에서 쓰기 권한을 가질 수 있게 해 주었다. 이 기능은
dba (1:1) 를 사용하는 realesable locking (특정 데이터 파일에 gc_files_to_locks 값을
0으로 지정하거나, gc_files_to_locks 값을 아예 설정하지 않을 경우 사용 가능) 에서만
적용되는 기능이다. Hashed lck은 ping 메카니즘을 이전 방식과 동일하게 사용하며, fixed lock은
9i RAC에서는 더이상 사용 할 수 없게 되었다. (잇점이 없기에)
앞에서 언급한 바와 같이 RAC 캐쉬 퓨전에서는 인스턴스가 dirty buffer의 사본을 사용할 수 있게
해 준다. RAC 캐쉬 퓨전에서 새로 등장한 개념은 past image라는 개념인데 이것은 버퍼의 이전
사본이 디스크게 기록되지 않은 상태를 말한다. 이 past image를 추적하기 위해 오라클에서는 전역
락과 로컬 락의 role과 BWR (block written redo)를 사용한다. global 과 local lock의 role을 명확히
하기 위해 Oracle 8i의 lock을 local lock으로 생각해보자. 이 경우 오라클 8i에서는 3가지 종류의
락 모드가 존재한다 :
N - Null
S - Shared
X - Exclusive
오라클 9i RAC의 lock모드를 언급할 때는 3 글자로 lock의 종류를 구분한다. 첫번째 글자는 lock mode를
나타내고, 두번째 글자는 lock의 role, 세번째 글자는 (숫자) 로컬 인스턴스에 past image가 있는지
여부를 나타낸다. 이와 같은 방식으로 lock의 종류는 다음과 같이 나타낼 수 있다 :
NL0 - Null Local 0 - 8i의 N과 동일 (past image 없음)
SL0 - Shared Local 0 - 8i의 S와 동일 (past image 없음)
XL0 - Exclusive Local 0 - 8i의 X와 동일 (past image 없음)
SG0 - Shared Global 0 - 글로벌 S 락, 인스턴스에 current block image 소유
XG0 - Exclusive Global 0 - 글로벌 X 락, 인스턴스에서 current block image 소유
NG1 - Null Global 1 - 글로벌 N 락, 인스턴스에서 past image 소유
SG1 - Shared Global 1 - 글로벌 S 락, 인스턴스에서 past image 소유
XG1 - Exclusive Global 1 - 글로벌 X 락, 인스턴스에서 past image 소유
lock이 최초로 획득된 경우, 로컬 role로 획득된다.만약 lock이 획득된 시점에 원격 인스턴스에 더티
버퍼가 존재한다면, 글로벌 락 role을 가지고 획득하며, 이 경우, 더티 버퍼를 모든 원격 인스턴스에서는
버퍼의 'past image'로 취급한다. 리커버리를 위해 past image를 가지고 있는 인스턴스에서는
lock을 점유하고 있는 마스터 인스턴스에서 lock을 해재했다고 알려 오기 전 까지는 past image를
버퍼 캐쉬에 저장한다. 버퍼내의 블럭이 폐기 될 때는 past image를 관리하는 인스턴스에서 DBR
또는'block written redo'를 redo 스트림에 기록하게 되는데 이것은, 블럭이 이미 디스크에
기록되었으므로, 이 인스턴스를 recovery하는데는 필요하지 않다는 것을 나타낸다.
RAC 클러스터의 3번 노드가 EMP 테이블의 블럭을 관장하는 락 엘리먼트 123을 소유하고 있다고
가정해 보자 :
사용자 C가 인스턴스 3에 연결된 상태에서 EMP 테이블에 대한 select를 하면서 SL0 락이
오픈 되었다 :
| Instance 1 | | Instance 2 | | Instance 3 |
| | | | | |
| Lock Held | | Lock Held: | | Lock Held: |
| on LENUM 123: | | on LENUM 123: | | on LENUM 123: |
| | | | | SL0 |
| | | | | |
Shared lock을 획득하는 것은 락 role에 아무런 영향을 미치지 않는다. 따라서 만약 인스턴스
2번에 연결된 사용자 B가 동일한 EMP 테이블에 대해 select를 하더라도, 인스턴스 2의 락
엘리먼트 123의 락 모드는 동일하다. ( S 락이기 때문임 ) 그리고 락의 role 또한 동일한데
버퍼 내용에 더티 상태로 변경된 내용이 없기 때문이다 :
| Instance 1 | | Instance 2 | | Instance 3 |
| | | | | |
| Lock Held | | Lock Held | | Lock Held |
| on LENUM 123: | | on LENUM 123: | | on LENUM 123: |
| | | SL0 | | SL0 |
| | | | | |
첫번째 exclusive 락을 획득하는 것은 락의 role에 영향을 미치지 않는데 이 경우에도
락 엘리먼트에 대한 더티 버퍼가 없기 때문이다. 따라서 만약 사용자 B가 EMP 테이블의 락
엘리먼트 123에 속하는 row를 update 하더라도, XL0이 획득되고, 이미 획득된
SL0 락은 제거된다. 이것은 8i OPS와 동일한 동작이다.
| Instance 1 | | Instance 2 | | Instance 3 |
| | | | | |
| Lock Held | | Lock Held | | Lock Held |
| on LENUM 123: | | on LENUM 123: | | on LENUM 123: |
| | | XL0 | | |
| | | | | |
원격 노드상에 더티 버퍼가 존재하는 락 에리먼트에 대해 exclusive 락을 회극하는 것은 캐쉬
