RE: 64 node Oracle RAC Cluster (The reality of...)

  • From: "Kevin Closson" <kevinc@xxxxxxxxxxxxx>
  • To: <oracle-l@xxxxxxxxxxxxx>
  • Date: Tue, 21 Jun 2005 21:03:33 -0700

 >specifically aimed at OCFS (which is another thing I don't 
>know much about!)

I do. OCFS is a "datafile container" 100% geared to 
direct IO. Not sufficient for executables, and
nobody has ever argued that fact.

Peter is aware of my opinion due to my postings
on the OakTable :-)  I think it is wrong to call
anything a filesystem unless it supports the
filesystem API (e.g., glibc on linux, Win32 on Windows). 
If there are barbs and bombs to be encountered simply
accessing files via the filesystem API (glibc) then 
let's just not call it a filesystem at all, and 
therefore not a cluster filesystem.

Summary? If the OS can't demand page text from 
binaries efficiently, it isn't a filesystem ...
'cause at that point it certainly wouldn't 
support mmap nor positional locks and would likely
be troubled by other bread-and-butter FS

Kevin Closson
Chief Architect, Database Solutions

>"Controlling developers is like herding cats."
>Kevin Loney, Oracle DBA Handbook
>"Oh no, it's not.  It's much harder than that!"
>Bruce Pihlamae, long-term Oracle DBA
>-----Original Message-----
>From: oracle-l-bounce@xxxxxxxxxxxxx 
>[mailto:oracle-l-bounce@xxxxxxxxxxxxx] On Behalf Of Mladen Gogala
>Sent: Wednesday, 22 June 2005 11:05 AM
>To: peter.sharman@xxxxxxxxxx
>Cc: kevinc@xxxxxxxxxxxxx; mwf@xxxxxxxx; 
>Rich.Jesse@xxxxxxxxxxxxxxxxx; oracle-l@xxxxxxxxxxxxx; Peter 
>Ross Sharman
>Subject: Re: 64 node Oracle RAC Cluster (The reality of...)
>On 06/21/2005 08:42:50 PM, Pete Sharman wrote:
>> Seriously, where a CFS is supported by the OS, why would you 
>do anything else for the ORACLE_HOME?
>> Pete
>Because CFS might not be the best fit for a myriad of small 
>files that need to be paged into the memory quickly. CFS may 
>not support anything but direct I/O, therefore not caching 
>$ORACLE_HOME/bin/oracle and shared libraries on 
>$ORACLE_HOME/lib, which means that almost all page faults will 
>be hard faults. On the other hand, if CFS does cache files as 
>is the case with UFS by SUN Microsystems, it needs the same 
>type of mechanism to synchronize the caches across the nodes 
>as are the ones used by oracle. That might perform well only 
>if background_dump_dest, user_dump_dest and core_dump_dest are 
>not on the same global file system. Also, resist temptation to 
>put archive destinations on CFS. Putting it on normal FS and 
>then sharing it over public connection by NFS is much faster 
>then by putting it on CFS.
>Mladen Gogala
>Oracle DBA

Other related posts: