Although some places are getting better.

When I was in downtown Atlanta over labor Day this year, I was able to get cellular phone and data services. The last 4 times I was there, it was nearly impossible to make a phone call, as for data, forget about it.

No. we were heading west.

@scottalanmiller Has? How'd that work? I can see @psx_defector drinking every drink before it gets to you. And that's when he's 3 feet way...:P

Transcript:

so my name is Scott Alan Miller that's
all I'll tell you about me because
honestly no one actually cares and so if
I talk about myself I'm boring and
narcissistic and one of the rules of
presentations is it could be boring it
could be narcissistic people so this
talk is on open storage I think it's
called open storage landscapes and I
think there's a lot of no one's quite
sure what I'm gonna be talking about
what I want to do here is not look at
specific technology implementations or
some real specifics but my goal today is
to have you come away with a different
vision a different appreciation for
storage and the storage landscape and
and exactly what your options are and
how to approach thinking about storage
in general when it's oranje that is a
major problem for all companies today
whether you're you know a tiny shop or
you're a giant enterprise that sort of
expensive storage is critical in reality
nothing is more important in your
business than your storage if you lose
everything from a processing perspective
and save your storage or your voice then
you're recoverable but if you lose all
your storage and keep your processing
you're screwed
so so storage is very very important but
we have a very we tend to look at it in
very strange ways kind of forgetting how
storage actually works under the hood so
I kind of want to start off by going
through some real basics of storage and
try to present it in a little bit of an
interesting way and then I want to talk
about the concepts of open storage and
how that applies to to your businesses
and some reference implementations of
how that might be something that you
would want to do and when that might
make sense because my format is very
very loose I don't work with slides
because they're distracting and I can
kind of scale the speed very easily so
I'm gonna be a lot more open to people
asking questions as we go try to keep it
kind of related to where I am but if you
want to jump in and it's kind of
ladies will tackle it as we go because I
could very easily you know miss
something that's valuable for everybody
so alright so in the beginning right
when when computers first began the very
first thing we needed was well we have
this computer so we need to store stuff
so the very first thing that we got was
the hard drive I realize this is not a
bit hard drive the hard drive was simple
right has one spindle it's store stuff
everybody understood that right very
very easy
there's nothing complex it's a physical
device and we attach it to a computer
so this is our server
we refer to servers as boxes we can
demonstrate them as
so what we did with it was our very
first server the first server because we
took this hard drive the first hard
drive ever
and we literally took a cable that
actually looked like that babe right I'm
stuck inside we're now directly attached
one drive directly in Tetris get
attached inside your server the
connection technology was very very
simple right we know today is scuzzy yes
there's something before it but
basically scuzzy was the beginning of
real storage it is a generally
simplistic technology but it does have
the ability to dress but to address more
than one drive so we started putting
more drives in service once we had more
than one driving service they said well
this is a problem so we took that drive
when we said this is a point of failure
we lose this drive we have nothing
obviously can have backups hopefully
you're following Daniels advice and have
made good backups and you can get a new
hard drive and keep going but we're also
concerned about not just recovery which
is where backup comes into play but
continuity of business which is where
with redundancy come from the place so
in order to get redundancy we said well
you know what if we had two drives which
we already had we had multiple drives
and we said well what if we built an
abstraction layer and put these together
half the entertainment just watching it
works doctor
so we take these hard drives and we
literally put them together in a way
that for an outside observer it's one
drive hands bigger but it's the same
thing right we call this rate so in this
case it's raid one so these drives in my
example are mirrored to each other but
not important so the key element here is
that when we use raid we are taking
multiple independent devices and we are
attaching them together in such a way
that when we then attach it to the
computer alright and there's a little
bit more than duct tape involved here
imagine this duct tape includes a RAID
controller
right and that RAID controller provides
that abstraction layer so that your
computer when it sees this block of
storage right it's still connected using
the same technology that the single
drive was connected with and is
connected in the same way and when the
server when the operating system looks
at your raid array it is looking at one
drive it has no idea how many physical
devices are in play it believes that
there is a single logical drive so this
was the first step of making enterprise
storage right and this was a pretty good
step once we got to this point we could
really do business with servers so what
happened from here was we started
getting a lot of these right oh that one
fell out
don't bite from that manufacturer and
and we said well we don't have a lot of
spaces in these boxes they get hot they
fill up with other things we want to
have a lot more than two drives I don't
have two drives my example but what if