퓨전에서 2단계로 넘어가도록 한다. 이 경우 인스턴스 1번에 연결된 사용자 A가
락 엘리먼트 123번에 속하는 EMP 테이블의 row를 update 할 때, 인스턴스 2번의 블럭은
버퍼 캐쉬 내에서 더티 상태로 존재하므로, 인스턴스는 블럭에 대한 사본을 생성 하여 인스턴스
1번에 전달 한 후, 인스턴스는 null 글로벌 락을 같는다 (page image)*. 동시에
인스턴스 1은 exclusive, globally dirtied lock을 획득하며, 인스턴스 2는 past
image를 유지한다.
* 인스턴스에서 블럭에 대한 자신만의 page image를 갖는 것은, 블럭이 디스크에 게록되지 않고
또, 마스터 노드에서 알려주기 전 까지는 이미지를 폐기 하지 않기 위해서이다.
| Instance 1 | | Instance 2 | | Instance 3 |
| | | | | |
| Lock Held | | Lock Held | | Lock Held |
| on LENUM 123: | | on LENUM 123: | | on LENUM 123: |
| XG0 | | NG1 | | |
| | | | | |
이제 인스턴스 3에 연결된 사용자 C가 EMP 테이블을 select 하려고 하며, EMP 테이블이
락 엘리먼트 123에 속해있는 경우를 살펴보자. 사용자 C가 select 명령을 수행 시키면
인스턴스 1의 락은, S 락으로 하향 조정되며, 가장 최근의 buffer 내 사본을 갖고 있으며
인스턴스 2번은 그 이전의 버퍼에 대한 past image를 유지하게 된다.
| Instance 1 | | Instance 2 | | Instance 3 |
| | | | | |
| Lock Held | | Lock Held | | Lock Held |
| on LENUM 123: | | on LENUM 123: | | on LENUM 123: |
| SG1 | | NG1 | | SG0 |
| | | | | |
이제 인스턴스 2번의 사용자 B가 EMP 락 엘리먼트 123에 대해 select를 하는 경우를 가정해
보자. 인트넛느 2는 다른 인스턴스로 부터 읽기 일관성 있는 버퍼의 사본을 요청하게 된다.
인스턴스 2번에 버퍼 내용을 전달해 줄 인스턴스는 다음과 같은 순서에 의해 결정된다:
1. lock에 대한 마스터 인스턴스.
2. 가장 최후의 page image를 S lock으로 점유하고 있는 인스턴스.
3. Shared local 상태로 lock을 점유한 인스턴스.
4. 가장 최근에 S lock를 부여 받은 인스턴스.
인스턴스 1번이 락에 대한 마스터 인스턴스라고 가정해 보자 ( 그리고 가장 최근의 past image를 소유 )
이 경우 인스턴스 2는 인스턴스 1번의 버퍼 캐쉬에 저장된 블럭 사본을 전달 받고, 인스턴스 2번이
SG1 (past image에 대한 일기 일관성을 갖춘 사본) lock을 획득한다. 나머지 노드는 동일한 상태가
유지된다:
| Instance 1 | | Instance 2 | | Instance 3 |
| | | | | |
| Lock Held | | Lock Held | | Lock Held |
| on LENUM 123: | | on LENUM 123: | | on LENUM 123: |
| SG1 | | SG1 | | SG0 |
| | | | | |
이제 인스턴스 3번에 연결된 사용자 C가 테이블의 락 엘리먼트 123의 row를 upate 하고자 한다고
가정해 보자. 사용자 C는 exclusive lock을 요청하며, 인스턴스 1번과 2번은 소유하고 있던 lock이
하향 조정된다. 결과적으로는 인스턴스 3번이 XG0 lock을 점유하게 되고 인스턴스 1, 2번은
NG1을 각각 점유 하게 된다:
| Instance 1 | | Instance 2 | | Instance 3 |
| | | | | |
| Lock Held | | Lock Held | | Lock Held |
| on LENUM 123: | | on LENUM 123: | | on LENUM 123: |
| NG1 | | NG1 | | XG0 |
| | | | | |
인스턴스 3번에서 체크 포인트가 발생하는 경우, 모든 더티 버퍼의 내용을 디스크에 기록하게 된다.
이 때 인스턴스 3은 마스터노드가 기록을 하였음을 다른 노드에게 알려주게 된다. 이 경우 인스턴스
1, 2번은 past image를 폐기 처분할 수 있게 되며 인스턴스 3번만 XL0 락을 점유하게 된다.
유의해야 할 점은 인스턴스 1, 2가 BWR(block written redos)을 각각의 리두 스트림에 기록하여
해당 블럭이 이미 디스크에 기록되었으며, 인스턴스 리커버리에는 필요하지 않음을 기록한다는 점이다.
| Instance 1 | | Instance 2 | | Instance 3 |
| | | | | |
| Lock Held | | Lock Held | | Lock Held |
| on LENUM 123: | | on LENUM 123: | | on LENUM 123: |
| | | | | XL0 |
| | | | | |
Example
Reference Documents
Note:139436.1 - Understanding 9i Real Application Clusters Cache Fusion
Note 144152.1 - Understanding 9i Real Application Clusters Cache Fusion Recovery
Note 139435.1 - Fast Reconfiguration in 9i Real Application Clusters

