Take a look at this:

http://en.wikipedia.org/wiki/Amazon_Web_Services

That's... uh... wow, what the heck?

OK, so some of this stuff appears to make sense. Other parts... don't. 
(Most especially, "Amazon Mechanical Turk". WTF?) I'm not really seeing 
what SQS would be useful for. And I'm also wondering just how much all 
this stuff actually *costs*...

-- 
http://blog.orphi.me.uk/
http://www.zazzle.com/MathematicalOrchid*

Post a reply to this message

I'm also loving the concept of "elastic clouds" and "computational 
clouds". I mean, really, lay off the acid, guys!

-- 
http://blog.orphi.me.uk/
http://www.zazzle.com/MathematicalOrchid*

Post a reply to this message

Orchid XP v8 wrote:
> (Most especially, "Amazon Mechanical Turk". WTF?)

The only WTF part of this for me is that people will actually work for a 
penny a click or a nickel a result.  But without telling us how much you 
actually looked into it (like whether you read Amazon's pages, etc) it's 
hard to help explain.

 > I'm not really seeing what SQS would be useful for.

Think of a renderfarm for POV renders. How do you manage machines vs 
renders?  (The main problem with SQS in that sense is that you can't easily 
set the time-out for retries.)

Lots of this stuff is used by Amazon internally. I suspect order fulfillment 
is using SQS internally, and putting products up on the store (like, writing 
descriptions or entering metadata) is using something like Turk.

 > And I'm also wondering just how much all this stuff actually *costs*...

CPU hours are a bit expensive, but comparable to renting a shared server 
somewhere else (except you get a "dedicated" server). The rest is really 
cheap.  Hit aws.amazon.com for details.

-- 
   Darren New, San Diego CA, USA (PST)
   Why is there a chainsaw in DOOM?
   There aren't any trees on Mars.

Post a reply to this message

>> (Most especially, "Amazon Mechanical Turk". WTF?)
> 
> The only WTF part of this for me is that people will actually work for a 
> penny a click or a nickel a result.  But without telling us how much you 
> actually looked into it (like whether you read Amazon's pages, etc) it's 
> hard to help explain.

Why would people do "work" (especially such mundane work) for a nickle? 
Why would companies pay for results which are likely to be garbage 
anyway? How does Amazon make money out of this? Which numpty thought 
this sounded like a good idea? ...the questions continue...

>> I'm not really seeing what SQS would be useful for.
> 
> Think of a renderfarm for POV renders. How do you manage machines vs 
> renders?  (The main problem with SQS in that sense is that you can't 
> easily set the time-out for retries.)
> 
> Lots of this stuff is used by Amazon internally. I suspect order 
> fulfillment is using SQS internally, and putting products up on the 
> store (like, writing descriptions or entering metadata) is using 
> something like Turk.

OK, so... it's a message delivery system. Where do the messages come 
from? Where does it deliver them to? What are they for? Why would you 
pay money for this? etc.

>> And I'm also wondering just how much all this stuff actually *costs*...
> 
> CPU hours are a bit expensive, but comparable to renting a shared server 
> somewhere else (except you get a "dedicated" server). The rest is really 
> cheap.  Hit aws.amazon.com for details.

Yeah, I'm currently looking at this.

One wonders how a small online shop such as Amazon ends up with such 
vast computational resources that it can profitably hire out the spare 
capacity, but still...

Post a reply to this message

Invisible <voi### [at] devnull> wrote:
> Why would people do "work" (especially such mundane work) for a nickle?
> Why would companies pay for results which are likely to be garbage
> anyway? How does Amazon make money out of this? Which numpty thought
> this sounded like a good idea? ...the questions continue...

I think you've hit the nail on the head.

 - Ricky

Post a reply to this message

Invisible wrote:
> Why would people do "work" (especially such mundane work) for a nickle? 

I don't know. Most of it seems to not be from people in third-world 
a-nickel-is-a-lot-of-money countries either.

> Why would companies pay for results which are likely to be garbage 
> anyway? 

You don't pay till you check the results. Or you send the same job out three 
times and make sure all three answers come back similar. Which isn't hard 
when it only costs a nickel to start with.

> How does Amazon make money out of this? 

They take a cut.

> OK, so... it's a message delivery system. 

Yes.

> Where do the messages come 
> from? Where does it deliver them to? What are they for? Why would you 
> pay money for this? etc.

Read the APIs.

> One wonders how a small online shop such as Amazon ends up with such 
> vast computational resources that it can profitably hire out the spare 
> capacity, but still...

