RE: 64 node Oracle RAC Cluster (The reality of...)
- From: "Kevin Closson" <kevinc@xxxxxxxxxxxxx>
- To: <oracle-l@xxxxxxxxxxxxx>
- Date: Wed, 22 Jun 2005 15:47:56 -0700
>Bill Bridge is involved in writing it. Whether it will be fast
>is an open question, because even Bill can't change the laws
>of physics.
I've known Bill for many years through my decade long engineering
work in the Sequent port of Oracle. So you must know then that
ASM is a marketing name for OSM. "Oracle Storage Manager",
became "Automatic Storage Management" magically with the flitter
of the marketing pen :-) Bill also spec'ed the ODM
library which interestingly only Veritas and PolyServe have
signed up for, and implemented... Anyway, the important things
about ASM I like. Things like the ability to migrate hot
segments between datafiles. That is, however, a feature
that could have just as easily been implemented in any
datafile type...FS,raw or ASM. The things I don't like
about ASM are the things that no datacenter should care
about. For instance, ASM was hijacked by marketing as
"the Veritas killer". It is no such thing, and no datacenter
should care whether Oracle kills Veritas or not. How does
Oracle "killing" Veritas help a datacenter deployment of Oracle
in any way? ASM is software RAID. And it doesn't do it as
good, nor as general purpose, as 3rd party solutions.
To understand my point, you have to accept the fact that
there is other software in the datacenter that has nothing
to do with Oracle and common-ground platform choices are
important.
>less complex than a full featured CFS. The functional
>requirements are a tiny subset of a CFS, and only precisely
>controlled "clients" will be able to request well defined
>services from ASM.
that is exaclty my point. so what do you do with the
myriads other data that can't get stuffed into ASM?
BTW, the code path for a CFS direct IO measures
in the sub-percentage difference from character special
raw device access. I've done way too many instruction
traces. If you want to account for lost processor cycles,
look first to the code that executes hundreds of thousands
of times per second, like Oracle spinlocks (latches).
I'm not saying that Oracle spinlocks are inefficient,
but slice the processor pie before talking about how
big any one certain piece might be.
--
//www.freelists.org/webpage/oracle-l
Other related posts:
