Jordi Espasa Clofent
2007-Jun-23 11:05 UTC
[Xen-users] XEN and Windows Guests in critical environment (hospital)
Hi all, One friend of mine are thinking about how to implement virtualization in their critical job environment (a hospital). The main "problem" is there are a lot of medical application builded in .NET tecnology; so, I view three possible options: 1. Win server with VMware and win guests (IIS to support .NET). 2. UNIX/Linux server with XEN (or XenEnterprise) and win guests 3. UNIX/Linux server with XEN (or XenEnterprise) and UNIX/Linux guests (Apache with mod_mono to support .NET) I think the next about each one option: 1. The current preference because stability and windows compatibility. 2. My personal preference, but I''m not sure about the performance of win guests on XEN Unix based system. 3. The "ideal" economical solution, but probably the more insecure in terms of stability. Let me to repeat: it''s a CRITICAL environment and will be not any error-edge. Any argued reasoning will be welcomed. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
James Harper
2007-Jun-23 12:15 UTC
RE: [Xen-users] XEN and Windows Guests in critical environment(hospital)
> Hi all, > > One friend of mine are thinking about how to implement virtualizationin> their critical job environment (a hospital). > The main "problem" is there are a lot of medical application buildedin> .NET tecnology; so, I view three possible options: > > 1. Win server with VMware and win guests (IIS to support .NET). > 2. UNIX/Linux server with XEN (or XenEnterprise) and win guests > 3. UNIX/Linux server with XEN (or XenEnterprise) and UNIX/Linuxguests> (Apache with mod_mono to support .NET) > > I think the next about each one option: > > 1. The current preference because stability and windows compatibility. > 2. My personal preference, but I''m not sure about the performance ofwin> guests on XEN Unix based system. > 3. The "ideal" economical solution, but probably the more insecure in > terms of stability. > > Let me to repeat: it''s a CRITICAL environment and will be not any > error-edge. > > Any argued reasoning will be welcomed.In theory, all of the above options you mentioned add a layer of complexity to the problem and therefore potentially reduce reliability. I''m guessing, but there is a good chance that the critical applications you speak of have been designed and tested in an un-virtualised environment running on a Microsoft operating system. (I''d question running critical applications in a Microsoft environment at all, but that''s another argument :) By taking the applications out of the environment they have been designed and tested in, you are almost certainly moving into an environment that the company that designed the software can''t or won''t support. What if you have a problem? Who are you going to call? You need someone that can drop everything they are doing and work on the problem. If you are running VMWare+Windows+.NET or Xen+Windows+.NET, and an application starts crashing, what do you do? The problem could be related to the hardware, related to the virtualisation layer (VMWare/Xen), related to windows, related to .NET, related to the actual application, or some combination of all of them. Your support agent needs to be able to work on the problem as a whole... I find it hard enough diagnosing obscure problems when virtualisation isn''t involved, and the last thing you want is one of the vendors throwing their hands in the air and saying ''Xen/VMWare''??? We don''t support that!!! My argument therefore is that if these applications are as critical as you say they are, give them exactly the environment they were designed for, or at least get the approval of the software developers and testers before you do anything. In fact I would have thought that critical applications like this would be supplied essentially as a ''black box'' solution, and if anything goes wrong the supplier deals with it... James _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Ligesh
2007-Jun-23 13:19 UTC
[Xen-users] Re: XEN and Windows Guests in critical environment(hospital)
On Sat, Jun 23, 2007 at 10:15:25PM +1000, James Harper wrote:> problems when virtualisation isn''t involved, and the last thing you want > is one of the vendors throwing their hands in the air and saying > ''Xen/VMWare''??? We don''t support that!!! >Why would you want to tell the application software folks that you are running it inside Xen? How can they even find out? Just let them login to the windows machine like they normally do, and even if they are experts on virtualization, I don''t think there is an easy way to determine that. Even the windows os itself has no idea it is running in a virtualized environment. In fact, that''s the entire point of virtualization--virtualize nothing or at the most, the kernel, and the apps all work transparently, since otherwise the entire thing becomes pointless. As for virtualization itself, you need to check the stability of Xen itself, which has only been tagged stable recently. Virtualization makes it easier to manage, and overall I would say increases the reliability, simply on account of being easier to keep track of, and also take backups, migrate in case of hardware failure etc. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Nico Kadel-Garcia