Small?  I suppose they started small.  Winter a year ago, they were 
fulfilling 43 orders per second the week before Christmas.

-- 
   Darren New, San Diego CA, USA (PST)
   Why is there a chainsaw in DOOM?
   There aren't any trees on Mars.

Post a reply to this message

Also... what precisely *is* a "web service" anyway? I can't find any 
useful answers online.

-- 
http://blog.orphi.me.uk/
http://www.zazzle.com/MathematicalOrchid*

Post a reply to this message

Orchid XP v8 wrote:
> Also... what precisely *is* a "web service" anyway? I can't find any 
> useful answers online.

It currently means a service provided to you over the internet. I.e., we'll 
run the program, you access it remotely.

Alternately, if you're talking about "Web Service", it could also mean using 
SOAP over HTTP to remotely invoke services at a remote machine. It's a bit 
of an overloaded term.

-- 
   Darren New, San Diego CA, USA (PST)
   Why is there a chainsaw in DOOM?
   There aren't any trees on Mars.

Post a reply to this message

Darren New wrote:

> It currently means a service provided to you over the internet. I.e., 
> we'll run the program, you access it remotely.
> 
> Alternately, if you're talking about "Web Service", it could also mean 
> using SOAP over HTTP to remotely invoke services at a remote machine. 
> It's a bit of an overloaded term.

Amazon appears to be talking about SOAP. It points to a W3 Schools 
tutorial, but it doesn't make any sense. (Neither does Wikipedia. But 
then, that's a horrid place to attempt to learn new technologies. It's 
reference material.)

-- 
http://blog.orphi.me.uk/
http://www.zazzle.com/MathematicalOrchid*

Post a reply to this message

Orchid XP v8 wrote:
> Amazon appears to be talking about SOAP. 

SOAP, conceptually, is pretty simple. You take the arguments to the 
procedure and wrap them up in XML. You put the function you're calling into 
the headers. You post the message (with appropriate crypto signatures if 
necessary) to the appropriate URL. You get back from the web server your 
answer, probably again wrapped up in XML.

WSDL is a way of specifying what arguments go where in the sending and 
returning XML.

Generally, you write the server end and have the IDE generate the WSDL file 
for it. You give the WSDL file to someone else, who hands it to *their* IDE, 
which writes the client stub. So it's basic remote procedure invocation, 
lathered up with a bunch of Web 2.0 buzzwords.

If you don't do the automation part, it's an awful technology. If you do the 
automation part, it's a convenient way of filling that niche. Most 
complaints about it are how hard it is to automate, or that it's the wrong 
way to do distributed computing.

In theory, SOAP supposedly runs over things other than HTTP. In practice, it 
doesn't, in part because they left out of SOAP some of the vital parts that 
HTTP provides (like cookies, for example).

-- 
   Darren New, San Diego CA, USA (PST)
   Why is there a chainsaw in DOOM?
   There aren't any trees on Mars.

Post a reply to this message

Darren New wrote:
> Orchid XP v8 wrote:
>> Amazon appears to be talking about SOAP. 
> 
> SOAP, conceptually, is pretty simple. You take the arguments to the 
> procedure and wrap them up in XML. You put the function you're calling 
> into the headers. You post the message (with appropriate crypto 
> signatures if necessary) to the appropriate URL. You get back from the 
> web server your answer, probably again wrapped up in XML.

So... you issue a command, and you receive a reply? And these are both 
formatted as XML?

Is the reply necessarily immediate?

If I'm understanding this correctly, that means that you can make Amazon 
start or stop a virtual server by sending a lump of XML to it over HTTP. 
Is that about right?

-- 
http://blog.orphi.me.uk/
http://www.zazzle.com/MathematicalOrchid*

Post a reply to this message

Orchid XP v8 wrote:
> So... you issue a command, and you receive a reply? And these are both 
> formatted as XML?

Yes. Think of it like a procedure call between machines.

> Is the reply necessarily immediate?

In theory, no. In practice, yes.

> If I'm understanding this correctly, that means that you can make Amazon 
> start or stop a virtual server by sending a lump of XML to it over HTTP. 
> Is that about right?

Exactly.  I have a library to do just that, only it uses the REST interface 
(which isn't really REST) to do so.

Also, http://console.aws.amazon.com/ lets you do it all from a web browser, 
or so they just announced. I haven't tried it.

-- 
   Darren New, San Diego CA, USA (PST)
   Why is there a chainsaw in DOOM?
   There aren't any trees on Mars.