we had to eat and drive right what if we
wanted to do that well why not put these
same drives outside the box it's not
where the term came from
so we came up with the ability to
literally take skuzzy same technology
nothing's changed we changed the cable
so there's a little bit more resilient
to our interference and we attached
drive outside of the box this is where
the term direct attached storage came
from kind of denote that this is what
we're doing as far as the server is
concerned they're still drives inside
the computer
nothing has changed to the servers
perspective it's just literally they're
outside the case rather than inside the
case you can do this with a running
server pop off the top yank a car drive
and as far as the cable will go you can
just dangle your ear drive but basically
what's happening in a physically managed
way so that you're not catching fire
shorting something out and this gave us
the next step of storage we were able to
do more advanced things but nothing has
changed as far as the server's concerned
it has one drive that's it and it's
everything is operating the same so then
at some point we said well I like how
this looks like scuzzy cable works out
really well so the later we said and
this could be anything right so early on
OA scuzzy but now there's lots of
options for this today we've got sass
we've got SATA we've got USB we have
firewire we have Fiber Channel right
lots of options but they all do almost
exactly the same thing which is very
little
so with the more advanced of these
technologies right we started saying
well we could instead of simply
attaching the drives to the server in a
way that the it's really really
physically directly attached that's not
necessarily useful we might want to have
our drives farther away well if we're
gonna do that we're going to need a
little bit more advanced communication
so scuzzy is traditionally it's very
much like I love this example it's
basically like Ethernet not switched
Ethernet that we use today
it's like Ethernet long cable vampire
taps everything communicates at the same
time interrupting if each other so
scuzzy was kind of inefficient and it
really wasn't going to scale so we moved
to different technologies for large
implementations
and the first thing that really came
into play was fiber champ and the thing
that makes fiber channel different than
scuzzy is that scuzzy is really a non
lair to networking connection right it
doesn't really have what we think of as
Ethernet addressing style machine level
addressing with fiber channel we have
machine level addressing so it's a full
layer 2 networks protocol which also USB
and firewire are just not on the same
scale so when we did that we said oh
well now we can switch these connections
and I don't have all the cabling here
that it would be necessary to
demonstrate this but what's interesting
is suddenly we could take all these
drives just like we had before and sit
them in a box and put a bunch of servers
over here and just have a switch in the
middle and connect the drives and
connect the servers and using the switch
we can tell which ones connected to
which and what we have is rudimentary
Sam right and the most simple devices
you can get a hold of like your neck
your SC 101 is literally two drives in a
little plastic container with the
smallest chip you've ever seen and the
only thing it does is act as a very very
simplistic drive controller and puts zzn
protocol on to the wire and that's it no
logic no nothing it is basically a
network card and you combine Network
written small network cards that attach
directly - I'm sure they exist for sass
but I've seen them for old-school
parallel data plug them in and you
literally have a network connection on
the drive nothing else no raid no
management no anything the computer
still sees the block device that it saw
when it's plugged in internally plugged
in externally plugged in through fibre
channel switch whatever so nothing has
changed as far as the computers
concerned
so what we started doing then right is
we said well now these drives are out
here this was kind of an experimental
stage this wasn't a stage that lasted
very long with all these old drives in
fact it said well if we took all these
drives and we put them into another box
another server and instead of being
rudimentary like the SC 101 we've got a
big box with an eye specifically got a
larger box so that we could hold lots of
drives right this box might end up being
several racks in size we can hold
hundreds of draws thousands of drives in
a single box and now we have all these
abstractions that already exists we have
the r8 abstraction that allows this
machine to remember this as a server to
see those drives as a single drive and
we have networking technologies like
fibre channel that allow this to share
out those drives as block devices to any
device that wants to read a block device
right so these are acting the same as
any other block device you're going to
see right whether it's we think of
storage as its own thing but storage
drives networking they're all blocked
devices to the system so they're all
acting exactly the same it's just the
way that we look at them so we get our I
don't have the switch here so just
imagine they're switching
okay
so we have our fiber channel connection
we're gonna attach it from this server
which has all the drives over to this
server which is where we want the data
so yes
yes
yes correct so this is a server that we
have made purposed to storage and this
is a server that's doing you know who
knows what the normal stuff probably
with no Drive or could have its own
drives plus these struts you can mix and
match so now we have dedicated devices
but as far as this server is concerned
it still sees a scuzzy connection a
fiber channel connection whatever so
we're still dealing with the the same
