[besson3c]

VIrtualization is really the thing these days. To me, under most circumstances it just doesn't make sense to build servers and test environments that are not virtualized any more - virtualization just provides so much useful facility (and lower costs than physical hardware, of course). As we start to expand the number of cores that are shipped on CPUs, virtualization just grows stronger and more compelling.

I know most of you are not into running servers, but I'm wondering at what point these two worlds sort of intersect enough to force Apple's hand in making some changes? I mean, I'm sure many of you would love to have access to older versions of OS X for supporting people, taking screenshots, running certain applications, etc. I thought I read about something like this for the iPhone/iTouch OS? You can already run OS X Server in the Mac versions of VMWare and Parallels (I believe). Is this the extent of Apple's strategy?

Where these worlds intersect is in the bad rush to start virtualizing network services. I'm not exactly sure what Apple did to permit virtualizing OS X Server in Parallels and VMWare, but if they ever plan to extend this OS X Server support to other VM hosts for whatever strategic reason, why not OS X client as well at some point?

The obvious reason has to do with people not buying Macs so that they can run OS X/OS X Server as a VM guest on a PC. Why couldn't Apple update OS X to detect whether it is running on a VM hypervisor and cripple the experience just enough to make this not worth anybody's while? Perhaps they could cripple 3D acceleration, prevent going beyond a certain screen resolution, cut off all support or updates, or *something*? It seems like Apple could make it a lot of work for people to run VM based Hackintoshes enough to make this no more of a cannibal to sales than the current Hackintoshes are. To me, the upsides of allowing this outweigh these downsides. Nobody is going to want to put a crippled version of OS X Server into production on a PC, and your Grandma users are not going to be content with the kludge of running OS X Client in a VM on a Hackintosh. Have the geeks that have currently setup their own Hackintosh really cut into Apple's bottom line a whole lot?

Do you think that eventually Apple will be forced, or will willingly go down this path at some point?

[Laminar]

http://forums.macnn.com/alternative-operating-systems/

[Chuckit]

I don't think anybody Apple will ever utter the phrase, "Oh, we'll just make the user experience worse and that will solve the problem."

[besson3c]

Originally Posted by Chuckit 

I don't think anybody Apple will ever utter the phrase, "Oh, we'll just make the user experience worse and that will solve the problem."

Good point, but couldn't you argue that not having testing environments provides a poorer user experience to the end user, indirectly? The issue of whether running OS X on non Mac VM hosts is only a part of the equation. The other part is allowing OS X client to run in the VM hosts that currently support running OS X Server. If they do so, the question of whether whatever magic was provided to permit this will be extended to other vendors as well.

[goMac]

Originally Posted by besson3c 

The obvious reason has to do with people not buying Macs so that they can run OS X/OS X Server as a VM guest on a PC. Why couldn't Apple update OS X to detect whether it is running on a VM hypervisor and cripple the experience just enough to make this not worth anybody's while? Perhaps they could cripple 3D acceleration, prevent going beyond a certain screen resolution, cut off all support or updates, or *something*?

They already do. People already hack past.

When I first started OS X Intel development and I was peeved at Apple for making me rent a developer Mac, I ran OS X under VMWare on a Windows Pentium 4 box.

Now of course I run OS X Server under VMWare on my Mac Pro for development. I'm not sure what the big deal would be for OS X under a VM. For low usage, it's fine (like a simple DB or web server). But both Linux and FreeBSD are perfectly capable of that.

I use OS X server for it's high end features, which are frequently CPU and RAM intensive if you're going to use them in any real capacity, which makes it not good for virtualization.

If you really felt dirty about using Linux or FreeBSD, you could just load Darwin. It has no VM restrictions.

[Salty]

Thing is if they cripple it how do we know then that it won't affect the user experience if the testing environment isn't accurate? I can see them soon adding the ability to run OS X client in a VM so long as you're on a Mac.

[Dakar V]

Wasn't Boot Camp supposed to kill OS X?

[besson3c]

Originally Posted by goMac 

They already do. People already hack past.

When I first started OS X Intel development and I was peeved at Apple for making me rent a developer Mac, I ran OS X under VMWare on a Windows Pentium 4 box.

Now of course I run OS X Server under VMWare on my Mac Pro for development. I'm not sure what the big deal would be for OS X under a VM. For low usage, it's fine (like a simple DB or web server). But both Linux and FreeBSD are perfectly capable of that.

I use OS X server for it's high end features, which are frequently CPU and RAM intensive if you're going to use them in any real capacity, which makes it not good for virtualization.

If you really felt dirty about using Linux or FreeBSD, you could just load Darwin. It has no VM restrictions.

What about VMs for testing particular versions of OS X though, and not just hosting general common network services (i.e. web, email, etc.).