2007-Jun-23 17:36 UTC
Re: [Xen-users] Re: XEN and Windows Guests in critical environment(hospital)
Ligesh wrote:> On Sat, Jun 23, 2007 at 10:15:25PM +1000, James Harper wrote: > >> problems when virtualisation isn''t involved, and the last thing you want >> is one of the vendors throwing their hands in the air and saying >> ''Xen/VMWare''??? We don''t support that!!! >> >> > > Why would you want to tell the application software folks that you are running it inside Xen? How can they even find out? Just let them login to the windows machine like they normally do, and even if they are experts on virtualization, I don''t think there is an easy way to determine that. Even the windows os itself has no idea it is running in a virtualized environment. In fact, that''s the entire point of virtualization--virtualize nothing or at the most, the kernel, and the apps all work transparently, since otherwise the entire thing becomes pointless. > > As for virtualization itself, you need to check the stability of Xen itself, which has only been tagged stable recently. Virtualization makes it easier to manage, and overall I would say increases the reliability, simply on account of being easier to keep track of, and also take backups, migrate in case of hardware failure etc. >Well, sending them a "sysreport" output might be a good hint, because it will name the Xen kernel. In fact, why would you want to lie about it? Not telling a vendor your actual setup is begging for them to suffer. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Ligesh
2007-Jun-23 20:16 UTC
[Xen-users] Re: XEN and Windows Guests in critical environment(hospital)
On Sat, Jun 23, 2007 at 06:36:53PM +0100, Nico Kadel-Garcia wrote:> Ligesh wrote: > > As for virtualization itself, you need to check the stability of Xen > > itself, which has only been tagged stable recently. Virtualization makes > > it easier to manage, and overall I would say increases the reliability, > > simply on account of being easier to keep track of, and also take > > backups, migrate in case of hardware failure etc. > > > Well, sending them a "sysreport" output might be a good hint, because it > will name the Xen kernel. In fact, why would you want to lie about it? > Not telling a vendor your actual setup is begging for them to suffer. >NO you need not. Do you tell your application vendor what CPU you are running on? Or what motherboard you are using? Xen is transparent. That''s the whole point; otherwise I don''t think it is useful. Xen should be treated as yet another hardware, and I don''t think it is necessary to tell anyone about it, unless of course, you are running a software that explicitly deals with hardware, in which situation you should probably not run Xen, unless you know what you are doing. For the app vendor, it is irrelevant if you are running it inside xen or on baremetal. Even for windows it doesn''t matter, so for something that sits so high up in the application chain, why would you want to complicate and confuse the easily confused people by bringing in information about a software about which they probably haven''t heard about? _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Nico Kadel-Garcia
2007-Jun-23 21:44 UTC
[Xen-users] Re: XEN and Windows Guests in critical environment(hospital)
Ligesh wrote:> On Sat, Jun 23, 2007 at 06:36:53PM +0100, Nico Kadel-Garcia wrote: > >> Ligesh wrote: >> >>> As for virtualization itself, you need to check the stability of Xen >>> itself, which has only been tagged stable recently. Virtualization makes >>> it easier to manage, and overall I would say increases the reliability, >>> simply on account of being easier to keep track of, and also take >>> backups, migrate in case of hardware failure etc. >>> >>> >> Well, sending them a "sysreport" output might be a good hint, because it >> will name the Xen kernel. In fact, why would you want to lie about it? >> Not telling a vendor your actual setup is begging for them to suffer. >> >> > > NO you need not. Do you tell your application vendor what CPU you are running on? Or what motherboard you are using? Xen is transparent. That''s the whole point; otherwise I don''t think it is useful. Xen should be treated as yet another hardware, and I don''t think it is necessary to tell anyone about it, unless of course, you are running a software that explicitly deals with hardware, in which situation you should probably not run Xen, unless you know what you are doing. >If I expect their useful help for deeply system integrated tools like Oracle, or graphics device drivers, or even kernel related functionality like Wacom graphics tablets and the requisite kernel module, you''re damned right I tell them. It''s relevant, and I''ve in fact helped debug hardware related issues with all of those professionally.