technology this is what's interesting is
that we start to think oh weird things
are happening but as far as the server
is concerned nothing has happened it's
still just that drive connected directly
so then after this stage right we said
okay so we're going to introduce another
abstraction layer to the network and
we're going to pump fibre channel over
tcp/ip so we introduced ice Guzzi which
is the fiber channels goes here
basically the same protocols very very
close I scuzzy is simply taking the
original scuffie protocol and
encapsulating it in tcp/ip so that I
scuzzy leaves this box
I'm sorry scuzzy leaves this box goes
into an encapsulation layer that allows
scuzzy be transported over tcp/ip and
then it gets taken out on this box and
its back to scuzzy so it's a device
layer it is still seeing a single scuzzy
drive connected nothing has changed so
that really brings us to modern storage
as far as a block level storage is
concerned I'm trying to avoid the
obvious words that every knows I'm
talking about so this is a server normal
server which is very very important it's
a normal storage server this is a normal
computing device and everything is
viewed as block devices if we take this
server and we then apply an interface to
it that makes it very very easy to
manage it makes it no longer look like a
normal server but look like a dedicated
storage server take away the options of
processing on it add in some
perhaps ease of management type things
we then call it appliance sized and we
refer to it as a sand but there's no
difference to a sand from a storage
server except that it has an interface
that makes it work not like the regular
server that's so so this is the really
interesting point we have all this
processing power here and so if we then
let this server not just share out the
lock level devices without touching them
with you know when we're dealing with
this at this level this machine is
effectively dumb it doesn't know what's
on the disks it doesn't know how to use
them it doesn't know how to read them it
can have if this could be a window
server and those drives that are in it
could be loaded with Linux file system
of Solaris file systems it can't read
them they can't do anything with them
but it can share them out because it's
purely a block level device if we then
add more power to this box more power
from a software perspective and said
okay we're gonna make it in this box
able to read these discs and understand
them then we can start getting even more
power because we can start doing things
here without sending it over there first
so we can add new protocols onto this
box that give us not block level sharing
we can still have that this is all still
going on but this box now can read these
drives right there this is new layer so
we add in a protocol that allows this to
share a file system layer instead of the
block layer and to do this obviously
this machine has to understand it too we
need to be able to put it onto the
network so specific filesystem
abstractions are made for this and we
know them today is NFS sips AFS these
are popular protocols for this what
makes them different than blooded block
level is that at the file system layer
this device can determine if there's
been changes to a file and only Sandover
changes it can make changes
you can do all kinds of things including
that's very very important
it can lock single file from you which
means that this server is contacting
this box and wants to write to a file
this server can lock that file and say
no one else can write to it which means
it for the first time we have a means of
having this connection go to more than
one server you can do this with block
level devices back in the early days
when we had literally hard drives inside
servers people would actually take this
scuzzy cable hook one end into one
server one hand into another server and
dangle a hard drive off the middle anak
would cause obviously disaster you have
to drive controllers it's like having
two steering wheels in the car and two
people driving without being able to see
each other or talk right one person
wants go this way one person's go that
way one first the fitment gas the
person's hitting the brake
you know deer runs out in the road each
one thinks a different directions way to
go you're gonna have a disaster and
that's what servers do if there's two
servers talking to a single hard drive
not aware of each other so there are
specific file systems that were designed
to be able to handle that where but each
server had to play nice there's no
gatekeeper so any server that decided to
mess with the data was going to and it
could make changes that the other one
didn't know about it could delete files
that the other one tried to protect they
could read files the other one said it
shouldn't be allowed to read there's no
way to control those things so there was
a gatekeeper when we're dealing with
file system level sharing we have that
gatekeeper we have the security at this
level where we control it we don't have
an open connection somewhere that
someone can do anything they can get
access to so at this point most people
know that a device that's doing this is
called a file server if we then in the
same manner of taking this storage
server adding an abstraction layer to it
so that it looks like a non-standard
server we call it ass and we do the same
thing with file a little storage and we
call it a mass at no point is this not a
traditional file server it is simply
applying sized so that it looks like a
different device and takes
some of the options of doing some
general things so the reason that I
wanted to run through that is because
quite often when dealing with Santa Ness
we actually think of the world in terms
of fan of that we say well we have
storage right should I have santur Nass
and in reality those aren't those aren't
really the questions we're asking what