I would argue that whatever high end features you use with OS X Server make it precisely ideal for virtualization. For starters, it makes more economic sense for companies to invest their money in a single VM cluster. Secondly, being a VM would make it easy to bring up multiple nodes of OS X Server. If your VM host is VMWare ESX Server, it will do very sophisticated load balancing and dynamic allocation of resources for you. This would be perfect for building little farms of OS X Server nodes for whatever it was that you were doing.

[besson3c]

Originally Posted by Salty 

Thing is if they cripple it how do we know then that it won't affect the user experience if the testing environment isn't accurate? I can see them soon adding the ability to run OS X client in a VM so long as you're on a Mac.

It all depends on what they cripple.

It could be that Apple doesn't have to cripple anything though, just make it illegal to run OS X in a VM on a PC. Many companies won't risk being audited, even if there aren't actually any auditors to audit them. Otherwise, maybe they can require a serial number or something...

There seem to be a number of different ways this can be solved.

[goMac]

Originally Posted by besson3c 

What about VMs for testing particular versions of OS X though, and not just hosting general common network services (i.e. web, email, etc.).

This is why I keep an OS X Server VM around. I would like it if Apple had a developer edition of OS X that allowed virtualization, but in since they give developers a copy of OS X Server for free, it's not a huge issue.

Originally Posted by besson3c 

I would argue that whatever high end features you use with OS X Server make it precisely ideal for virtualization. For starters, it makes more economic sense for companies to invest their money in a single VM cluster. Secondly, being a VM would make it easy to bring up multiple nodes of OS X Server. If your VM host is VMWare ESX Server, it will do very sophisticated load balancing and dynamic allocation of resources for you. This would be perfect for building little farms of OS X Server nodes for whatever it was that you were doing.

I don't know why. What would OS X server bring to the table that would be suitable for virtualization? I mean, it has some nice admin tools, but the licenses are $1000 a pop. It's not economical, just as you don't usually see hosts doing Windows Server VM's, for the same reason.

[besson3c]

Originally Posted by goMac 

This is why I keep an OS X Server VM around. I would like it if Apple had a developer edition of OS X that allowed virtualization, but in since they give developers a copy of OS X Server for free, it's not a huge issue.

I'm talking about OS X client though... For proper testing and accurate screenshots, using OS X Server as a substitute for client is not adequate, no?

I don't know why. What would OS X server bring to the table that would be suitable for virtualization? I mean, it has some nice admin tools, but the licenses are $1000 a pop. It's not economical, just as you don't usually see hosts doing Windows Server VM's, for the same reason.

Well, there actually are a ton of Windows VMs around here because we have a site license.

How about Final Cut Server? Calendar Server?

Even for fairly mundane tasks like using Server to manage Mac based labs or running an LDAP directory it is still far cheaper to bring up a new VM guest on an existing VM cluster than it is to provide dedicated Mac hardware and keep it powered and racked. If that machine is sitting idle all the time, no big deal if your VM host is intelligent enough to allocate resources that aren't being used elsewhere. Idle VMs are going to be far, far cheaper than physical hardware.

You also have to look at it from a staffing perspective. If you need to hire Mac guys, taking racking and maintaining physical hardware out of their job descriptions allows you to adjust your staffing accordingly. It also makes physical relocation and sharing builds between sites far easier too. I'm not sure if tools exist for cloning a physical machine into a VM by using a .dmg file dump/clone of the physical machine, but obviously making VM images available is trivial.

What would be really cool is the day that the network can be counted on for fast enough bandwidth (or a direct fiber connection between sites is cheap enough) to do stuff like running off of VM images hosted on a remote site..

[goMac]

Originally Posted by besson3c 

I'm talking about OS X client though... For proper testing and accurate screenshots, using OS X Server as a substitute for client is not adequate, no?

I've only heard of a few instances when it is not adequate, usually dealing with the testing of other server software. Myself I have never run into any.

OS X Server does not have a different enough GUI that would make testing difficult.

Originally Posted by besson3c 

Well, there actually are a ton of Windows VMs around here because we have a site license.

How about Final Cut Server? Calendar Server?

You would not want to run Final Cut Server on a VM.

Calendar Server does not require a Mac server. The new 10.6 push server might, but I think that's supposed to be an open standard, isn't it?

Originally Posted by besson3c 

Even for fairly mundane tasks like using Server to manage Mac based labs or running an LDAP directory it is still far cheaper to bring up a new VM guest on an existing VM cluster than it is to provide dedicated Mac hardware and keep it powered and racked. If that machine is sitting idle all the time, no big deal if your VM host is intelligent enough to allocate resources that aren't being used elsewhere. Idle VMs are going to be far, far cheaper than physical hardware.

