Configuring HP SIM for Optimal Performance in Large Environments
Tuning disk input and output
Disk tuning – RAID 5 performance
Ensure
that
there
are
no
other
bottlenecks
(CPU,
NIC,
or
memory)
before
adding
more
disks. Typically,
the
SCSI
or
Fibre
Channel
is
not
saturated,
but
rather
there
are
not
enough
disks configured.
Ensure
that
you
place
the
log
files,
indexes,
and
database
files
on
different
array
sets.
The
RAID
stripe
size
is
written
to
each
system
in
the
array
or
to
the
group
of
systems
in
the
RAID
set.
In
other
words,
if
you
have
a
stripe
size
of
128
KB
and
four
disks
in
a
RAID
5
array,
the
stripe
size
is
divided
across
all
the
disks
in
the
array.
Each
chunk
is
written
to
each
drive
in
the
array.
So
if
you
have
three
systems
in
a
RAID
5
array,
three
chunks
are
written.
Logically,
only
two
of
the
three
drives
are
receiving
your
data.
The
other
drive
is
receiving
the
XOR
parity
data,
which
is
rotated
among
the
disks.
For
RAID
5,
set
the
low-level
raid
format
chunk
size
so
that,
when
multiplied
by
the
stripe
width
(number
of
drives
in
the
array
minus
one),
the
product
is
equal
to
the
typical
input/output
size
you
use
from
the
server
operating
system’s
perspective.
For
example,
if
you
determine
your
input/output
writes
are
occurring
8
KB
at
a
time
using Performance
Monitor
and
you
have
a
three-disk
RAID
5
set,
chunk
size
must
equal
to
8
KB
for
two drives,
or
a
4
KB
chunk.
If
this
is
not
possible,
set
the
chunk
size
to
the
next
larger
size
available.
Now,
each
8
KB
write
fills
an
entire
stripe
set
(user
data)
and
the
parity
data
is
written
on
the
same
stripe
set
on
the
last
disk.
Keep
in
mind
that
this
parity
info
rmation
rotates
around
the
disks
in
the array.
Set
the
relationship
between
the
RAID
stripe
size
and
the
formatted
sector/cluster
size
so
that
the
operating
system
input/output
writes
to
the
cluster
size
with
the
proper
chunk
size.
Set
the
allocation
unit
size
of
the
NTFS
partition
to
8192.
Tuning
a
single
disk
To
optimize
the
performance
of
single,
large
hard
disks
in
heavy
throughput
server
environments, RAID
is
the
solution
of
choice,
with
throughput
distributed
over
multiple
drives
and,
more
particularly, by
the
application
of
large
amounts
(32
MB
or
greater)
of
write
back
cache
on
the
hardware
RAID controller.
However,
many
people
are
moving
to
more
fault-tolerant
configurations
with
full management
server
mirroring
and
full
fail-over
for
a
more
dynamic,
high-availability
solution.
This solution
deals
with
any
type
of
management
server
failure,
not
merely
a
hard
disk
failure.
In
this environment
it
is
unnecessary
and
costly
to
mirror
complete
RAID
systems.
Additional tips:
Use only one disk partition per physical disk (lowers head movement)
Defragment the disk on a regular basis
Use the proper allocation size (match it to your workload)
Configuring the network
Generally, no specific tuning is required to get a NIC in Microsoft Windows Server 2008 to run at
full speed, but consider the following:
Do
not
use
autodetect.
Set
your
network
system,
clients,
and
management
server
NICs
to
the
exact
settings
you
want.
In
many
cases,
autodetect
provides
less
than
optimal
performance.
Foregoing
autodetect
also
ensures
that
your
network
systems
are
working
properly
at
the
speed
you
desire.
If
they
are
not,
connectivity
problems
are
likely
to
occur.