we should be asking is do we need
block-level storage or do we need files
at the lowest or and that's the big
question if you're running a database
and it needs to be able to interface
directly with the devices because it
does something really complex which is
basically have its own file system and
ignore the actual so you know if you're
running IBM db2 IBM db2 to these raw
devices supports discs because it has
its own file system that it's only for
the database has no other purpose so it
has to have block level access so they
can do anything at one for the drive
head but if you're dealing with normal
file sharing right Word documents Excel
documents all the stuff that users have
piled up everywhere yes you can do that
at the block level and attach it to a
single device yes you can go get really
expensive commercial file systems that
allow you to share that out like Oh F Oh
CF from pork right and GFS too from Red
Hat but you're getting into running big
UNIX boxes filled with do those things
so not really effective for that but if
you're running sis or NFS you can
connect all kinds of desktops to it you
can you know all the things that you
know how to do you can do so choosing
block storage or file system level
search is really really question and at
the end of the day you have a file
server one way or another that's doing
that work so at that point I'm just
gonna let people ask questions now and
they have prompt so I'm not sure if
they're like falling asleep for has
questions or
yep
yes okay
actually what's funny is typically mass
cost about 150 to 200 dollars and Sam
starts at under a hundred it's and
started close to 30 but it's not the
same people think of they think of Santa
thick the fan is actually a little one
so the names are actually awful right
network-attached storage and storage
area network
I hate these terms partially because
they're confusing FRC because they're
wrong storage area network is really
just a term for block level storage and
network attached storage is really just
the name for filesystem level storage so
when you're doing a block level you have
to attach as a device and when you're
doing filesystem level you have to have
a special virtual file system driver
like sis or NFS that allows you to put
it onto the network and share it over
the normal network
the idea was Sam in theory the reason
the word exists is because when they
first did it with fiber channel fibre
channel was its own thing you didn't
communicate over fibre channel for other
tasks so it was a storage area network
but then later very quickly and actually
at the time as well people would take
nass devices file servers because we
didn't call them NASA back then and put
them on a dedicated network with its own
switch connected to the servers on
dedicated NICs just for storage well
obviously that's a storage area network
using Nass so that the terms can overlap
even when you're doing dedicated
networks so that's why I hate the terms
but when we say Sam and we sometimes say
San protocols doing Fiber Channel I
scuzzy s CEO ata those things all right
SC o Ethernet sorry
and we use Nass to refer to the SIS NFS
AFS today after that okay cool
anything else before
in the back
yes a a SAN is like a LAN in that it is
a network yes
yep and I worked at a company that
actually had what we called the D bat it
was a database area network it was a
dedicated network in the same way as her
storage area network except it was
dedicated to database communications so
all the databases were on here
Ethernet switch Ethernet service so who
basically did the same thing really well
so there was question up here
I scuzzy is not as I'm aware any more
noisy then I'm assuming they're there
thinking it's broadcasting I'm not aware
that there being any significant
broadcast traffic unless they weren't
running switched Ethernet which would be
the bigger problem
yeah it's um no it's it's really it's
the it's DCP and it's non broadcast so
it's point-to-point same as any any
other for exalt that nature you know
yeah
it could be actually that's really good
point so for a customer like that
regardless of the noisiness of the
protocol when you have traditional
Sandpiper Channel will call traditional
your network is obviously completely
dedicated to that
but what's highly recommended is if
you're going to do I scuzzy or any other
you know sand that leverages commodity
networking as we like to say so Ethernet
you want dedicated Hardware and
dedicated network for that even so
because it's it's not necessarily noisy
but it is a really high volume
traditionally I mean you wouldn't bother
with it if it wasn't a certain life so
you want to have switches that aren't
doing anything but that in the same way
you would have done a fiber channel just
because you switch to I scuzzy from
fiber channel doesn't mean that you
should leverage your traditional land
that's already existing to do so because
you still have the same volume concern
you just want to put it on a dedicated
switch and treat it in the same manner
which is nice because when we move to I
scuzzy verses fiber channel you can't
afford to go faster you can afford to
have more redundancy you can get better
switches keeper and quite often it's
really really popular the more important
something is Frank your shortage area
networks the most important thing in
your business it's really really common
to jump straight to we need layer 3
switches we need lots and lots of
management we need that the biggest less
expensive things we can find the reality
is you probably don't care that much
about monitoring your surgery Network
you might but most of the time you're
actually gonna be better served getting
a cheaper unmanaged switch because what
you care about is latency and throughput
and so the the less management there is
the less layers there are the less
that's going on you don't wanna be v
landing on the switch that's handling