Why? Why not create multiple LDAP trees on one server? By adding VM's, you're just adding more overhead and making it inefficient.

Originally Posted by besson3c 

You also have to look at it from a staffing perspective. If you need to hire Mac guys, taking racking and maintaining physical hardware out of their job descriptions allows you to adjust your staffing accordingly. It also makes physical relocation and sharing builds between sites far easier too. I'm not sure if tools exist for cloning a physical machine into a VM by using a .dmg file dump/clone of the physical machine, but obviously making VM images available is trivial.

Yes they do, but I'm not seeing why VM's are required to consolidate services, especially on the Mac.

Originally Posted by besson3c 

What would be really cool is the day that the network can be counted on for fast enough bandwidth (or a direct fiber connection between sites is cheap enough) to do stuff like running off of VM images hosted on a remote site..

Sure but why? Storage is cheap.

[besson3c]

Originally Posted by goMac 

I've only heard of a few instances when it is not adequate, usually dealing with the testing of other server software. Myself I have never run into any.

OS X Server does not have a different enough GUI that would make testing difficult.

So the System Preference panes are exactly identical?

You would not want to run Final Cut Server on a VM.

Why not? I think you are equating a VM to equal slow. A VM is slower *on the same hardware*, but I think you are failing to realize that if you put a machine on monster VM cluster it might be faster than many physical machines providing the host is adequately powered. For starters, a SAN is definitely going to be faster than SATA drives in terms of I/O.

Calendar Server does not require a Mac server. The new 10.6 push server might, but I think that's supposed to be an open standard, isn't it?

No idea about 10.6 push server, but last I checked Calendar Server requires OS X and LDAP for delegation support.

Why? Why not create multiple LDAP trees on one server? By adding VM's, you're just adding more overhead and making it inefficient.

See above. VM does not equal slower necessarily.

Yes they do, but I'm not seeing why VM's are required to consolidate services, especially on the Mac.

I don't understand what you mean.

Sure but why? Storage is cheap.

What if you need to access the same data in real time from multiple sites? What if SATA drives won't do and you want to invest in a single SAN for your company?

[goMac]

Originally Posted by besson3c 

So the System Preference panes are exactly identical?

Would that matter? What would be stopping you from reverting the look back?

Originally Posted by besson3c 

Why not? I think you are equating a VM to equal slow. A VM is slower *on the same hardware*, but I think you are failing to realize that if you put a machine on monster VM cluster it might be faster than many physical machines providing the host is adequately powered. For starters, a SAN is definitely going to be faster than SATA drives in terms of I/O.

I'm not sure I get it. Something like Final Cut Server has built in user support, built in job management support, why would you even want to virtualize it? You're putting something that tries to manage CPU usage on top of something trying to manage CPU usage. It just sounds like a disaster in the making.

Originally Posted by besson3c 

No idea about 10.6 push server, but last I checked Calendar Server requires OS X and LDAP for delegation support.

It uses CalDav, which is an open standard, and implemented on other platforms.

Originally Posted by besson3c 

See above. VM does not equal slower necessarily.

How would it not? What you're saying is that a VM isn't slow as long as you add a bunch of extra hardware to make it not slow. But this will equal less hardware somehow...

Originally Posted by besson3c 

I don't understand what you mean.

What I'm saying is that if you have a network that needs multiple LDAP trees, put up a server with multiple LDAP trees. Why do you need to virtualize each LDAP tree?

Originally Posted by besson3c 

What if you need to access the same data in real time from multiple sites? What if SATA drives won't do and you want to invest in a single SAN for your company?

I guess... but why not just boot the server from the remote image?

[besson3c]

Originally Posted by goMac 

Would that matter? What would be stopping you from reverting the look back?

Have you been reading what I've been writing? What if you want to provide screenshots for support purposes?

I'm not sure I get it. Something like Final Cut Server has built in user support, built in job management support, why would you even want to virtualize it? You're putting something that tries to manage CPU usage on top of something trying to manage CPU usage. It just sounds like a disaster in the making.

No it's not. The Final Cut Server job management support works with the underlying hardware that is available to it. When resources are allocated dynamically by a VM host, the guests don't see "oh wait, I have more RAM available to me now", or "oh, I have a faster processor now", the resources are being allocated at a much lower level so that the guests aren't even aware of this. It takes some time to wrap your head around undoing everything you know about physical hardware and thinking virtually, but basically the guests are allocated a number of processors and memory as a sort of "you can use up to this" resource limit. There is no clock speed limitation. If you have a ridiculously high powered VM host and you have offered, say, 16 processors to the guest, if every other VM guest is idle you can count on your single working VM guest to absolutely haul ass. You can push the performance of a VM guest beyond anything you could get out of a single physical machine.

