Cloud computing ( was Re: Linux Petition )

[JLMS]

I know my employers would be reluctant to look at using google
calendar as they'll be concerned about confidential
data/meetings/shared documents being on a system outside of our own
control, but I'm sure this is more to do with perception and adapting to

change.
The same thing could be said of any part of a business using 'cloud' services.

And, quite frankly, should be.
Andy

Well, that is going to change, so better one jumps now into the bandwagon.

All major companies (Oracle, Microsoft, Red Hat, Amazon) are pushing
very decisively on the direction of cloud nirvana.

All of them are mindful of the obvious concerns of seasoned technical
people, their bosses and lawmakers, right now at least Red Hat knows
they should do something about where your data stays in order to
comply with the respective legislation, I think Amazon is going that
way as well and I would be shocked if the others are not working about
this, Oracle is shamelessly talking about Solaris 11 as a "cloud
computing" operating system (and the revamping of their network
management utilities plus solid virtualization and storage management,
i.e. ZFS, would indicate they are serious about it and have the tools
to deliver).

The question is: if all your data and communications are properly
secured (encrypted disks in Red Hat, Solaris work quite well for
example, and with SSDs and forever more powerful machines the
traditional technical penalties for encrypting are becoming less of an
issue), what exactly would be wrong with having your data in the so
called cloud?

Many companies have all or part of their IT infrastructure in somebody
else's datacentre already and they are escalating from nought to
millions of users in the comfort of scalable hosting services provided
by somebody like Amazon, so you are going to get more and more Sys
Admins, middle managers and bean counters that will be familiar and
comfortable with this way of working (and lets face it: who really
wants the mess of dealing with a data centre if your core business is
something else entirely different?)

I think the writing is in the wall, the next set of skills one will
need to have is familiarity with how to work with hosts (physical and
virtual) that are located goodness knows where ensuring the safety of
the data being handled...
--
Gllug mailing list - ***@gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug

[Nix]

Post by JLMS
All major companies (Oracle, Microsoft, Red Hat, Amazon) are pushing
very decisively on the direction of cloud nirvana.

Not only. Also 'you buy boxes from us to do ${database services} ${local
searches} et al. That's sort of the oppoosite of this cloudy thing (at
least it is if you make sure those boxes can't send arbitrary stuff out
onto the network, which you probably should if you know what's good for
you).

At least the Oracle versions of these boxes and very probably the Amazon
and Google versions too are Linux boxes. (This is about as far from a
secret as it is possible for anything to be.)

(disclaimer: not speaking for my employer, duh)

Post by JLMS
Many companies have all or part of their IT infrastructure in somebody
else's datacentre already

Part of me is wondering how much of all this was orchestrated by
lawyers. I mean, the first time something goes wrong or another
relationship with one of those suppliers gets mucked up, the legal
fights are going to be absolutely *epic*. Megaupload has nothing on
this. If it were just a hosting company, that's one thing, but Oracle,
MS and Amazon at least do lots of other things too and the first two
trade with basically everyone. Any of those relationships going awry
with a customer who is also a cloud customer could lead to...
interesting (I-want-to-be-very-far-away) territory. (I'm sure lawyers on
all sides have already prepared for this and minor skirmishes have
already happened, but do you seriously believe there isn't going to be a
masssive well-publicised fight sooner or later? I don't.)

[Tethys]

Post by JLMS
Well, that is going to change, so better one jumps now into the bandwagon.

That's about the worst possible reason I can think of to make a
technical decision.

Post by JLMS
All major companies (Oracle, Microsoft, Red Hat, Amazon) are pushing
very decisively on the direction of cloud nirvana.

Yes, because they see profit there. No other reason. I won't be
doing so because, well, there be dragons.

Post by JLMS
The question is: if all your data and communications are properly
secured (encrypted disks in Red Hat, Solaris work quite well for
example, and with SSDs and forever more powerful machines the
traditional technical penalties for encrypting are becoming less of an
issue), what exactly would be wrong with having your data in the so
called cloud?