your eyes Guzzi you want it to be its
own not not virtual LAN you want to be a
LAN
if you need to be LAN get another switch
and use it physically actually segmented
because you don't want that overhead on
your storage because your storage
latency matters everything in your
company
so you want that to be as fast as
possible and which is actually cheap
that's the wonderful thing about speed
on Ethernet pretty much the cheaper you
get not consumer devices but with him
within a given product range generally
the cheaper they are the faster they are
anyone else before going
so you're talking about what where I
said there there are file systems that
allow you to do that commercial positive
them okay
I believe they're simply generally
referred to as shared access file
systems
maybe someone actually knows what the
generic term is for that family of but I
do know that Red Hat and Oracle are the
key players with that and I believe the
Veritas with the XFS does that as well
but I'm not a hundred percent sure I'm
definitely not an expert on TX of that
but GFS too is Red Hat's product so if
you look up go to Red Hat look at GFS -
I believe it's actually available for
free from them but it's a dedicated
these are file systems
so anything that attaches to that with
Sam has to have that file system driver
so you have to look for a vendor that's
gonna support whatever devices you're
using but yep that's the big one
but I don't work we'll have one - all
right all right okay so
a lot of people are familiar in the
community we've talked about the Sam SP
right the Sam St which I did not name is
not an actual product but it is what I
like to call a reference implementation
of open storage and the reason that we
came across with this is because in a
lot of conversations we talked about
with with companies and with people in
community you know people want out well
you want to put in the sand right so
they go to a vendor
everybody's gotta stay on these days
they go to the vendor and they say well
I need well you know I need block-level
storage so they come out with really
really expensive storage products and if
you're a giant fortune 500 it probably
makes sense because your storage devices
right when I worked at the bank our
storage devices are in the petabytes
right they have entire data centers
dedicated to the storage and we have an
OSI 192 running to other facilities to
carry those the fibre channel / - right
so we can lose an entire data center our
storage is still there and that sorta
just replicated to another data center
over OC 192 so unbelievable amounts of
storage there's no way you're gonna go
build this at home yourself
that's where players like EMC Clarion
come in and and Hitachi and build entire
rooms that makes sense but when you're
looking at more reasonable numbers of
storage you start getting into the space
where you're using traditional
technologies completely including
chassis so that's really what matters
here so I'm gonna give a little story on
this this is kind of the back story of
how the Santa Fe came and came into
being I work for where I'm a consultant
for of a major fortune 10 it's hard to
be a minor fortune-telling foot and they
were doing a global commodity grid you
can read about it online we're well
known they were over 10,000 notes and we
push an unbelievable amount of computing
through that lots of different lines of
businesses use it a lot of people like
to call cloud it is not its
high-performance grid computing it's
very related to clouds that's not the
same thing but we so it said it's an
application layer virtualization not a
operating system layer virtualization
r500 realization so that's kind of where
those government
so we run several dozen and maybe a few
score applications on this ten thousand
note grid to be able to back that grid
we don't have any storage on the nodes
except for the operating system just
makes it easy they boot locally but all
their data comes in from somewhere else
and then gets saved somewhere else we do
cache locally just approximately on this
we were working with we won't name names
but a very very very major storage
appliance member we had the second
largest product that they made it costs
really close to three hundred thousand
dollars per unit we worked with them we
brought up a new part of our grid and
the load demand on the grid turned out
to be higher than this device could
supply not from a throughput necessarily
actually from an eye off standpoint it
just couldn't handle it with the
spindles ahead and so we approached some
vendors and at the time another vendor
in the server space had brought out I
guess I'll name it son had brought out
what they called thumper which is a 48
drive for you server to processors 48
drives for you for you chassis
it's a traditional chassis you go to
your data center it looks like a regular
for you server nothing weird it just has
a lot of drive bays and they were
pushing this as a sort of think of this
in retro term let's go back to old
storage stop thinking that they actually
this is where the term open storage came
from when they really suffer son said it
is time to rethink storage storage
devices that everyone's been buying
Sanon ass are just servers that have
lots and lots of drives attached to them
well why not just buy a normal server
and use that because when we make normal
servers we can make lots of them faster
than your price goes way way
now when you buy sand and NASA labeled
devices you tend to get products that
are not sold on the same types of
quantities as commodity servers and
sometimes you use proprietary software
to do some of the cool features and this
drives the price through the roof
they are also considered a non commodity
so their margins are much higher the
margins on a traditional server you're
looking at the major players you know HP
Dell whatever Dell does not make a
thousand dollars off every server
