 | | | who 's got the biggest and the fastest? | who 's got the biggest and the fastest? 2007-11-05 - By Robyn
Back
Thanks Greg ...
Yes, I've gotten the marketing pitch from a few different vendors and I can
see through some of it, but I can also see the potential for the right kind
of data and a specific kind of business need. And according the marketeers,
some of the big players in data warehousing are making the switch (big names
were dropped) and Larry Ellison has even acknowledged that Oracle needs to
come up with something to beat the newest upstarts. Of course, some of the
information being presented is clearly misleading; the challengers imply
that Oracle maintenance is much more labor intensive than it really is these
days and they neglect to mention that Oracle has many of the same features
that they are bragging about.
So now I'm reading through Winter Corps white papers and looking over their
top 10, but 2005 was a while ago and I'm curious .... who's biggest and
fastest now and what do they run? Have any of them really made the switch to
shared nothing and mpp? If they're using Oracle, how do they do it? (for
once, I probably do need RAC.)
And the crux of my question - Is there a point where Oracle becomes the less
optimal solution?
I'm in a very preliminary evaluation stage at this point, but estimates for
data loads are in the neighborhood of over a TB of raw data per day and it
will need to be accessed fast and adhoc to some extent. What are the
options to consider?
Robyn
On Nov 5, 2007 11:37 AM, Greg Rahn <greg@(protected)> wrote:
> Sounds like you've been getting the marketing pitch...be careful and
> don't get caught up in it. Stick to the technicals.
>
> Just to clarify, Oracle isn't OLAP based, unless you are using the
> OLAP feature and functionality.
>
> Just curious, what "very big" and "very fast" to you? These words are
> a bit subjective and mean different things to different people.
>
> Oracle is a significant player in the DW/DSS space and has been for a
> number of years. Check out the Winter Corp VLDB reports. In 2005,
> Yahoo had a 100 TB db on Oracle. I'm sure its grown since then.
>
> An Oracle system can be architected to do whatever you want. You just
> have to define what *it* is. Start with good db design and smartly
> apply features where it can help. Probably the most important thing
> in choosing hardware for an Oacle DW/DSS system is to have a balanced
> system. This means that the CPU and I/O bandwidth are in a optimal
> ratio to each other. Optimal meaning that for your workload, both
> resources are exhausted at approximately the same time.
>
> Don't be fooled that brute force table scans are the way either.
> There are certainly times when they are required, but remember the
> phrase "work smarter, not harder"? It applies here too. The fastest
> I/O is the one that never happens. Partitioning is your friend.
>
> Hope that helps...
>
> On 11/5/07, Robyn <robyn.sands@(protected)> wrote:
> > morning everyone,
> >
> > Anyone out there familiar with setting up very big and very fast Oracle?
> If
> > you could start from scratch, how would you architect your system? What
> are
> > some of the options that should be considered?
> >
> > Can a well designed Oracle system beat the proprietary data warehouse
> type
> > specialists who claim that Oracle is OLAP based and therefore unable to
> > match their speed with big data?
> >
> > Thoughts, comments and war stories appreciated ...
> >
> > Robyn
>
>
> --
> Regards,
>
> Greg Rahn
> http://structureddata.org
>
Thanks Greg ...<br><br>Yes, I've gotten the marketing pitch from a few
different vendors and I can see through some of it, but I can also see the
potential for the right kind of data and a specific kind of business need. 
; And according the marketeers, some of the big players in data warehousing are
making the switch (big names were dropped) and Larry Ellison has even
acknowledged that Oracle needs to come up with something to beat the newest
upstarts. Of course, some of the information being presented is clearly
misleading; the challengers imply that Oracle maintenance is much more labor
intensive than it really is these days and they neglect to mention that Oracle
has many of the same features that they are bragging about.
<br><br>So now I'm reading through Winter Corps white papers and looking
over their top 10, but 2005 was a while ago and I'm curious .... who's
biggest and fastest now and what do they run? Have any of them really made the
switch to shared nothing and mpp? If they're using Oracle, how do
they do it? (for once, I probably do need RAC.)
<br><br>And the crux of my question - Is there a point where Oracle becomes the
less optimal solution? <br><br>I'm in a very preliminary evaluation
stage at this point, but estimates for data loads are in the neighborhood of
over a TB of raw data per day and it will need to be accessed fast and adhoc to
some extent. What are the options to consider?
<br><br>Robyn<br><br><br><div class="gmail_quote">On Nov 5, 2007 11:37 AM, Greg
Rahn <<a href="mailto:greg@(protected)">greg@(protected)</a>
> wrote:<br><blockquote class="gmail_quote" style="border-left: 1px solid rgb
(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
Sounds like you've been getting the marketing pitch...be careful and<br>don
't get caught up in it. Stick to the technicals.<br><br>Just to
clarify, Oracle isn't OLAP based, unless you are using the<br>OLAP feature
and functionality.
<br><br>Just curious, what "very big" and "very fast" to
you? These words are<br>a bit subjective and mean different things to
different people.<br><br>Oracle is a significant player in the DW/DSS space and
has been for a
<br>number of years. Check out the Winter Corp VLDB reports. In
2005,<br>Yahoo had a 100 TB db on Oracle. I'm sure its grown since
then.<br><br>An Oracle system can be architected to do whatever you want.  
;You just
<br>have to define what *it* is. Start with good db design and smartly<br
>apply features where it can help. Probably the most important thing<br>in
choosing hardware for an Oacle DW/DSS system is to have a balanced<br>system.
This means that the CPU and I/O bandwidth are in a optimal
<br>ratio to each other. Optimal meaning that for your workload, both<br
>resources are exhausted at approximately the same time.<br><br>Don't be
fooled that brute force table scans are the way either.<br>There are certainly
times when they are required, but remember the
<br>phrase "work smarter, not harder"? It applies here too.
The fastest<br>I/O is the one that never happens. Partitioning is
your friend.<br><br>Hope that helps...<br><div><div></div><div class="Wj3C7c">
<br>
On 11/5/07, Robyn <<a href="mailto:robyn.sands@(protected)">robyn.sands@(protected)
.com</a>> wrote:<br>> morning everyone,<br>><br>> Anyone out there
familiar with setting up very big and very fast Oracle? If<br>> you
could start from scratch, how would you architect your system? What are
<br>> some of the options that should be considered?<br>><br>> Can a
well designed Oracle system beat the proprietary data warehouse type<br>>
specialists who claim that Oracle is OLAP based and therefore unable to
<br>> match their speed with big data?<br>><br>> Thoughts, comments
and war stories appreciated ...<br>><br>> Robyn<br><br><br></div></div>
<font color="#888888">--<br>Regards,<br><br>Greg Rahn<br><a href="http:/
/structureddata.org" target="_blank">
http://structureddata.org</a><br></font></blockquote></div><br>
|
|
|