Post a reply to this message

Darren New wrote:
>> Is the reply necessarily immediate?
> In theory, no. In practice, yes.

And by "immediate" I mean "returned over the same TCP connection the request 
went over."

SOAP designers will tell you how wonderful and flexible it all is. SOAP 
users don't do any of that because nobody has bothered to do any libraries 
to support SOAP over SMTP and things like that.

-- 
   Darren New, San Diego CA, USA (PST)
   Why is there a chainsaw in DOOM?
   There aren't any trees on Mars.

Post a reply to this message

>> So... you issue a command, and you receive a reply? And these are both 
>> formatted as XML?
> 
> Yes. Think of it like a procedure call between machines.

So it's a way of issuing server commands without needing a human to 
press the button?

>> Is the reply necessarily immediate?
> 
> In theory, no. In practice, yes.

Rather like the way HTTP responses are supposed to be "immediate"? 
(I.e., the browser hangs if it isn't.)

> Exactly.  I have a library to do just that, only it uses the REST 
> interface (which isn't really REST) to do so.

What's REST then?

> Also, http://console.aws.amazon.com/ lets you do it all from a web 
> browser, or so they just announced. I haven't tried it.

Yes, I noticed that. Currently only EC2 though. (?)

-- 
http://blog.orphi.me.uk/
http://www.zazzle.com/MathematicalOrchid*

Post a reply to this message

Orchid XP v8 wrote:
>> Yes. Think of it like a procedure call between machines.
> 
> So it's a way of issuing server commands without needing a human to 
> press the button?

Think of it like a procedure call between machines. :-)  I don't know what 
"button" you're talking about. "Buttons" are UI features. SOAP is a 
programming library mechanism. SOAP is a mechanism where you can write 
something like
    X = floob("Hello", 23)
and get back some data structure in X that was generated when "floob" and 
"Hello" and "23" were passed to some server somewhere. It doesn't matter 
whether that function call is invoked by pushing a button, a timer, a CLI, etc.

> Rather like the way HTTP responses are supposed to be "immediate"? 
> (I.e., the browser hangs if it isn't.)

Yes.

>> Exactly.  I have a library to do just that, only it uses the REST 
>> interface (which isn't really REST) to do so.
> 
> What's REST then?

Representational State Transfer.   A set of rules that almost nobody who 
claims to use REST actually follows.  But most everyone who calls their 
interfaces "REST" but which aren't means "just like SOAP, only without the 
ability to automate stub creation."  That is, you send us an XML request, 
and we return an XML answer, but you have to write all the code to parse out 
the results from the answer as a special case for each such service.

>> Also, http://console.aws.amazon.com/ lets you do it all from a web 
>> browser, or so they just announced. I haven't tried it.
> 
> Yes, I noticed that. Currently only EC2 though. (?)

As I said, I haven't looked.  The "s3fox" extension gives you access to 
Amazon S3 with a decent UI from firefox.

-- 
   Darren New, San Diego CA, USA (PST)
   Why is there a chainsaw in DOOM?
   There aren't any trees on Mars.

Post a reply to this message

> Think of it like a procedure call between machines. :-)

Normally prorams can't call functions in other programs. (Unless you 
count the operatin system.) And certainly not in programs runnin on 
another machine.

I guess that's what this technology brings to the table. (Whatever the 
technology is actually called...)

OOC, is there some better way rather than tunnelling over HTTP? Surely 
somebody has already come up with sane designs for this?

>> What's REST then?
> 
> Representational State Transfer.   A set of rules that almost nobody who 
> claims to use REST actually follows.

Sounds delightful.

-- 
http://blog.orphi.me.uk/
http://www.zazzle.com/MathematicalOrchid*

Post a reply to this message

Orchid XP v8 <voi### [at] devnull> wrote:
> > Think of it like a procedure call between machines. :-)

> Normally prorams can't call functions in other programs. (Unless you 
> count the operatin system.) And certainly not in programs runnin on 
> another machine.

  In many object-oriented languages, such as for example objective-C,
dynamic binding is done with a messaging system. In other words, when
you want to call the member function of an object (even if you don't
know the exact type of that object, but only have an abstract handle
to it), what happens internally is that a "message" is sent at runtime
to that object, and the object can accept it (in practice by the object's
type implementing the proper member function) or not. In practice this is
very much like a member function call (except that you don't necessarily
need to know that the object supports that function call).