yourself is up
they make twenty dollars right when you
buy the time you get all the discounts
done their margins are low so they're
not ripping off on a server there it
cost them a lot to deliver that to you
so you want to be buying those devices
if you can help it because that's where
your value is coming from when you go to
appliance size products you generally
have to pay a lot for the main master
the name sent and so so what Sun did
something actually came in and worked
with us and they knew they weren't
getting this account but they worked
with us anyway because they hated the
other vendor we were competing against
and and we said to them we we really
feel that this this device that we have
is very very expensive and doesn't seem
to be doing as much as we could just do
with a regular file server and son said
absolutely a regular file server gives
you all these advantages there's the
commodity you can tune the operating
system you can pick the operating system
you can pick every component and you can
do it much cheaper and they actually
flew in the guy who invented ZFS to talk
to us about it it was awesome and so we
said well we went to the client and we
said we would like to do an engineering
study and we want the storage vendor
involved I said ok they ordered the
largest product that they made the
largest mass device on the market
there's a couple years ago so it's this
figure now this was the it was a half
million dollars and it was installed and
tuned by the storage vendors own
engineers they had a lot of money
because we weren't buying one we're
looking to buy like it doesn't
so they brought in a lot of resources to
make sure this was going to beat
anything we could do we took two people
from the engineering team with a couple
hours
we took a commodity server that we had
now is it was a large server at the time
it was a four-way Opteron box but it
would be considered a small server today
it's probably about 1/3 the processing
capacity of what you would get for run
for $5,000 today so still a decent
server but and that's the time pretty
impressive but nothing if we just fold
it remember joint we loaded Red Hat
Linux on it no tuning we got normal
storage nothing special we set it up
with NFS which is exactly how they were
connecting to the other box and we did
and before we ran it we projected what
was going to happen that we knew there
were threading issues on the processing
side of the storage vendors product
because it was not an open box and they
could not update their operating system
for the latest kernel which they needed
to do because they weren't making their
own operating system they were getting
it for another vendor and they didn't
have time to rework the operating system
and we had the ability to run the latest
Red Hat which had great threading
support which was needed to be able to
push the I office and when we ran it not
only did our at the time $20,000.00
solution which cost about it was
literally do you be able to put together
for about two to three thousand today I
expect not only do we outperform a half
million dollar server turn tuned by
their engineers but instead of flat
telling and having them we actually have
all the performance curves we have no
idea what the capacity of the open
scratch-built box was because we the
grid could not request enough I off step
pressure in the half-million-dollar
device not only plateaued but when it
went on the plateau for very long it
actually shut down
so the the potential savings here were
not just millions of dollars of
purchasing but that this product met the
need while this product did not it was
easier for us to manage because we
didn't have to have dedicated storage
people we use the skill set we already
had we already have the engineers for
this will just manage it along with all
the other servers it will look exactly
like all the other services so this
experience and for those who wonder no
they didn't go with that they went with
the expensive solution in the project
help the welcome to the fortune 10 so
what that prompted was later when when
Niagara started looking at storage
options and we started having a lot of
conversations in the community about how
do you do storage how do you make it
cost-effective what do you do when you
have all these needs and needs
flexibility and I can't afford these
other products oh we looked at the
product market and we said wow you know
you go to any major server vendor right
ones that are here ones that aren't
anyone who's a top-tier vendor and they
have these chassis that are very
affordable that have a lot of space for
desks some have more than others some
have different price points but they're
all relatively affordable and powerful
stable and manageable and they fit into
your infrastructure just like everything
else you can go get third-party disks
for them some support that a little bit
better than others but most of them have
complete open support for any gift you
want into it you can put lots of disks
into them you control their speed you
control their configuration to control
the models if you have a specific drive
vendor that you're very very comfortable
with you can pick them to get you all
that and you're building systems for a
few thousand dollars that not only might
outperform a 30 or 40 or $100,000
commercial appliance eyes sand or nest
device but you also have more control
over it you have
the event this is the most important
thing with any computing device remember
it's just a server there's no magic
right everybody thinks well I'm gonna
get its and I can let everything else
fails cuz the sand won't fail but the
sand is just a server like everything
else right there are better ones they're
cheaper ones but it's just a server
right it's always subject to the
forklift risk right but someone's gonna
drive the forklift into that one box and
it just absolutely happens right from