Someone else has your data. Not that they'll be able to get at it
if it's suitably encrypted[1], but they can deny you access to it.
For that reason alone, I won't be putting any of my data in the
cloud now, or for the forseeable future. Or probably ever.

Tet

[1] Although most data in the cloud isn't.
--
Gllug mailing list - ***@gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug

[Simon Wilcox]

Post by Tethys
Someone else has your data. Not that they'll be able to get at it
if it's suitably encrypted[1], but they can deny you access to it.
For that reason alone, I won't be putting any of my data in the
cloud now, or for the forseeable future. Or probably ever.

That doesn't mean that you shouldn't put your data in the cloud, it just
means that you had better make sure you have a good backup of it. Which
of course weakens the business case for data in the cloud but other
factors may still make it a good business choice.

S.
--
Gllug mailing list - ***@gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug

[Karanbir Singh]

Post by Simon Wilcox

Post by Tethys
Someone else has your data. Not that they'll be able to get at it
if it's suitably encrypted[1], but they can deny you access to it.
For that reason alone, I won't be putting any of my data in the
cloud now, or for the forseeable future. Or probably ever.

That doesn't mean that you shouldn't put your data in the cloud, it just
means that you had better make sure you have a good backup of it. Which
of course weakens the business case for data in the cloud but other
factors may still make it a good business choice.

not sure if you and Tethys are talking about the same thing... His point
is more about privacy and ownership/access of data : backups wont help
if everyone else gets access to your data.

I think there is merit on both sides. Encryption works to some level to
mitigate the idea of data being worth something to anyone else other
than the key holder, but then on the flip side you need key instance in
the cloud in order to consume the data locally anyway. So threat
remains. not all data needs to be private, eg: the data hosted by the
guys at GDS ( gov.uk ) can mostly be ( and should be ) very visible.
Till such time as my personal details get added in there, and they have
an auth mechanism that allows my tax and personal details to make their
way into the 'system'. at that point, *I* really want them off AWS and
into gurantee'able secure storage, processing and interface.

No worries about council publications all being in the cloud,
distributed from there, accessed from there.

- KB

[John Edwards]

On Mon, Apr 23, 2012 at 12:43:05PM +0100, Karanbir Singh wrote:
<snip>

Post by Karanbir Singh
not sure if you and Tethys are talking about the same thing... His point
is more about privacy and ownership/access of data : backups wont help
if everyone else gets access to your data.
I think there is merit on both sides. Encryption works to some level to
mitigate the idea of data being worth something to anyone else other
than the key holder, but then on the flip side you need key instance in
the cloud in order to consume the data locally anyway.

<snip>

Data can be encrypted in storage (disks) and transit (network), but
I don't think it can be encrypted in memory if you want to use it.

That means that a superuser on virtual host can read the memory
of any guest machine. That will include a lot of unencrypted
sensitive data, although it would not be easy to sort through.

I'm not aware of any way of preventing this, but would happy to
be proved wrong.

[Duane Griffin]

Post by John Edwards
Data can be encrypted in storage (disks) and transit (network), but
I don't think it can be encrypted in memory if you want to use it.
That means that a superuser on virtual host can read the memory
of any guest machine. That will include a lot of unencrypted
sensitive data, although it would not be easy to sort through.
I'm not aware of any way of preventing this, but would happy to
be proved wrong.

AIUI (and to be clear, I don't, really) this is an active research
problem which probably *is* soluble, although not in anything
approaching practical terms at this point in time.

See http://crypto.stanford.edu/craig/easy-fhe.pdf, which I think was
published in CACM a couple of years ago.

Post by John Edwards
#---------------------------------------------------------#
#---------------------------------------------------------#

Cheers,
Duane.

[Karanbir Singh]

Hi John,

Post by John Edwards

Post by Karanbir Singh
I think there is merit on both sides. Encryption works to some level to
mitigate the idea of data being worth something to anyone else other
than the key holder, but then on the flip side you need key instance in
the cloud in order to consume the data locally anyway.

Data can be encrypted in storage (disks) and transit (network), but
I don't think it can be encrypted in memory if you want to use it.

right, thats what I meant about needing the key used for encryption
needing to be available locally if there is going to be a need to
process the data locally in the cloud.