  It's perfectly possible to send these messages between procedures, and
I assume even if these procedures are running in different computers. So
basically it ends up being one program calling the member function of an
object in another program (possibly running on a different computer).

  Here's some documentation about remote messaging in objective-C:
http://tinyurl.com/9339sk

-- 
                                                          - Warp

Post a reply to this message

Orchid XP v8 wrote:
> I guess that's what this technology brings to the table. (Whatever the 
> technology is actually called...)

Exactly. The general term is "remote procedure call", or "RPC".

> OOC, is there some better way rather than tunnelling over HTTP? Surely 
> somebody has already come up with sane designs for this?

Yes, and yes.  And, to answer your next question, yes.

>>> What's REST then?
>>
>> Representational State Transfer.   A set of rules that almost nobody 
>> who claims to use REST actually follows.
> 
> Sounds delightful.

To be a little less obtuse, it's a few guiding principles that make it 
possible to cache stuff and supposedly figure out interfaces on your own. 
Such principles include using a URL to identify every resource, and to have 
the body of resources include URLs to refer to other URLs; but the REST 
moniker requires much more to work "properly", including (for example) 
storing no "state" at the server.  In other words, HTML over HTTP, "done 
right" in some important senses, so that you can cache the results properly 
and etc.

People then take anything that encapsulates the request over HTTP in a way 
that assigns different URLs to resources to be "REST". That's a tiny and 
fairly useless part of it.

SOAP can't be "REST" because the URL identifies only the server and the 
function name and arguments go in the body of the request.

-- 
   Darren New, San Diego CA, USA (PST)
   Why is there a chainsaw in DOOM?
   There aren't any trees on Mars.

Post a reply to this message

>> And I'm also wondering just how much all this stuff actually *costs*...
> 
> CPU hours are a bit expensive, but comparable to renting a shared server 
> somewhere else (except you get a "dedicated" server). The rest is really 
> cheap.  Hit aws.amazon.com for details.

Well now... It seems that even if you store some absurd amount of data 
like 1GB (think how many years that would take to upload...) it costs 
less than 50p/month. Which is nothing. (I currently pay £5/month.)

On the other hand, 30 days of instance time on EC2 is almost 80$. 
Exchange rates vary, but this compares wildly unfavourably with my 
current hosts' demands of £15/month.

The verdict: S3 is very, very cheap. EC2 is absurdly expensive. (If 
you're a company, the reliability guarantees and lack of initial outlay 
could make sense. But for an individual, it's a no-brainer: don't do it!)

For about £200 you could *buy* a half-decent PC. If you leave it running 
for 1 year, it will have cost you [slightly more than] £200. If you run 
EC2 for 1 year, it will cost 964.15$, which I would suggest is slightly 
more than £200, whatever the exchange rate is.

Post a reply to this message

Invisible wrote:
> Well now... It seems that even if you store some absurd amount of data 

> like 1GB (think how many years that would take to upload...) it costs 
> less than 50p/month. Which is nothing. (I currently pay £5/month.)


Yep. I think my last bill was seventeen cents.

> On the other hand, 30 days of instance time on EC2 is almost 80$. 
> Exchange rates vary, but this compares wildly unfavourably with my 
> current hosts' demands of £15/month.

Is that for a machine where you can install your own OS and such? Or is t
hat 
for just a web host, where you're (for example) sharing an Apache server 

with others?

> The verdict: S3 is very, very cheap. EC2 is absurdly expensive. (If 
> you're a company, the reliability guarantees and lack of initial outlay
 
> could make sense. But for an individual, it's a no-brainer: don't do it
!)

It's also designed for you to rent it briefly, not for a long time. I.e.,
 if 
your needs aren't "elastic", it might not make sense. If you need six 
machines today, three tomorrow, and one next week, it makes more sense. I
f 
you want to rent a machine for three hours to do a render, it makes a lot
 of 
sense.

> For about £200 you could *buy* a half-decent PC. If you leave it r
unning 
> for 1 year, it will have cost you [slightly more than] £200.

Well, you would need the connectivity too, which I understand is pretty 
expensive where you are. :-)

-- 
   Darren New, San Diego CA, USA (PST)
   Why is there a chainsaw in DOOM?
   There aren't any trees on Mars.

Post a reply to this message

>> Well now... It seems that even if you store some absurd amount of data 
>> like 1GB (think how many years that would take to upload...) it costs 
>> less than 50p/month. Which is nothing. (I currently pay £5/month.)
> 
> Yep. I think my last bill was seventeen cents.

To quote the bitch from Friends, "ooo, that's *interesting*!"