This is the beauty of a virtualized environment. I take it from the way you think about these sorts of problems that you haven't spent a whole lot of time in big enterprise lately (I don't mean that in a condescending way either). If you did, you'd see why everybody is going apeshit over this whole virtualization thing. We are in the process of moving data centers, and most groups here are working on leaving their old hardware behind in the old data center and virtualizing it. In a matter of years, it is feasible that entire data centers will be transformed from racks of various single physical machines to a single VM cluster.

This is a very big deal, the cost savings are quite substantial. If I were a jerk I would turn the tables on you and say that this is an occasion where *you* should do your homework, but fortunately I'm such a nice guy...

There is one area where cost savings are not substantial, and this is a problem that has yet to be solved. That is with storage... VM clusters are designed around SANs traditionally, and therefore adding disks to a SAN is much more costly than simply adding another SCSI or SATA drive to a RAID array. Therefore, if you don't really need fast I/O virtualizing a service and using expensive disk is not very cost effective. Still, even with this problem running a VM cluster *still* works out to be faster than running a bunch of physical machines if you do the math.

What could be used to bridge this gap is iSCSI. iSCSI will allow you to create mass storage based on cheap disks. ZFS will also be an important part of this equation too, although a lot of people are still shying away from it on account of it still being young and slightly buggy (if you Google for ZFS related problems you will find them). Even without ZFS though, iSCSI is a much cheaper option than a SAN. Slower, but effective for services that don't require fast I/O. Perhaps some VM clusters will start to incorporate a hybrid storage system - SAN for guests that need fast I/O, iSCSI for guests that just need a disk.

It uses CalDav, which is an open standard, and implemented on other platforms.

I'm fully aware of that, but that doesn't change anything I've said. Calendar Server uses XML, LDAP, or Kerberos as a means of authenticating users, and last I checked calendar delegation required LDAP/OS X Server. The Calendar ACLs are managed by the server, they are not a part of the CalDav protocol.

How would it not? What you're saying is that a VM isn't slow as long as you add a bunch of extra hardware to make it not slow. But this will equal less hardware somehow...

Yes, it does. VM clusters make efficient use of idle resources - that's where the savings enter the picture.

[goMac]

Originally Posted by besson3c 

Have you been reading what I've been writing? What if you want to provide screenshots for support purposes?

Then you install back in whatever preference pane it is you want to take a screenshot of. It's hardly the end of the world.

Originally Posted by besson3c 

No it's not. The Final Cut Server job management support works with the underlying hardware that is available to it. When resources are allocated dynamically by a VM host, the guests don't see "oh wait, I have more RAM available to me now", or "oh, I have a faster processor now", the resources are being allocated at a much lower level so that the guests aren't even aware of this. It takes some time to wrap your head around undoing everything you know about physical hardware and thinking virtually, but basically the guests are allocated a number of processors and memory as a sort of "you can use up to this" resource limit. There is no clock speed limitation. If you have a ridiculously high powered VM host and you have offered, say, 16 processors to the guest, if every other VM guest is idle you can count on your single working VM guest to absolutely haul ass. You can push the performance of a VM guest beyond anything you could get out of a single physical machine.

This is the beauty of a virtualized environment. I take it from the way you think about these sorts of problems that you haven't spent a whole lot of time in big enterprise lately (I don't mean that in a condescending way either). If you did, you'd see why everybody is going apeshit over this whole virtualization thing. We are in the process of moving data centers, and most groups here are working on leaving their old hardware behind in the old data center and virtualizing it. In a matter of years, it is feasible that entire data centers will be transformed from racks of various single physical machines to a single VM cluster.

What I'm saying is why. Why would you virtualize that? What is the point? What does virtualization add? Do we virtualize an operating system for each user on our machines?

Final Cut Server needs lots of cores. It needs lots of speed. Virtualization causes an immediate speed hit. And, if you want to deallocate or allocate resources on the fly, it'll do that. So what's the point of virtualizing it?

Heck, if you want to dynamically enable and disable certain functions or different servers, just use Netboot images. You can reallocate resources with no CPU loss.

Originally Posted by besson3c 

This is a very big deal, the cost savings are quite substantial. If I were a jerk I would turn the tables on you and say that this is an occasion where *you* should do your homework, but fortunately I'm such a nice guy...

There is one area where cost savings are not substantial, and this is a problem that has yet to be solved. That is with storage... VM clusters are designed around SANs traditionally, and therefore adding disks to a SAN is much more costly than simply adding another SCSI or SATA drive to a RAID array. Therefore, if you don't really need fast I/O virtualizing a service and using expensive disk is not very cost effective. Still, even with this problem running a VM cluster *still* works out to be faster than running a bunch of physical machines if you do the math.

I'm not saying you have to add disks to a SAN. I'm saying you can take advantage of a SAN better without virtualization. Even if the SAN is on the other side of the world.

