A long answer…. Sorry …
I started working for Oracle Public Sector back when the X2’s were first coming
off the line. All I did during my five years at Oracle was work with customers
installing and trying to maintain Exadata. I did site surveys, helped them
figure out how to patch their systems (many of them were government entities
that could not have outside internet attached to their Exadatas so they could
not use Platinum support).
I left Oracle about 7 months ago (as have, I’m afraid to say, many talented
people) after working there for over 5 years. All I can say is that within my
group, there was a talent disappearance the likes I’d not seen in a long time.
These were all voluntary departures. I’ll leave the why’s up to your
imagination.
Now, I bleed Oracle red, but I also understand that part of Oracle is this
sales and marketing machine that one has to take with a grain of salt at one
time. So, as I read the specs and benchmarks, I expected that a certain amount
of tweaking here, there and other voodoo was contributing to the numbers I saw.
All that being said, let me share a couple of experience with Exadata – and
then I’ll share my list of reasons.
Then, I participated in my first POC with a very large customer. In this case,
they had a *daily* report that they were legally required to submit to certain
government offices. The problem is that this *daily* report would take *three*
days to run. We decided as part of our POC that we would take on this report.
In quick order, we sucked up the customers data set, and imported it into a
half rack X2. The database was created with the default Exadata template and no
parameter changes were made save some memory optimizations and setting some
system statistics to the Exadata standardized settings. Back in those days,
Exadata nodes were limited to some 96GB of memory each as I recall – and this
was the one real concern we had – we would have liked more memory to allocate
to the SGA and the PGA’s of each node.
We then proceeded to run the report. Which finished in something under 2
minutes. Frankly, all of us were sick to our stomachs at first, since our first
question was, what went wrong… we expected serious performance gains, but
really – three days to 2 minutes. The client and our team poured over the
numbers in the report and the end result – the report was exactly correct.
So – to answer the question asked – my top selling points (more than 3 I’m
afraid)
* Disclaimer – Any machine can be configured to perform like crap. I’ve seen
cases where legacy customer code caused major problems for Exadata simply
because their code consumed all the CPU’s on the compute nodes due to
incredible bad coding practices. Bad coding practices are just that – wherever
you are.
1. Performance beyond the extreme. The X2’s had some Achilles heels with
respect to performance. For example, bulk data loads could be slower than one
would really like in the older generations. Future X-generations addressed many
of these issues and performance in both the data warehouse and OLTP arenas, as
well as things like bulk loading of data – is simply phenomenal. More often
than not, you can drop in your code without changes and experience serious
performance gains. Add in certain features such as parallel processing,
compression, etc… and let your eyes water.
Additionally, Exadata adds additional features not available in the regular
database product, such as additional forms of compression that offer
significant savings in terms of disk space storage.
2. Patching – Seriously, look at your data center right now. SANs and
Disks, switches, servers, software as far as the eye can see, etc..etc… How
often is every component of your data center patched by a recent patch? How
complex is the process of patching your entire data center? How many vendors do
you have to deal with, how many conflicts do you need to deal with? Are outages
required? As a result of all of this, how secure are all the components of your
data center?
With Exadata – you have a data center in a box. You have quarterly patch sets
that are released that provide the most current patches that have been tested
and approved by Oracle. Not everything is patched every quarter – the
InfiniBand switch is patched in-frequently, for example – but when the patches
do come they are regression tested, unit tested and system tested.
While there have been issues, bugs and occasional problems with patches (such
as when patching the InfiniBand switches would take the whole cluster down
(oops) – in general my experience (well over 75+ Exadata patches under my belt
over the last six years) is MUCH smoother than the alternative.
3. Community – You might have noticed that Oracle is a bit picky about
what you can, and cannot, add to an Exadata system. The objective is some
uniformity in the Exadata community. Thus, when one customer experiences a
problem the root cause of that problem, and it’s solution, is easily
disseminated to the entire community in either a fix or some note that
documents the problem along with a solution. It is this community that I think
gives a better sense of security when applying the quarterly PSU’s. My approach
was if there was a development system – I’d apply the quarterly patch pretty
quickly to let it bake. If I didn’t have a test system, I’d watch what was
happening in the community and wait for the patch to bake a month and a half or
so before I’d start looking to apply it.
4. High availability to the extreme – We did a demo for a government
entity once with Exalogic and an Exadata full rack. While the thing was up and
running, we had the senior leader present at the meeting pull just about any
cable or disk he wanted. We opened the front and back of the box, generally
pointed out what was in there and asked him to point to cables he would like us
to remove. We pulled two disks he pointed too, one InfiniBand cable and as I
recall one power supply cable. Exadata chugged away without skipping a beat.
Also, the failure notifications showed up, the automatic dispatch for the disks
happened, etc.
There is a down side to this though… I was at another POC for a customer with a
loaner Exadata which had been moved around between a few customers already. We
had had some issues with the disks during setup, but they had settled down and
started working. We had been working for 4-weeks on their POC. A day or two
before I was to leave, I went to the computer center to get my coat (I’d had to
wear a coat in there it was so cold!) As I walked by the full Exadata Rack, I
noticed two drives – and the lights on the drive told the tale – they had
failed.
Because of the rush to setup the machine, we had not setup the hardware
monitoring – so we didn’t know the drives had failed. Yet, the POC went off
without a hitch and performance was, again, awesome. The point is that HA can
lead you down this path of over confidence…. If another disk or two had failed
in the right place, we might have had problems. So, monitoring is critical –
even in HA environments.
5. Platinum support – FREE – don’t have time to do the quarterly upgrades?
Platinum support – a free option you can sign up for – will do them for you
four times a year. Additionally, they will do a database migration for you for
no cost (except that it represents 2 of your quarterly upgrades last time I
checked). They do this all for free, coordinating times with you. They also do
it online. Further, Platinum provides some monitoring services for your Exadata
box. Did I mention it was free?
6. Red Stack enables performance customization – By virtue of the fact
that Oracle owns the stack, they can include performance customizations that
other vendors simply can’t do. The way the cells process predicates (WHERE) at
the storage layer, removing a significant number of rows that have to travel
over the wire and be processed in the SGA is simply something other vendors
cannot do.
7. The ability to run OVM on Exadata was just coming into being when I
left Oracle. I’ll confess that for a long time I was not a fan of the OVM. My
position has now changed and I see the incredible benefits of the Virtual
Machine – even in production.
8. Support Priority – As an Exadata customer – you get support priority.
There are some who have complained that they have not seen that benefit
materialize in any significant way, but all my interactions with support – when
using an Exadata CSI – have resulted in excellent support. I think that it’s as
much as knowing HOW to work support as it is anything else.
9. Easier administration – that’s my opinion of course – it’s a subjective
topic. I think Grid Control is a requirement here.
Ok… I’ll stop there….:)
Now my director is asking for the three top bullet points to take to his
boss. On why we should go to Exadata.