>> On the other hand, 30 days of instance time on EC2 is almost 80$. 
>> Exchange rates vary, but this compares wildly unfavourably with my 
>> current hosts' demands of £15/month.
> 
> Is that for a machine where you can install your own OS and such? Or is 
> that for just a web host, where you're (for example) sharing an Apache 
> server with others?

That's for root SSH access to a Linux box that you can install whatever 
you want on. (Although presumably you can't change the OS - how would 
you talk to it while you're installing it?) Several flavours of Linux 
availbable.

It is a *virtual* server though - your stuff runs in a virtualisation 
system. (They claim to guarantee a certain minimum performance level 
though.)

If you want a *dedicated* server (i.e., a real physical box that's just 
for you) it's drastically more expensive - like £90/month or something.

>> The verdict: EC2 is absurdly expensive.
> 
> It's also designed for you to rent it briefly, not for a long time. 

As with any kind of renting, there will always be a point where renting 
becomes more expensive than buying. The question is only where that 
point is.

If you just want to *run* stuff, EC2 sets the tipping point really, 
really low. (E.g., the POV-Ray thing I'm currently trying to do will 
probably take at least a week, more likely a month. And I want to do a 
whole series of others afterwards. For the price EC2 want, I could 
probably buy a killer PC several times over.)

The advantace of EC2 is either
- You only want it for a little while.
- You want to do something that requires massive amounts of Internet 
bandwidth.

Buying a PC which out-performs EC2 in compute terms only isn't 
expensive. Beating the connectivity EC2 is likely to have would be... 
uh... "expensive".

> If you want to rent a machine for three hours to do a 
> render, it makes a lot of sense.

It does? Surely it would make more sense to just run it on my own PC for 
3 hours, while I go watch TV or something. Costs £0. :-P

>> For about £200 you could *buy* a half-decent PC. If you leave it 
>> running for 1 year, it will have cost you [slightly more than] £200.
> 
> Well, you would need the connectivity too, which I understand is pretty 
> expensive where you are. :-)

Maybe that's it: Rendering stuff doesn't require gigabit Internet access!

-- 
http://blog.orphi.me.uk/
http://www.zazzle.com/MathematicalOrchid*

Post a reply to this message

Orchid XP v8 wrote:
> That's for root SSH access to a Linux box that you can install whatever
 
> you want on. (Although presumably you can't change the OS - how would 
> you talk to it while you're installing it?) Several flavours of Linux 
> availbable.

Impressive.  (Cost-wise, I mean.)  I guess this is one of the great wins 
of 
virtualization.

> It is a *virtual* server though - your stuff runs in a virtualisation 
> system. (They claim to guarantee a certain minimum performance level 
> though.)

Same with Amazon, I expect. They talk about it as "equivalent to a ...." 
so 
I assume it's a shared image thingie.  Especially since you boot it by 
giving it an image split into a bunch of parts on S3, so there's got to b
e a 
pretty sophisticated boot loader going on there already.

> As with any kind of renting, there will always be a point where renting
 
> becomes more expensive than buying. The question is only where that 
> point is.

Agreed. It's also designed for you to (for example) spin up more servers 
for 
your shopping carts during the christmas rush.

> If you just want to *run* stuff, EC2 sets the tipping point really, 
> really low. (E.g., the POV-Ray thing I'm currently trying to do will 
> probably take at least a week, more likely a month. And I want to do a 

> whole series of others afterwards. For the price EC2 want, I could 
> probably buy a killer PC several times over.)

Sure.  I'm not trying to defend ECC's pricing, mind. :-)

> Buying a PC which out-performs EC2 in compute terms only isn't 
> expensive. Beating the connectivity EC2 is likely to have would be... 
> uh... "expensive".

You pay for the bandwidth, too. Not a lot, but again you have to measure 
how 
much you need and what it'll cost.

>> If you want to rent a machine for three hours to do a render, it makes
 
>> a lot of sense.
> 
> It does? Surely it would make more sense to just run it on my own PC fo
r 
> 3 hours, while I go watch TV or something. Costs £0. :-P

Assuming you didn't have one you could render on, of course.

-- 
   Darren New, San Diego CA, USA (PST)
   Why is there a chainsaw in DOOM?
   There aren't any trees on Mars.

Post a reply to this message

>> That's for root SSH access to a Linux box that you can install 
>> whatever you want on. (Although presumably you can't change the OS - 
>> how would you talk to it while you're installing it?) Several flavours 
>> of Linux availbable.
> 
> Impressive.  (Cost-wise, I mean.)  I guess this is one of the great wins 
> of virtualization.