Hi, how are you ?
Well, Oracle8i greatly improved scalability for read/write applications through the introduction of Cache Fusion. Oracle9i improved Cache Fusion for write/write applications by further minimizing much of the disk write activity used to control data locking.
It�s this.
If you still having doubt, please call me.
Regina Vidal

RAC Cache Fusion

Hello,
RAC databases have multiple instances. Here we assume there are two nodes (Node A and Node B).
If suppose node A reads data block (abc) from disk and update it as "xyz" but not committed. At the same time node B wants to read the same data block. So here comes cache fusion technology which transfers data "xyz" from node A instance to node B instance through inter connect mechanism.
My doubt is since the changed data is not been committed in node A then how node B can view the data "xyz"? Only the old image should be transferred to node B.
Is my understanding is correct? Any one can explain this.
Thanks
Gokul

hi,
hope this helps
http://advait.wordpress.com/2008/07/21/oracle-rac-10g-cache-fusion/
regards,
Deepak

Oracle RAC on VMWare Fusion

Hi All,
Has anyone installed Oracle RAC on VMWare Fusion ? Can you please, pass links or documentation ?
Thanks

Hi All,
Has anyone installed Oracle RAC on VMWare Fusion ? Can you please, pass links or documentation ?
Thanks

Oracle RAC Course Institues in India (Hyderabad)

Hi
I am interested in doing the Oracle RAC course , can someone suggest me any good institutes in India that are offering the course specially in hyderabad .
thanks
Sri