Originally Posted by besson3c 

What could be used to bridge this gap is iSCSI. iSCSI will allow you to create mass storage based on cheap disks. ZFS will also be an important part of this equation too, although a lot of people are still shying away from it on account of it still being young and slightly buggy (if you Google for ZFS related problems you will find them). Even without ZFS though, iSCSI is a much cheaper option than a SAN. Slower, but effective for services that don't require fast I/O. Perhaps some VM clusters will start to incorporate a hybrid storage system - SAN for guests that need fast I/O, iSCSI for guests that just need a disk.

Thank you for the lesson on file systems, but this still doesn't have anything to do with what I'm saying. The usage of a SAN does not require virtualization.

Originally Posted by besson3c 

I'm fully aware of that, but that doesn't change anything I've said. Calendar Server uses XML, LDAP, or Kerberos as a means of authenticating users, and last I checked calendar delegation required LDAP/OS X Server. The Calendar ACLs are managed by the server, they are not a part of the format.

I don't think Caldav makes any assumption on how users are authenticated. That's how it works on OS X Server, yes, but other servers implement this differently...

Originally Posted by besson3c 

Yes, it does. VM clusters make efficient use of idle resources - that's where the savings enter the picture.

In some specific cases, yes. If your a web host for multiple clients who can't have touching OS's, then yes, it does make sense. For most other cases, I still don't think it makes sense. If you're running a Final Cut Server rack, no, it makes little sense at all. Final Cut Server has built in load balancing across nodes. If you're a serious organization running servers, usually your servers are always pegged anyway. Even if they're not, you at least need to have the hardware in place to handle the maximum possible load without any conflicts for grabbing the CPU. I can't image, say, running any sort of serious file server nodes off of a VM for a SAN. I worked for an organization that did this sort of thing, and there were some days where those nodes were all pegged. This is not something you'd want to virtualize. Even if you had a day where those nodes weren't pegged, you've got to be ready for the next day where they could be. And you don't want it setting on the same hardware as your mail server when they both get pegged.

[Laminar]

I thought boring threads were supposed to be kept out of the Lounge.

[besson3c]

Originally Posted by goMac 

Then you install back in whatever preference pane it is you want to take a screenshot of. It's hardly the end of the world.

So what, find the install media of an older version of an OS, find the pref pane, and install it onto whatever OS version you are using? How would you even know that there was a difference in the way the screens look without being able to actually sit down in front of the OS version in question? Even if you did this, all of the widgets surrounding the pref pane, and possibly the widgets in the pref pane itself would look different aesthetically, no?

You have this very strange habit of really clinging to a particular solution in your head and not letting it go. Sure what you are describing would work, but having access to these older OSes in a VM guest would be much more convenient for places that need to provide a lot of support documentation. I'm not really sure why you feel the need to debate this, to be honest... Do you have a difficult time admitting to being wrong? I'm not gloating, I'm wrong all the time, just asking...

What I'm saying is why. Why would you virtualize that? What is the point? What does virtualization add? Do we virtualize an operating system for each user on our machines?

Final Cut Server needs lots of cores. It needs lots of speed. Virtualization causes an immediate speed hit. And, if you want to deallocate or allocate resources on the fly, it'll do that. So what's the point of virtualizing it?

Heck, if you want to dynamically enable and disable certain functions or different servers, just use Netboot images. You can reallocate resources with no CPU loss.

I've already answered that. I'll reiterate:

1) You don't have to invest in dedicated hardware for the next generation of Final Cut Server infrastructure and can consolidate on a single VM cluster. If need be you can dedicate physical machines in the cluster to certain guests

2) You can make the infrastructure as fast or as slow as you want - there is no real ceiling in terms of performance

3) Redundancy in having multiple nodes and as many nodes as your physical hardware allows. If you want to expand how many nodes you can offer or how fast they can go, you just add hardware to the underlying cluster, clone your VM, assign it an IP, and put it into your pool.

4) Consolidated high speed storage for the company

Again, it takes time to wrap your head around everything being virtualized, I went through the same process, so none of this is intended to sound condescending...

I'm not saying you have to add disks to a SAN. I'm saying you can take advantage of a SAN better without virtualization. Even if the SAN is on the other side of the world.

Thank you for the lesson on file systems, but this still doesn't have anything to do with what I'm saying. The usage of a SAN does not require virtualization.

I'm not sure where you are coming from here, but what I wrote was just me rambling about a cost related problem with VM clusters in general, it wasn't specific to what we were talking about.

I don't think Caldav makes any assumption on how users are authenticated. That's how it works on OS X Server, yes, but other servers implement this differently...

Right, so like I said, it is up to the server to provide calendar ACLs, and last I checked Calendar Server did not support delegation on non OS X/LDAP setups.

