|
|
|
Oracle Hardware Upgrade Tips
Oracle Database Tips by Donald BurlesonNovember 5, 2015
|
Update:
Please see these updates to Oracle hardware
architectures and
the
costs and benefits of server deconsolidation.
The rapid pace of improving hardware technology has an
important influence on the landscape of the IT manager. Hardware depreciates at
an astonishing rate, and the CIO must understand how to maximize their hardware
investment in the volatile world of Oracle database technology.
This article will examine the realities of hardware
advances and show how the Oracle systems manager can make an intelligent
decision about when to move to faster hardware platforms.
Your worthless machine room
Oracle system managers are often surprised to learn that
their multi-million dollar investment five years ago is almost worthless. While
the IRS tax rules allow for rapid depreciation of computer hardware, the real
depreciation is often greater than the depreciable rate, and the manager must
face the reality that their 2 acre machine room is practically worthless. Just
look at the radical changes in the past few decades:
- 1960s
- Only the largest of corporations could afford their own data processing
center, and all small- to mid-sized companies had to rent CPU cycles from a
data center in order to automate their business processes.
- 1970s
- Small UNIX-based servers existed, such as the PDP-11. However, they were
considered far too unreliable to be used for a commercial application.
- 1980s
- In 1981, the first commercial personal computer (PC) was unveiled, and
practically overnight, computing power was in the hands of the masses.
Software vendors rushed to develop useful products that would run on a PC,
and the introduction of VisiCalc heralded the first business application
outside the mainframe domain.
- 1990s
- Oracle appears and relational databases dominate the IT market. Large
shops have hundreds of small UNIX-based computers for their Oracle
databases.
- 2000s
- Monolithic servers reappear, and Oracle shops undertake a massive server
consolidation. By 2015, servers with 256 processors run hundreds of Oracle
instances.
- 2010s
- Disk becomes obsolete, and all Oracle database are solid-state. Hardware
costs fall so much that 70 percent of the IT budget is spent on programmers
and DBAs.
Many IT managers find that the cost of removing ancient 5
year old hardware from their data center outweighs the residual value. See my
notes on
evaluating an Oracle hardware environment.
The bad news
Hardware depreciates rapidly, regardless of use, and it's
not uncommon to see a 95% depreciation every five years.
In addition to declining costs we also see an exponential
improvement in hardware speed:
-
RAM -
Historically, RAM I/O
bandwidth grows one bit every 18 months, making the first decade of the 21st
Century the era of 64-bit RAM technology
-
CPU - Because CPU speed continues to outpace
memory speed, RAM subsystems must be localized to keep the CPUs running at
full capacity
-
Disk - Disk
storage costs fall
10x every year, and I/O bandwidth capacity doubles every ten years.
Many Oracle shops are lucky if they can find someone to
rip-out the aging servers in return for the hardware. Many servers gain new life
on eBay, where you can peruse a fine
selection of new and obsolete servers.
That old Sun E25k that cost over $200k a few years ago can
now be found on eBay for under $10k!
The good news - better throughput
The good news is that the Oracle system manager will see
several real benefits from an upgrade to brand-new Oracle hardware:
-
Faster throughout - Today's CPU's are far faster
than their 5-year old predecessors, and 64-bit servers can be a Godsend to
shops that suffer from RAM constraints. See my notes here on the benefits
of 32-bit vs. 64-bit processing for Oracle servers. Also, solid-state disk
(SSD) is becoming increasingly affordable, leading some shops to abandon
mechanical platters in favor of solid-state Oracle storage.
-
More real estate - The CIO will gain significant
real estate on their machine room floor when acres of individual
minicomputers are replaced with compact grid server blades or large
monolithic servers. Instead of the old-fashioned 'one database, one
server' approach, we now see servers that that can support dozens of
applications. See my notes here for the benefits from Oracle server
consolidation.
When do I upgrade my Oracle hardware?
By quantifying the intangible costs of managing obsolete
hardware, the Oracle systems manager can predict the increasing costs. These
intangibles include the lost work time from slow response time, the increased IT
costs from manual intervention, and the lost goodwill from customers who
experience delays.
Oracle systems managers are reluctant to spend on human
intervention
Once quantified, finding the date to upgrade the hardware
becomes a classic cost-benefit analysis, with the CIO weighing the relative
costs in the 'maintain vs. upgrade' decision.