DB Experts offers Best training for Oracle RAC 11g R2 at Hyderabad
Oracle RAC 11g R2 Contents
Introduction
     Introduction of clusters
     Types of cluster
     Components of a cluster
     Hardware requirements
     Define Oracle RAC
     List the advantages of using RAC
     RAC architecture and components
     SGA structures
     Background processes
     RAC internals and overview of cache fusion
Oracle 11g R2 RAC Installation (RHEL 5)
     Operating system preparation for RAC
     Introduction to network hardware
     Private and public network configuration
     Pre installation configurations
     Configuring DNS for SCAN
     Attaching shared storage
     Preparing attached shared storage for ASM Diskgroups
     Installing Grid Infrastructure
     Configuring ASM
Parameters required to configure ASM
Understanding SCAN
     Installation of 11g R2 RDBMS software
RAC Database Creation
     Storage options for RAC Database
     Understanding Parameters
     Creating Disk groups for Database, FRA and archive log
     Creating RAC Database using DBCA
     Creating RAC Database manually
     Policy managed and admin managed databases
Administering RAC and DB
     Using ASMCA and ASMCMD
     ASM administration
     ASM Disk groups metadata backups
     Configuring ACFS
Enabling and Disabling CRS
     Startup and Shutdown of CRS
     Enabling and Disabling Databases
     Startup and Shutdown of Database
     Startup and shutdown of Instances
     Important log files and monitoring
     checking various events
     Management of OCR and Voting disks
     Adding more OCR and Votingdisks
     Backup of OCR and voting disks
     Restoring OCR and voting disks
     Changing locations of OCR and voting Disks
     missing ocr scenario
     changing SCAN IP and port numbers
     Creating and droping redo log threads
     modifying parameters
Managing Services and TAF, Load Balancing
     Introductions to services
     Creating different services
     Startup and shutdown of services
     Understanding and Configuring TAF
     Configuring Oracle Net services
     Configuring Load Balancing
     relocating services
Managing Backup and Recovery
     Configuring Archivelog
     Configuring FRA
     Configuring RMAN
     Configuring Recovery Catalog
     Performing Hot Backups uing RMAN
     Recovery Using RMAN
     Some RMAN recovery scenarios
Database Cloning
     Cloning RAC DB to Single Instance DB using RMAN
     Cloning Single Instance DB to RAC DB using DBCA
Adding and Deleting Nodes
     Adding a new node to RAC
     preparing a node
     preinstallation configuration
     extending clusterware
     extending oracle home
     adding asm and DB instances
     Deleting a node from RAC
     Removing instance
     deinstallation of oracle
     removing crs services
     deinstalling clusterware
Patching and Upgradation
     Applying patches
     Upgrading CRS from 10g to 11g
     Upgrading DB from 10g to 11g
Installing and configuring CHM
Migrating Non RAC DB to RAC DB
     Assuming non RAC DB using File System and
     RAC DB using ASM storage
Some important cache fusion related views
Faculty : Mr Basavraju
Contact : 9866465379
address : F 204, KVR Enclave, Ameerpet
Hyderabad

Solaris 10 Installation for Oracle RAC 10g Rel2

Hello,
I have bought Enterprise 4000 and 250 in order to install and explore Oracle 10g Rel2 RAC.
In this regard I need some guide lines for
1. Installinig Solaris 10
2. Should I use ASM or other other volume manager?
If anybody can give me a starting document, it will be great.
Thanks,
Javed

I'm assuming you mean running RAC with or without Sun Cluster? If so, the answer is that there would almost no difference in most cases. The only case I can think of would be if you had a sub-optimal interconnect without Sun Cluster where Sun Cluster's clprivnet would have given you striping and availability for free.
I am currently working on a Blueprint that describes the important differences between the two configurations (with and without SC). They can broadly be summarised as:
Sun Cluster gives you:
* Better data integrity protection
* Faster, more reliable node failure detection
* Makes your name space homogeneous, simplifying installation and device management (DID structure), so no need for messy symbolic links
* Gives you a highly available, striped cluster interconnect for the Cache Fusion traffic. (No need for tricky IPMP or link aggregation configurations.)
* Allows you to use volume managers link Solaris Volume Manager or VxVM
* Provides support for shared QFS as a file system for all Oracle objects, data includes (while still allowing ASM)
* A substantial collection of Sun written and supported agents to manage other applications you might also have on your cluster, e.g. NFS, SAP, Apache, etc, etc.
Hope that helps,
Tim
---

Oracle RAC environment

Hi,
I have 4 virtual machines based on XEN. Each of them have Centos 5 OS, 1GB RAM, 10GB disk space. I download all necessary staff for install Oracle RAC, but I wondering what infrastructure I should implement? I want install RAC for test functionality like Cache Fusion, TAF and do some AWR statistic.
What do you think, which choice will be the best:
* raw devices or OCFS2?
* use one xen machine for storage data or partition disk on each machine and use ASM?
* other..
If you have own experience please help me choose best configuration.
Best,
tutus

Hi,
i don't have experience with Xen in oracle RAC but i do use Vmware Fusion on my MAC machine.
** raw devices or OCFS2?*
-- for manageability use OCFS2. raw device is difficult to manage unless you want to try it.
** use one xen machine for storage data or partition disk on each machine and use ASM?*
-- 1. yes you can create a new Xen machine for your storage try considering using Openfiler
-- 2. remember that the disk should be shared on both Linux machine and you cannot just create disk partition on each box.
if you just want to test RAC, why not use vmware its very easy to configure even sharing the disk between nodes.
regards,
baidba

Oracle RAC and SGA size

Hello,
I read somewhere that for Oracle RAC environments, it is necessary to size the SGA 10% - 15% bigger than it would be for a normal, single-instance environment. This would be necessary because of the Cache Fusion feature, which creates a global virtual cache between all nodes of the RAC.
I ask you: Does this make any sense, or it is not correct to add 10%-15% to the SGA, compared to a single-instance environment?
Thanks for your attention.