Post by John Edwards
That means that a superuser on virtual host can read the memory
of any guest machine. That will include a lot of unencrypted
sensitive data, although it would not be easy to sort through.

so yes, need to trust hypervisor - but in some cases, also trust all
your neighbours to some degree. its been possible to bring down a
physical machine from inside a xen domU.

Also, since the state of what is visible inside the domU is controlled (
and potentially manipulated ) from the physical host, it makes things a
lot less secure.

[Andy Millar]

Not specifically following on from anything so far in this thread, but:

http://vmblog.com/archive/2012/04/24/dirty-disks-raise-new-questions-about-cloud-security.aspx

Andy
--
Gllug mailing list - ***@gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug

[Simon Wilcox]

Post by Karanbir Singh

Post by Simon Wilcox

Post by Tethys
Someone else has your data. Not that they'll be able to get at it
if it's suitably encrypted[1], but they can deny you access to it.
For that reason alone, I won't be putting any of my data in the
cloud now, or for the forseeable future. Or probably ever.

That doesn't mean that you shouldn't put your data in the cloud, it just
means that you had better make sure you have a good backup of it. Which
of course weakens the business case for data in the cloud but other
factors may still make it a good business choice.

not sure if you and Tethys are talking about the same thing... His point
is more about privacy and ownership/access of data : backups wont help
if everyone else gets access to your data.

I think we are, although specifically only about denying access to data.
I agree that preventing unauthorised access to that data is a different
issue.

If you rely on the cloud provider to maintain your backup as well as
your primary data you risk being denied access to both backup and
primary if anything happens to that provider (goes bust, gets shut down
by the feds etc).

Simon.
--
Gllug mailing list - ***@gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug

[Bernard Peek]

Post by Simon Wilcox
If you rely on the cloud provider to maintain your backup as well as
your primary data you risk being denied access to both backup and
primary if anything happens to that provider (goes bust, gets shut
down by the feds etc).

Assuming you use the same cloud provider for live services and backup.
That would be unusual I think. Of course there's always a possibility
that there's a single point of failure somewhere and you don't always
know where they might be hiding.

[Simon Wilcox]

Post by Bernard Peek

Post by Simon Wilcox
If you rely on the cloud provider to maintain your backup as well as
your primary data you risk being denied access to both backup and
primary if anything happens to that provider (goes bust, gets shut
down by the feds etc).

Assuming you use the same cloud provider for live services and backup.
That would be unusual I think. Of course there's always a possibility
that there's a single point of failure somewhere and you don't always
know where they might be hiding.

Very true but I suspect that many customers will be attracted by the
data durability claims of many cloud providers, or that "all data is
fully backed up". I fear that using the same cloud provider for both
live and backup is probably not as unusual an occurrence as we might think.

Simon.
--
Gllug mailing list - ***@gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug

[David Colon]

Post by Tethys
Someone else has your data. Not that they'll be able to get at it
if it's suitably encrypted[1], but they can deny you access to it.
For that reason alone, I won't be putting any of my data in the
cloud now, or for the forseeable future. Or probably ever.

The same caveat applies if your servers are in a data center that you do
not own. Your physical and network access can be removed by the ISP at any
time. You still need to maintain backups of your data in a separate
location.

David

[Karanbir Singh]

Post by Tethys
Someone else has your data. Not that they'll be able to get at it
if it's suitably encrypted[1], but they can deny you access to it.
The same caveat applies if your servers are in a data center that you do
not own. Your physical and network access can be removed by the ISP at
any time. You still need to maintain backups of your data in a separate
location.

Sure, but for very different levels of 'barriers to access'. For someone
to get access to your data in a colo facility the amount of effort would
( should! ) be a lot higher than being able to access something from a
shared resource. Besides, there would be potential to monitor for
external access to your data in a colo facility, in the cloud you might
never find out how else got access to it.

[Simon Wilcox]

Post by Tethys
Someone else has your data. Not that they'll be able to get at it
if it's suitably encrypted[1], but they can deny you access to it.
For that reason alone, I won't be putting any of my data in the
cloud now, or for the forseeable future. Or probably ever.
The same caveat applies if your servers are in a data center that you
do not own. Your physical and network access can be removed by the ISP
at any time. You still need to maintain backups of your data in a
separate location.