some real example and so when you cut
the cost dramatically when $30,000 was
it was a consideration but now you can
do the same thing for $5,000 don't cut
your budget by $25,000 cut your budget
by $20,000 and get two of those and now
you can now they can you can use them in
the same way you do with anything we're
done and that doesn't have to be on a
scale - it could be on a scale of 50
right you were gonna buy 50 or 25
commercial sands now buy 50 of these and
build things and that's that's an option
when you get really really big right it
starts to maybe not make sense really
large sands have capacity to put lots
more drives and and they're much more
manageable and on a really massive scale
so if there are price points and there
are feature points where traditional
sands start to make a lot of sense but
they almost never do when you're in a
capacity range where you are working
with a single traditional commodity
server chassis capacity as a lot of ways
basically if you have a normal server
you can buy off-the-shelf from your
preferred vendor wherever you're working
with now and once you're working with
some white box builder and then stop and
go get it enterprise bender either if
you're dealing with an enterprise vendor
go to them get their price for the
chassis that makes sense it's almost
always a to you I know
Dell is here and they've got a box that
holds 24 2.5 inch drives in a to you
right pretty unbelievable
if it's 24 2.5 inch drives meets your
needs you've got all that storage and
that it's that's potentially really fast
sure
well before I answer that exact question
because this actually came up last night
almost exactly the same thing right so
when I talk about storage I often talk
about Red Hat because that's how we do
storage which is not actually true we do
a little bit of Red Hat most of our
storage is actually solaris cousins that
are io throughput but in either those
cases you're dealing with an operating
system that chances are the majority of
you whether that's 51 percent or 80
percent I don't know but most people in
this community are not unix proficient
it's not part of your infrastructure
it's not something you manage on a daily
basis if it is it's definitely
consideration but if it's not it doesn't
matter because Windows is an absolutely
wonderful storage platform in the same
way that UNIX is and it's just in this
example is that we ran UNIX because
that's that's what we did were for doing
administration windows make some really
powerful storage stuff they do I scuzzy
they do cess they do NFS it's all free
it's all included you're not buying
extra products and they're sips if
you're doing Active Directory
Integration it's by far the easiest to
deal with works the best most reliable
but if you don't want to go with Windows
as your storage and you want to go with
someone like Red Hat as an example you
have lots of options even if you don't
have the in-house expertise there are
lots of MSPs who will do that obviously
pretty much you can always find an
atmosphere do something for you know
what someone killed you find it but
there really are your storage devices
are something that need relatively
little monitoring they need monitoring
but you're not you're probably not
concerned with you know capacity
planning other than the amount of
storage and you can watch that right
Spiceworks will monitor it and tell you
how much it's being used so as long as
you're dealing with that kind of stuff
you're probably not dealing with in a
normal small business situation you know
CPU capacity memory capacity you've got
more than enough in the smallest box
that those dozen companies like Red Hat
if you actually get Red Hat commercially
you will get support from Red Hat
themselves right or if you're getting
Susa from Novell they don't sell in the
same way that Windows does Windows is
based on a license and the commercial
Linux players are based on support
contracts so that support is okay or
those I know a lot of people in here
like I'm going to
canonical
the exact same thing for free play with
it and when you're ready to go live you
can contact canonical and yet torsional
support directly comes em as a primary
vendor or to any number of msps who
would provide that support and of course
you get training anything else that's
your question okay
I do have opinions on them I have run
across them um I don't like their
pricing structure I feel that their they
are hurting themselves in that way I
think they should have a free version
that is more scalable but as a product
based on open Solaris if it fits into
your price range and it's not that
expensive right it's a very if you're
looking at this range of stuff this is I
think it's a really good product I have
not used it in the commercial capacity
so there may be some gotcha so I'm not
familiar with but the underlying
technology it is open Solaris is awesome
right lots and lots of power lots and
lots of flexibility lots of options and
very easy to manage and that's something
I should mention is the attenti is a
it's a NASA appliance right I can't
believe I forgot some ends of it so we
have traditional servers file services
just you know windows a red hat or
Solaris whatever and you're doing
everything and then we have the
appliance that's right you can go to
neck here you can go to Buffalo you go
to Drobo you can go to EMC and ecologic
and HP everybody right everybody has
these full appliances but there's also a
middle ground or you're using the
hardware like the commodity hardware
from anybody and then applying the
operating system that is a appliance
operating system so next enta is a great
example of that it's one that's built on
open Solaris 3 NASA is the same type of
thing completely free built on FreeBSD
and open filer is the same thing that
built on connery Linux which
unfortunately is a very unpopular
version of Linux and it's not monitored