In some specific cases, yes. If your a web host for multiple clients who can't have touching OS's, then yes, it does make sense. For most other cases, I still don't think it makes sense. If you're running a Final Cut Server rack, no, it makes little sense at all. Final Cut Server has built in load balancing across nodes. If you're a serious organization running servers, usually your servers are always pegged anyway. Even if they're not, you at least need to have the hardware in place to handle the maximum possible load without any conflicts for grabbing the CPU. I can't image, say, running any sort of serious file server nodes off of a VM for a SAN. I worked for an organization that did this sort of thing, and there were some days where those nodes were all pegged. This is not something you'd want to virtualize. Even if you had a day where those nodes weren't pegged, you've got to be ready for the next day where they could be. And you don't want it setting on the same hardware as your mail server when they both get pegged.

These problems are all solved with proper design of your VM infrastructure. Like I said, if you want to devote physical machines to certain guests, you can. If your I/O is being pegged, you need more of it (you can get SANs with absolutely ridiculous amounts of I/O). If you have reached what your SAN will offer, perhaps you need to start a new cluster on a new SAN.

Where you are wrong is thinking that there are somehow inherent unsolvable problems with a VM infrastructure that would make scaling of resource intensive systems impossible. It is true that there are more bottlenecks to think of in thinking about both the physical and virtual layers, but these problems can be solved, and people are getting better at solving them.

What VM host did the organization you worked for use that managed to saturate its I/O? What were they using for storage?

[turtle777]

Originally Posted by Laminar 

I thought boring threads were supposed to be kept out of the Lounge.

Seconded.

WTF does Besson keep posting highly technical OS threads in the Lounge ?

-t

[besson3c]

Originally Posted by turtle777 

Seconded.

WTF does Besson keep posting highly technical OS threads in the Lounge ?

-t

I wish I understood the obsession some of you have over the categorization of threads, it seems borderline neurotic to me at times.

[mattyb]

Not sure about the OS X space, but my humble experience of large companies was that they bought (for example) : two large mainframes in a sysplex (a cluster) and had prod, dev, test, validation environments as seperate LPARs (logical partitions) on those two machines. For the UNIX platform it was three machines, prod, dev/validation and test. If prod failed, then it would failover to one of the the other machines giving a reduced service until the prod machine was back up. This was the same for the mainframe setup.

Virtualization got popular because idiots calling themselves Architects decided that for each project using Windows (SQL Server or IIS) they needed a box. Pretty soon the Windows servers were taking up more space than the mainframe + UNIX + network equipment. Fook !! Lets virtualize !! Blah blah blah, marketing, stupid decisions, a bit of golf and a few months later there's a big push for virtualization.

I can see the need for virtualization for developers, but with the world going web based, the average user just isn't going to need it. The other downside of virtualization (that I came across) is that the licensing for products becomes impossible. Example : Oracle on a Virtual Logical Partition (VLPAR). The physical machine has 8 dual core CPUs. Each VLPAR has 1.5 CPUs allocated to it, yet you tell the VLPAR that the 1.5 is equal to two 'virtual' CPUs (Oracle asks how many CPUs you have on the machine when you install it - there is no choice for 1.5 CPUs ). Even the sales guys from Oracle don't know how to bill us. Virtualization, at least in the datacentre, is to me a sign of IT's failure. Basically it reeks of mis-management.

[Railroader]

Originally Posted by besson3c 

I wish I understood the obsession some of you have over the categorization of threads, it seems borderline neurotic to me at times.

I don't think you know how to lounge properly.

Shouldn't this thread be in the Apple OS forum?

[Chuckit]

Originally Posted by mattyb 

I can see the need for virtualization for developers, but with the world going web based, the average user just isn't going to need it.

I'd think that makes it more useful, not less. For instance, it seems like more and more hosts are offering virtualized servers these days. And the kind of massive power that computers now offer is overkill for most purposes, so if someone could find a way to split the family computer into five and effectively give everybody their own $60 computer, that would be a pretty good deal.

[besson3c]

Originally Posted by mattyb 

Not sure about the OS X space, but my humble experience of large companies was that they bought (for example) : two large mainframes in a sysplex (a cluster) and had prod, dev, test, validation environments as seperate LPARs (logical partitions) on those two machines. For the UNIX platform it was three machines, prod, dev/validation and test. If prod failed, then it would failover to one of the the other machines giving a reduced service until the prod machine was back up. This was the same for the mainframe setup.

Virtualization got popular because idiots calling themselves Architects decided that for each project using Windows (SQL Server or IIS) they needed a box. Pretty soon the Windows servers were taking up more space than the mainframe + UNIX + network equipment. Fook !! Lets virtualize !! Blah blah blah, marketing, stupid decisions, a bit of golf and a few months later there's a big push for virtualization.