If you're running your own servers of course you send backups off site.
You'd be insane not to.

In addition, you should make sure that the offsite location isn't
somewhere to which you can be denied access at the same time by the same
provider. e.g. not another datacentre operated by the same provider
where you'll be locked out of both if the provider goes bust.

S.

[Jason Clifford]

Post by Simon Wilcox
If you're running your own servers of course you send backups off
site. You'd be insane not to.
In addition, you should make sure that the offsite location isn't
somewhere to which you can be denied access at the same time by the
same provider. e.g. not another datacentre operated by the same
provider where you'll be locked out of both if the provider goes bust.

Backups are important (essential even!) however people often forget that
a backup is useless unless you can use it in a timely manner so it is
just as essential to ensure you can gain access to and bring into use
any backup very quickly and to actually test this with some frequency.

This means that copying to another data centre is good only to the
extent that you can then bring services into operation at that backup
data centre quickly.

Similarly copying to a backup at home or in the office is only any good
if you can quickly get that backup into service in a meaningful way.

A lot of backup facilities are useful only for storage but not for
actually using the data in situ. If you have a backup of several hundred
GB it may take days or even weeks to transfer that data to a new site
and so restore service. If that happens very often you may as well not
bother as your business is already dead.



--
Gllug mailing list - ***@gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug

[JLMS]

Post by Tethys

Post by JLMS
Well, that is going to change, so better one jumps now into the bandwagon.

That's about the worst possible reason I can think of to make a
technical decision.

Well, the day implementations are solely based on technical reasons it
will be heaven on earth and I'll sing hallelujah!

In the meantime the bean counters are being lobbied with insistence
about the advantages of "the cloud" (and have many real life business
cases, I wonder how many of those hot startups actually have their own
datacentre, for many young bright things under 25 a datacentre may be
as exotic a concept as a time shared system was to me when I started
to learn programming on a shiny new 8086 compatible PC, but my first
boss swore that UNISYS would dominate the computing industry for years
to come with their new mini computers that fitted on top of a desk,
thus replacing mainframes but I digress... )

So my contention point is that the cloud is coming, so better one gets
ready for it.

In many situations it is immensely cheaper to ask somebody else to
host your development and prototyping for example, or use massive
computer power that would otherwise be out of one's reach.

I just can't imagine a University that would not want to use cloud
providers for example: start instances of powerful machines for your
students during the term, shut them down during holidays and pay
nothing when you are not using them. Yes I know, it is a like a
mainframe, but as far as the user is concerned, this time you can
actually turn the thing off whenever you want.

You can also use thousands of cores to throw at a problem for a few
thousand pounds instead of expending hundreds of thousands or even
millions of pounds on a system of comparable power.

Post by Tethys

Post by JLMS
All major companies (Oracle, Microsoft, Red Hat, Amazon)  are pushing
very decisively  on the direction of cloud nirvana.

Yes, because they see profit there. No other reason. I won't be
doing so because, well, there be dragons.

And your point is? What is wrong with them making a profit if they are
offering a platform that may be possibly improving the way one uses
computing resources?

Of course they will make a profit, that is not worth even mentioning,
the question worth asking is if what they are offering makes my life
as a Sys Admin, IT head, or whatever, easier, or if it saves me money
or both within reasonable constraints that ensure my data remains
safe.

Post by Tethys

Post by JLMS
The question is: if all your data and communications are properly
secured (encrypted disks in Red Hat, Solaris work quite well for
example, and with SSDs and forever more powerful machines the
traditional technical penalties for encrypting are becoming less of an
issue), what exactly would be wrong with having your data in the so
called cloud?

Someone else has your data. Not that they'll be able to get at it
if it's suitably encrypted[1], but they can deny you access to it.

Not as long as you pay :-)

The same thing happens with collocation in a 3rd party data centre
and most people have got over that, the next natural step is to use
the flexibility provided by somebody that virtualizes and slices those
resources.