In sum, the costs of upgrading to new hardware are the same
as an initial acquisition, giving management a huge incentive to squeeze every
possible hardware resource. See my notes on
Oracle performance, hardware & RAM tuning optimization
Postponing a Oracle hardware upgrade
When faced with a one-time expenditure of millions of
dollars, many Oracle systems managers seek short-term strategies to defer this
cost. Considering that the opportunity cost of capital (at 10%), the manager
can safely spend $100k per year in short-term 'band aids' to reduce the stress
on their aging server. For full details, see my
book "Oracle
Tuning: The Definitive Reference",
As we have seen, hardware costs fall rapidly while human
costs remain relatively static. This leads to a condition there the human costs
of manual tuning activities outpaces the costs of faster hardware. For example,
shops which are I/O-bound can defer a server upgrade by buying super-fast
solid-state disk arrays. See my additional notes here on hardware-tuning for
Oracle databases.
During initial system acquisition, the Oracle DBA performs
the initial database tuning, using powerful global adjustments to optimize the
database workload. Holistic Oracle tuning grabs the 'low hanging fruit', and
the Oracle manager makes a conscious decision to defer the labor-intensive
Oracle tuning activities (e.g. tuning hundreds of individual SQL statements).
However, when face with a huge expenditure for a hardware
upgrade, it can be economically feasible to have the DBA grab the 'high-hanging
fruit', going after tuning activities with a smaller incremental benefit. These
might include:
- Multiple blocksizes
- Incurring the additional management overhead of employing multiple
blocksizes can improve utilization for scarce RAM resources, providing
between 5%-10%. Oracle guru David
Aldridge
notes a test where is noted a 6% reduction with larger index blocksizes,
a significant difference, especially to hardware stressed shops.
- Marking a tablespace
as read-only - You can mark Oracle
tablespaces as
read-only. Real-world benchmark tests show that making a
tablespace read-only produces a small but measurable benefit in response
time. While this benefit is not of interest to shops with an excess of
computing power, it can postpone a hardware upgrade.
- Task-level tuning
- When obsolete hardware and increasing volume causes hardware stress,
emergency DBA tuning techniques can be worthwhile. Oracle author
Chris Lawson notes that nothing is off-limits when trying to squeeze
more cycles out of an aging server environment, an approach he calls 'tuning
as a life-or-death endeavor:
"Given my new
motivation, all limits were off. I considered all kinds of crazy
ideas" server changes, init.ora changes, disk changes, etc."
Lawson goes on to describe a creative solution to solve
an acute RAM-related hardware problem:
"Since the bottleneck was
mostly due to disk i/o, any improvement in disk access would directly affect
the problem job. Since we knew the exact SQL, we could pre-run SQL at
duplicated the "real" Sql that would shortly follow. That is, we would
pre-cache many of the blocks that would shortly be needed.
- Detailed SQL
optimization - Provided that the
global settings are optimized for your workload, the DBA can use techniques
to identify
'exceptional' SQL for tuning. For example, if the server is stressed
with disk enqueues, the DBA can extract those SQL statements that incur the
most disk activity. The DBA can then undertake tuning to fetch the rows
with the minimum amount of data buffer touches, using tools such as
materialized views and specialized indexes.
To guarantee a smooth transition to new Oracle hardware,
it's always advisable to consult Oracle experts who can demonstrate a proven
track record of success. There are many perils and pitfalls when we upgrade to
new hardware, and I recommend avoiding any consultancy that does not publish the
names, credential and experience of their staff.
Choosing your Oracle hardware upgrade support
The web is full of puffery and self-appointed Oracle
experts, and the IT manager must be very careful. Don't take any claims of
"extensive experience" at face value, and always ask for real-world references
when choosing expert Oracle support. Many Oracle shops seek these credentials
for an Oracle hardware upgrade consultant:
-
Trustworthiness - Make sure that your consultant
has no fiduciary interests with hardware vendors and seek consultants who
have been vetted, such as a Oracle expert with an active SECRET clearance.
-
Training - See "real" software engineers with
degrees in electrical engineering and advanced study (masters degree) in
software engineering.
-
Experience - Look for consultants who publish
their resume's and offer real-world references.
Remember, not all "experience" is the same, and some Oracle
professionals count at-home learning as work experience. See my notes here
on
evaluating the credibility of Oracle experts on the web.
|
For me, I carefully vet all of my
Oracle upgrade experts, and my
clients like to use Oracle certified
professional experts who are also software engineers. |
|