I can see the need for virtualization for developers, but with the world going web based, the average user just isn't going to need it. The other downside of virtualization (that I came across) is that the licensing for products becomes impossible. Example : Oracle on a Virtual Logical Partition (VLPAR). The physical machine has 8 dual core CPUs. Each VLPAR has 1.5 CPUs allocated to it, yet you tell the VLPAR that the 1.5 is equal to two 'virtual' CPUs (Oracle asks how many CPUs you have on the machine when you install it - there is no choice for 1.5 CPUs ). Even the sales guys from Oracle don't know how to bill us. Virtualization, at least in the datacentre, is to me a sign of IT's failure. Basically it reeks of mis-management.

That's a very limited view, no offense.

What about services that rely on pooling and load balancing, i.e. require more than one physical machine to serve requests at any given time? What about companies with many different departments who can not agree upon what should be installed on the mainframes? What about sharing resources across all users on the mainframe? What about different projects requiring different versions of the operating system?

I could go on, but it's not as simple as every company only needing a few mainframes and virtualizing being invented to solve what could be solved as you describe.

[besson3c]

Originally Posted by Chuckit 

I'd think that makes it more useful, not less. For instance, it seems like more and more hosts are offering virtualized servers these days. And the kind of massive power that computers now offer is overkill for most purposes, so if someone could find a way to split the family computer into five and effectively give everybody their own $60 computer, that would be a pretty good deal.

Isn't it kind of ironic that we can begin to entertain going back to the mainframe + thin client model, where computing began in the first place?

Really though, this makes more and more sense every day. I think the first place it makes sense is replacing office settings and all of the various workstations devoted to word processing, spreadsheeting, emailing, and data entry to thin clients. In a room of, say, 50 of these sort of machines, there are a *lot* of computing resources being wasted. If each machine cost the company $500, that's $25,000 spent right there. Even at $300/pop, that's $15,000. For such a light load, this can be easily replaced by the underlying hardware running as a VM host.

Also consider that the initial computer sticker price is not all there is to factor in. There is the cost of power and repairs as well, the cost of disposal, etc. Granted thin clients will require some power, and granted some hardware will fail on a VM host, but if you can cut back on this a little bit there is the potential for savings there as well.

Besson3c's prediction: this is where things are going

[Doofy]

Of course, if every major company on the planet replaced their PCs with a fat server + thin clients model, that'd mean less computers sold. Which means less economy of scale for the computer (and parts) manufacturers. Which means that that Mac Pro that you require for your artistic endeavours will end up costing the same as a house.

No thanks.

[turtle777]

Originally Posted by besson3c 

I wish I understood the obsession some of you have over the categorization of threads, it seems borderline neurotic to me at times.

If you weren't the attention whore that you are, you wouldn't post the threads in the Lounge, but where they really belong. This a OS discussion, plain and simple.

But you always want maximum attention, so everything just gets dumped in the Lounge. Really awesome.

-t

[besson3c]

Originally Posted by Doofy 

Of course, if every major company on the planet replaced their PCs with a fat server + thin clients model, that'd mean less computers sold. Which means less economy of scale for the computer (and parts) manufacturers. Which means that that Mac Pro that you require for your artistic endeavours will end up costing the same as a house.

No thanks.

That's only if that change happens overnight.

More consumer options and increased competition are usually a good thing, and there may be a day where a thin client makes more technological sense in the home and costs less (especially when/if the average number of computers in a house is more than one). I think that homes are the last place this model will work, but it is feasible that you'll be able to buy your own VM host appliance for the house one day. That is still a lot of households to do business with and the number of people using computers overall is increasing everyday, so I wouldn't worry to much about these shifts affecting you in a negative way.

[Chuckit]

Originally Posted by Doofy 

Of course, if every major company on the planet replaced their PCs with a fat server + thin clients model, that'd mean less computers sold. Which means less economy of scale for the computer (and parts) manufacturers. Which means that that Mac Pro that you require for your artistic endeavours will end up costing the same as a house.

No thanks.

Because everybody is buying Mac Pros for their employees to surf the Web and use email? If anything, this might actually mean more high-end computers like the Mac Pro are sold in place of the low-end crap, since then all the power could actually be put to efficient use.

[besson3c]

Originally Posted by turtle777 

If you weren't the attention whore that you are, you wouldn't post the threads in the Lounge, but where they really belong. This a OS discussion, plain and simple.

But you always want maximum attention, so everything just gets dumped in the Lounge. Really awesome.

-t

You still haven't made a case as to why this actually matters, but just to restore some order to your world I will entertain this... The OS X forum is for talking exclusively about OS X, this thread is about Apple's decisions as a company. The alternative OS forum is for talking about non-Apple and/or non-Mac stuff, this relates to Apple. So, by process of elimination that leaves us with the Lounge.