Post by Tethys
For that reason alone, I won't be putting any of my data in the
cloud now, or for the forseeable future. Or probably ever.

You don't have to move all your infrastructure to the cloud, there are
so many cases in which "the cloud" is a better solution to many
problems that I can't understand the negativity.

During prototyping of a solution for example you would not be using
real data, but would be able to size which machine you would need, do
tests, change loads, etc, almost at the click of a button.

In some of my previous jobs it took months to provide one machine from
the moment the money was released to the point one was booting it for
the first time. Now you can shorten that time by an order of
magnitude, and once you are done with the development & prototyping
you are not left wondering what will you do with all that equipment
you no longer need.

Gosh, you can provide a server farm with a script from the command
line: no pushing servers around, no messy cabling, power consumption
problems, etc, etc, etc ad nauseam.

There are even companies that in a similar fashion to a price
comparison site will allow you to chose the cheapest or fastest or
less loaded solution in an automatic manner using parameters provided
by the user.

And if all that fails, you can even ask that they kick everybody out
of a piece of hardware and let you use it exclusively.

When I first built a LAN, or connected my PC to the Internet my gut
screamed "this is the future mate, get ready", well, after several
years of silence the gut is screaming again, so I''l follow the advice
of this sage that has served me well....

Post by Tethys
Tet
[1] Although most data in the cloud isn't.

Well, that would be a dereliction of duty of the Sys Admin, the
various cloud providers I have talked to often suggest as a good
practice the encryption of your data volumes. At least in the
Linux/Solaris world this is easy peasy, since in both cases one can
have encrypted volumes (I don't know AIX, or HPUX, but I suppose it
would be the same...).
--
Gllug mailing list - ***@gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug

[John Hearns]

Post by JLMS
You can also use thousands of cores to throw at a problem for a few
thousand pounds instead of expending hundreds of thousands or even
millions of pounds on a system of comparable power.

I very much agree with you there.
Especially in the biosciences market, cloud is being pushed as a
solution for problems like gene sequencing.
As a for instance, see this article on a 50 000 core machine which was
assembled as a demo on Amazon:

http://www.hpcinthecloud.com/hpccloud/2012-04-24/cycle_spins_up_50_000-core_cluster_in_amazon_cloud.html

Cloud also is very good for companies which have a 'bursty' workload.
You might have workstations and storage within your company, and the
software packages you need to do some design or simulation work. When
your consultancy wins a contract to (say) simulate vehicle crashes you
could hire cloud resources to do that.
Software vendors are already providing on-demand licenses which can be
accessed via a web server and a key, rather than the tradiational
model of running a Flexlm server within your company (yes, I know you
can VPN tunnel etc.)
I can think of one well-known company based in west London which does
just this sort of on-demand licensing.

Also re. assembling supercomputers on the cloud, I was very impressed
by a short talk from Amazon at the Intel Sandybridge launch in London.
They now provide large machine instances with a 10 gigabit connection,
which you need if you are going to run parallel MPI type codes across
many systems.
--
Gllug mailing list - ***@gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug

[Bernard Peek]

Post by JLMS
You can also use thousands of cores to throw at a problem for a few
thousand pounds instead of expending hundreds of thousands or even
millions of pounds on a system of comparable power.

That's correct but misleading. You can rent a vast number of cores for a
limited period for less than the cost of building a comparable hardware
solution.

If you intend to use a vast number of cores for a very long time then
it's still cheaper to buy rather than rent. Except that if you build the
system yourself you have it on the books as an asset that you can only
write down at a rate agreed by the tax man. Cloud solutions are always
paid for from the operating budget not the capital budget. That has tax
advantages, until the Chancellor changes the rules again.