by anything and the valve management
stuff is funky so that's unfortunate but
and there is a fourth player and I can't
remember their name he's definitely the
small tear in that left hand from HP
used to be one of those players and when
they got bought by HP they kind of
two combined hardware so they kind of
moved into that side instead of being in
the software space so but for people who
want to work with Linux and don't want
to work with Linux or want to look at
BSD and don't know what the BSD those
solutions give you those operating
systems with those operating systems
advantages and disadvantages without
having to know those operating systems
and one actual caveat to mention is if
you're gonna work with open file are
very powerful it is the best for
replication of any product along with
all the big commercial ones this
replication is phenomenal but there's no
interface for it you are you will need a
senior Linux admin fees to do that with
freedom the only couple hours we are in
officially in the QA I think we have
five minutes I need five minutes of
questions
ready now
well so full disclosure company I work
for is a partner with Netgear so we have
to say we'd love ready now but we love
ready now definitely it my personal
preferences if you're gonna be working
in the space where you want an appliance
mask and you you know you want all the I
just want to buy it right I don't want
to be ready NASA's a really really great
product is based on Linux it does not
have dr BD replication we are pushing
them for that that doesn't mean they'll
do it but we have pushed for other
things we are doing it so there are some
caveats with readiness and I'm not
allowed to tell you but so I'm not gonna
mention what the caveat are but I can
tell you since I didn't tell you what
they are that they're going away in
December so readiness it's a great
product and we've priced it versus
building a Sam st and it's within like
10 percent of cost and there is someone
on my team who runs one and was it yes
I'm sorry Don
I have I don't have experience on it so
I'm I can't really compare it to
anything unfortunately can't really
answer that very well
uh-huh
that
if you're getting the 24 bay from from
Dell the 2.5 inch chances are if you're
buying that unit it's because you want
15 K drives chances are just because you
selected that chassis that's like why
that chassis exists you don't have to
when you're choosing both your your raid
levels right and everybody knows but
mostly you know probably that I
absolutely hate raid 5 but the reality
is if you're in a archival situation and
it's not a live system and it's backed
up and all you want it to be is online
most of time and you're willing to take
a little bit higher risk raid 5 can save
you a lot of money I would not use it
for online systems but I would use it
for Nearline which is kind of a lot of
small businesses don't do near line
storage so but when it comes to actually
selecting your spindles it's really a
question of price and versus I ops right
and so if you know if you're gonna go
with SATA
you just have to have more of them but
they cost less so that could be very
beneficial and typically you're gonna
get more storage while you do it so you
might be like oh here's the option for
SAS at 15k here's the option for SATA at
7.2 K and at the price point where it
gives you the performance you need this
one likely is gonna give you 2 or 10
times the actual storage capacity that
might be a winner but it also might be
so many doesn't fit in the chassis you
want to get and there's a little looser
so and as you have more drives they are
more likely to fail right just 20 drives
are more likely to fail them - so there
are risks there but just doing a
calculation of performance is really the
only the only factor there there's no
there's no guaranteed thing and a lot of
commercial Santa nests are only SATA
because they just add more of them
well so with raid5 compared to and it's
not just right by the way it's called
the r8f family which is right two three
four five and six they use what's known
as a soar calculation and what that does
is there's obviously a stripe across the
disks and you get great capacity out of
it and that's why it was is why they
spent the effort to invent it the way
that that works is the rate controller
has to do whether it's software hardware
has to do a lot of work to make that
stripe work and because of that the rate
controller becomes a significant point
of failure compared to rate an array one
which doesn't have as or calculation so
the risk that you get beyond performance
issues the sort calculation causes
performance issues additionally but
performance is a arguable point right do
you care about performance do you not
care about performance but losing your
data everyone cares about and I have had
first hand which will really convince
you but I also know other companies who
have had raid controller failure on a
raid F array node rise fail everything
lost because the RAID controller freaked
out and has a destructive operations
where it destroys all the disks great
one and rate ten do not have a
destruction operation to perform on the
disks well they do a rebuild it is a
mirror if they were to mirror a good
disk it would build a new healthy disk
but if a raid 5 attempts to rebuild an
unhealthy system it will destroy a
healthy one
and so if raid F fails its action in
failing is to scrap everything and I've
seen I have definitely seen that
firsthand caused by chassis shudder in a
in a data center it was a server that
was in use for years drives came in and
out of contact over we're not there's a
period of minutes or a period of hours
it kicked off multiple rebuild
operations and one of them just posed
tire array so when we found it they had
all receded themselves and we had six
healthy discs to help you rate
controller and no data I think we're at
it and we're done