I don't really wish to debate this assessment, people have already complained about this thread being "boring", but hopefully you will be comforted to know that I just didn't randomly put this in the Lounge without any thought.

[Doofy]

Originally Posted by Chuckit 

Because everybody is buying Mac Pros for their employees to surf the Web and use email?

No. Because the crap inside the Mac Pro is inside those office PCs. Since it's in those PCs, the manufacturer can bulk produce and thus we get it cheaper for our machines.

[mattyb]

Originally Posted by Chuckit 

I'd think that makes it more useful, not less. For instance, it seems like more and more hosts are offering virtualized servers these days. And the kind of massive power that computers now offer is overkill for most purposes, so if someone could find a way to split the family computer into five and effectively give everybody their own $60 computer, that would be a pretty good deal.

I didn't explain myself very well. I can see virtualization as a bonus for companies offering services, but not for the average Joe - besides the developers that is. I'm no fortune teller but I reckon that we'll more likely see the ultra-thin-client model permanently connected to the internet with just a browser. The average Joe won't have a computer (not one that he administers) in his house.

Originally Posted by besson3c 

That's a very limited view, no offense.

What about services that rely on pooling and load balancing, i.e. require more than one physical machine to serve requests at any given time? What about companies with many different departments who can not agree upon what should be installed on the mainframes? What about sharing resources across all users on the mainframe? What about different projects requiring different versions of the operating system?

I could go on, but it's not as simple as every company only needing a few mainframes and virtualizing being invented to solve what could be solved as you describe.

Again, I didn't explain myself very well.

The sysplex is a cluster, so load balancing is 'built in' so to speak. In fact we had the mainframe on three different networks, one external, one internal, and one test. Yes, the mainframe can do this.

In my experience IT depts tend to offer services to other departments. For example a database with an app server, 100 users simultaneously, 200G initially, 15% growth etc etc. Economies of scale in ordering licenses, machines etc make centralization in an IT dept more cost effective than each dept doing their own thing. How many companies have you come across with several email systems, CRM, and ERP for example. Yes IT does tend to offer different databases, different OSes, but this has more to do with the company's history, Project Management bias and the flavour of the month, etc. than to do with different depts wanting different systems. When for example a HR dept decides that SAP's payroll module is needed, then the IT dept is more than likely going to be heavily involved in the underlying OS, database, backup, network config etc. As for sharing resources in the mainframe, well the small mainframe that I worked on (approx 900 MIPS) managed to run factories in France, Spain and the US. Seven factories that made trucks, engines, cabins, etc etc. Plus the spare parts system, finance and for a while an aging email system before it was replaced with Exchange. We also had several datawarehouses, developers in India plus websites on the beast.

We did at one point offer different versions of the OS, but it was decided that managing different version of different operating systems was not cost effective. In the IT depts that I worked in there were several operating systems already (z/OS, AS400, VMS, various UNIXes and Windows). Supporting this many operating systems is difficult enough, when you have to support different versions of different operating systems, IT management becomes unmanageable. This is IMHO good management, use the same tools, the same procedures, the same standards. I'm not saying that we were really good at our jobs and we were the best (far from it) but if you want to provide services and ensure that your costs are under control, you cannot just let anyone run anything at any time.

Just to clarify my response to Chuckit, there was and still is a trend in the mainframe world where instead of using z/OS on a Logical partition(LPAR), you install z/VM. You can then have Linux 'instances' running on top of z/VM. I can remember reading that some smart arse started thousands of Linux instances on one LPAR. This is an old story but apparently IBM have been increasing mainframe MIPS sales for the past few years by pushing this very ability.

[besson3c]

mattyb: the model you are describing doesn't quite apply to the way we are thinking of virtualization...

Instead of providing everybody the choice of what OS and system they want to run and trying to support them somehow, this is about providing *resources* for departments to *assume responsibility* over whatever it is they want to run. Whatever you want to run in your VM is up to you, and really not of huge concern to the VM admins whose only concern is to carve up the total amount of resources available into smaller pieces to divvy out to the users. So, there is really no software consolidation considerations here...

[mattyb]

OK, you could probably give depts a VM for their own use, but even stuff like this needs to be locked down somewhat. Some idiot that I worked with chose the same IP address as the primary DNS when re-imaging his own machine. This guy was a technical guy as well, so he had full admin access to his PC. You can imagine the fun that was had by all and sundry.

I'll stop ranting and chill out, its not been a fun Friday afternoon at work.

[besson3c]

No worries! AFAIK, big companies generally have network operations and DNS admins to allocate IPs and make sure that people don't just take IPs that are in use, so I'm not sure what sort of locking down you had in mind?