[JLMS]

Post by JLMS
You can also use thousands of cores to throw at a problem for a few
thousand pounds instead of expending hundreds of thousands or even
millions of pounds on a system of comparable power.
That's correct but misleading. You can rent a vast number of cores for a
limited period for less than the cost of building a comparable hardware
solution.
If you intend to use a vast number of cores for a very long time then it's
still cheaper to buy rather than rent. Except that if you build the system
yourself you have it on the books as an asset that you can only write down
at a rate agreed by the tax man. Cloud solutions are always paid for from
the operating budget not the capital budget. That has tax advantages, until
the Chancellor changes the rules again.
--
Bernard Peek

Absolutely. There are many things for which using a cloud solution is
not optimal, but dismissing the option off hand looks reckless to me.
--
Gllug mailing list - ***@gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug

[Richard W.M. Jones]

Post by JLMS
The question is: if all your data and communications are properly
secured (encrypted disks in Red Hat, Solaris work quite well for
example, and with SSDs and forever more powerful machines the
traditional technical penalties for encrypting are becoming less of an
issue), what exactly would be wrong with having your data in the so
called cloud?

Unless you've worked out an efficient homomorphic encryption scheme, I
have to tell you that encrypting the disks on your cloud machines
makes not the slightest bit of difference to the security of your
cloud data.

Rich.

[James Courtier-Dutton]

Post by Richard W.M. Jones

Post by JLMS
The question is: if all your data and communications are properly
secured (encrypted disks in Red Hat, Solaris work quite well for
example, and with SSDs and forever more powerful machines the
traditional technical penalties for encrypting are becoming less of an
issue), what exactly would be wrong with having your data in the so
called cloud?

Unless you've worked out an efficient homomorphic encryption scheme, I
have to tell you that encrypting the disks on your cloud machines
makes not the slightest bit of difference to the security of your
cloud data.

Encrypting disks provides protection for "data at rest"
i.e. It protects when the computer is switched off. So in the cloud it
protects when the vm instance is shut down.
I agree, that for a cloud based always on service, not a lot of time is
spent in the off state.

[JLMS]

Post by James Courtier-Dutton

Post by Richard W.M. Jones

Post by JLMS
The question is: if all your data and communications are properly
secured (encrypted disks in Red Hat, Solaris work quite well for
example, and with SSDs and forever more powerful machines the
traditional technical penalties for encrypting are becoming less of an
issue), what exactly would be wrong with having your data in the so
called cloud?

Unless you've worked out an efficient homomorphic encryption scheme, I
have to tell you that encrypting the disks on your cloud machines
makes not the slightest bit of difference to the security of your
cloud data.

Encrypting disks provides protection for "data at rest"
i.e. It protects when the computer is switched off. So in the cloud it
protects when the vm instance is shut down.
I agree, that for a cloud based always on service, not a lot of time is
spent in the off state.

But one is supposed to be encrypting communications also (VPN, ssh, SSL, etc).

I don't see what is left uncovered ...
--
Gllug mailing list - ***@gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug

[Bernard Peek]

Post by JLMS

Post by James Courtier-Dutton
Encrypting disks provides protection for "data at rest"
i.e. It protects when the computer is switched off. So in the cloud it
protects when the vm instance is shut down.
I agree, that for a cloud based always on service, not a lot of time is
spent in the off state.

But one is supposed to be encrypting communications also (VPN, ssh, SSL, etc).
I don't see what is left uncovered ...

As I see it this depends on how you use the cloud. If you use it simply
as a data store then it's securable. Data must be retrieved from the
cloud storage and decrypted on a local machine. That's not how I
understand the usual usage to be. Data is held in the cloud, possibly
encrypted. Then an access key is uploaded to the cloud where it is used
to decrypt some or all of the data. The cloud now has the encrypted
data, the key and the plaintext. It's compromised.

[Andy Smith]

Hello,

Post by JLMS
But one is supposed to be encrypting communications also (VPN, ssh, SSL, etc).
I don't see what is left uncovered ...

The keys for your encrypted data exist in the memory of the virtual
machine, which is readable by whoever has access to the metal.

I would say however that most attacks are simplistic and that
encrypting data prevents a lot of the simple attacks.

e.g. the recent Linode exploit where many tens of thousands of $
equivalent of bitcoins were stolen relied upon the attacker using a
bug in Linode's web interface to shut the VPS down and reset its
root password. That wouldn't have worked if the filesystems were
encrypted and also would not have worked if the actual bitcoin data
files were encrypted.

Cheers,
Andy