Indeed yes.

>> It is a *virtual* server though - your stuff runs in a virtualisation 
>> system. (They claim to guarantee a certain minimum performance level 
>> though.)
> 
> Same with Amazon, I expect.

Yeah. Although the fact that it takes a minute or two to start up makes 
me think it's spending time trying to find a free machine to run it on. 
Maybe each physical machine runs up to X virtual machines or something? 
(I would imagine transfering a runnin image from place to place would be 
infeasieble, even in a datacenter.)

>> As with any kind of renting, there will always be a point where 
>> renting becomes more expensive than buying. The question is only where 
>> that point is.
> 
> Agreed. It's also designed for you to (for example) spin up more servers 
> for your shopping carts during the christmas rush.

Indeed. If you're a company and you're large enough to actually need 
*more than one* server at short notice, EC2 totally makes sense. For an 
individual user who just wants some POV-Ray time... EC2 does not, even 
remotely, make sense.

>> Buying a PC which out-performs EC2 in compute terms only isn't 
>> expensive. Beating the connectivity EC2 is likely to have would be... 
>> uh... "expensive".
> 
> You pay for the bandwidth, too. Not a lot, but again you have to measure 
> how much you need and what it'll cost.

You can easily buy a machine, put it in your kitchen, and leave it 
running all day. But you can't easily get seriously low latency to the 
Internet. EC2 gives you that. ;-)

>>> If you want to rent a machine for three hours to do a render, it 
>>> makes a lot of sense.
>>
>> It does? Surely it would make more sense to just run it on my own PC 
>> for 3 hours, while I go watch TV or something. Costs £0. :-P
> 
> Assuming you didn't have one you could render on, of course.

Well, you need to own a PC in order to design it in the first place. ;-) 
Plus, as I say, buying a PC works out vastly cheaper than using EC2 if 
all you want is computer time.

Now, how does GPGPU change this picture? ;-)

-- 
http://blog.orphi.me.uk/
http://www.zazzle.com/MathematicalOrchid*

Post a reply to this message

Orchid XP v8 wrote:
> Yeah. Although the fact that it takes a minute or two to start up makes 
> me think it's spending time trying to find a free machine to run it on. 

It might be that, altho I'd expect it's more a matter of transferring the 
multi-gigabyte OS from S3, decompressing and decrypting it, putting it on 
the disk as a virtual drive, and then booting it up.

> Maybe each physical machine runs up to X virtual machines or something? 

That would be my guess, especially since they guarantee a certain amount of 
performance, and they let you pick how many CPUs you'll have, and so on. :-)

> (I would imagine transfering a runnin image from place to place would be 
> infeasieble, even in a datacenter.)

Shouldn't be *too* hard, if all you're talking about is moving the disk 
image. It's not like you get to hook up special devices or anything. It 
would lose all the TCP connections, most likely.

> individual user who just wants some POV-Ray time... EC2 does not, even 
> remotely, make sense.

Unless you want to spin up 20 machines for one hour to do a trace, say. :-) 
Or if you don't have access to the sort of thing you're talking about.

> Well, you need to own a PC in order to design it in the first place. ;-) 

True. Depends of course on how studly your machine is. You started this by 
complaining you couldn't let the render run overnight, didn't you? :-)

-- 
   Darren New, San Diego CA, USA (PST)
   Why is there a chainsaw in DOOM?
   There aren't any trees on Mars.

Post a reply to this message

On Wed, 14 Jan 2009 22:11:27 +0000, Orchid XP v8 wrote:

> Yeah. Although the fact that it takes a minute or two to start up makes
> me think it's spending time trying to find a free machine to run it on.
> Maybe each physical machine runs up to X virtual machines or something?

There are systems out there that allow dynamic reallocation of physical 
resources for virtual systems, so for example if you have 5 VMs running 
on a single physical box and the performance spikes on one of the VMs, 
the others can be dynamically moved to another physical system even while 
running.

Some of the features of large-scale virtualization farms are quite 
impressive these days in being able to do that.  "Workload management" is 
the term I hear our marketing folks (yeah, Novell has marketing folks!) 
use.

The server farm that we use for our practical exams currently doesn't do 
that, but I'm talking with the implementation team about how we could 
dynamically scale if we needed to, and there's a solution available from 
a company we bought in 2008 (which of course should make it incredibly 
cheap for us to use internally).

Post a reply to this message