I read it on metalink... on the foruns. But i didin't save the link... still looking for it
I'm doing a study about a new instance that will be created on an environment. I'm just analyzing the memory that will be necessary for the new instance + application, acording to the minimum requirements of the application that will be installed in the environment (this application will use this new instance).

Admin and dump directories for oracle RAC

I have conflicting environments running RAC on Solaris Sparc. We have a shared file system that contains the dump directories for both instance in the cluster.
both instance are writing to the same directorie for the alert log and other trace files. alert_<sid1>.log and alert_<sid2>.log are in the same directory. Should these be seperated on put these directories on local file system on the nodes? My instance is crashing on a regular basis, just wondering if this could cause a problem.
thanks in advance.

afaik , the ora-00600[12333] could'nt be crashing the instance but rather indicates a communication mismatch between the
client and the server or a networking issue .. Usually those are ignorable ..
however the other error you indicated ora-00600 [kclfadd_1] could be responsible for Instance crashes ..
Looks like this issue only affects 64 bit versions of Oracle and this affects Oracle 10.2.0.2 ..
one-off patches are available for this issue on Linux , HP Tru64 , AIX , Solaris SPARC and Solaris x86 { all 64 bit platforms only }
This issue is an Oracle bug and caused due to corruptions in the cache element structures used by LMS's for cache fusion ..
These crashes might have nothing to do with the Shared bdump / udump locations you have ..
My advice would be to apply the one-off patch if it's available for your platform on 10.2.0.2 or upgrade to 10.2.0.[3/4]
where this bug has been fixed .
Vishwa

Oracle RAC Wait events

Sun OS 10
Oracle 10.2.0.5
We we are running 2 node RAC and we frequently seeing the following waits in the top 5 wait event
cr request retry
gcs log flush sync
Couldn't locate these events in the database reference
http://download.oracle.com/docs/cd/B19306_01/server.102/b14237/waitevents.htm
Thanks
Saravanan

gcs log flush sync is similar to log file sync in standalone:
from - http://orainternals.files.wordpress.com/2010/02/riyaj_advanced_rac_troubleshooting_rmoug_2010_ppt.pdf (you might have more luck opening this one)
Gcs log flush sync
- But, if the instances crash right after the block is transferred to
other node, how does RAC maintain consistency?
-Actually, before sending a current mode block LMS process will
request LGWR for a log flush.
- Until LGWR sends a signal back to LMS process, LMS process
will wait on ‘gcs log flush’ event.
- CR block transfer might need log flush if the block was
considered “busy”.
- One of the busy condition is that if the block was constructed by
applying undo records.
cr request retry in some cases means that the message was lost and re-requested... this is tied to interconnect - either udp issues (like truncated udp packets or packets sent out of order), the session was lost on the other node, or the node restarted quickly... could also mean your nic might be flaky or something happening on the switch. If this is a big concern then you'll need to have someone look at the flow on in the interconnects as this is specific to cache fusion.

Oracle RAC interfaces

HI,
While ORACLE 10g R2 CRS Installation , why should we give public interface information.
I understand we give private interface information because it should be used for Inter instance communication.
But why do we need to specify public interface.Is this poublic interface gonna be used for Cache fusion at all?
Thanks
Pramod

When one server goes down, the other server will take up both virtual IPs on the public interface, ensuring there is no delay in failover.
The interface type indicates the purpose for which the network is configured. The supported interface types are:
Public—An interface that can be used for communication with components external to Oracle RAC instances, such as Oracle Net and Virtual Internet Protocol (VIP) addresses.
Cluster_interconnect—A private interface used for the cluster interconnect to provide interinstance or Cache Fusion communication.
If you set the interface type to cluster_interconnect, it affects instances as they start up and changes do not take effect until you restart the instances.

Enable Cache Fusion in Oracle RAC

Similar Messages

Maybe you are looking for