> For the app vendor, it is irrelevant if you are running it inside xen or on baremetal. Even for windows it doesn''t matter, so for something that sits so high up in the application chain, why would you want to complicate and confuse the easily confused people by bringing in information about a software about which they probably haven''t heard about >Nonesense. I''ve *been* the vendor, and it makes a big difference what kernel you''re using. Lying to them just confuses the issues, and I''ve run into *that*, too. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Ligesh
2007-Jun-23 22:06 UTC
[Xen-users] Re: XEN and Windows Guests in critical environment(hospital)
On Sat, Jun 23, 2007 at 10:44:15PM +0100, Nico Kadel-Garcia wrote:> Ligesh wrote: > > > For the app vendor, it is irrelevant if you are running it inside xen or > > on baremetal. Even for windows it doesn''t matter, so for something that > > sits so high up in the application chain, why would you want to > > complicate and confuse the easily confused people by bringing in > > information about a software about which they probably haven''t heard about > > > Nonesense. I''ve *been* the vendor, and it makes a big difference what > kernel you''re using. Lying to them just confuses the issues, and I''ve > run into *that*, too. >We are talking about application written in dotnet. That''s pretty much 3 layers above the kernel. I had very clearly stated that if you are running kernel level apps, then it is better you stick with real hardware, or even if you are migrating, you get the proper experts to do it. We are talking about pure user-level applications, like the one in the Original Post--a dotnet medical app --which would mostly be dealing with accounting and database. So obviously, it has nothing to do with Xen or the hardware. If you are running apps that has device drivers, then the situation is completely different, but that''s obvious. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Nico Kadel-Garcia
2007-Jun-24 01:01 UTC
[Xen-users] Re: XEN and Windows Guests in critical environment(hospital)
Ligesh wrote:> On Sat, Jun 23, 2007 at 10:44:15PM +0100, Nico Kadel-Garcia wrote: > >> Ligesh wrote: >> >> >>> For the app vendor, it is irrelevant if you are running it inside xen or >>> on baremetal. Even for windows it doesn''t matter, so for something that >>> sits so high up in the application chain, why would you want to >>> complicate and confuse the easily confused people by bringing in >>> information about a software about which they probably haven''t heard about >>> >>> >> Nonesense. I''ve *been* the vendor, and it makes a big difference what >> kernel you''re using. Lying to them just confuses the issues, and I''ve >> run into *that*, too. >> >> > > We are talking about application written in dotnet. That''s pretty much 3 layers above the kernel. I had very clearly stated that if you are running kernel level apps, then it is better you stick with real hardware, or even if you are migrating, you get the proper experts to do it. >Fair enough. but you''d be *amazed* at some of the fun and games that user-level applications do that are seriously affected by kernel subtleties. Proxy cache performance, for instance, is massively affected by subtleties of the "select" function.> We are talking about pure user-level applications, like the one in the Original Post--a dotnet medical app --which would mostly be dealing with accounting and database. So obviously, it has nothing to do with Xen or the hardware. If you are running apps that has device drivers, then the situation is completely different, but that''s obvious. >OK, be clear about that then. I still think it''s a bad idea to hide your full configuration from your vendor. The conflicts can really surprise you at odd moments, and I''ve seen them happen. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Ligesh
2007-Jun-24 03:14 UTC
[Xen-users] Re: XEN and Windows Guests in critical environment(hospital)
On Sun, Jun 24, 2007 at 02:01:25AM +0100, Nico Kadel-Garcia wrote:> Ligesh wrote: > OK, be clear about that then. I still think it''s a bad idea to hide your > full configuration from your vendor. The conflicts can really surprise > you at odd moments, and I''ve seen them happen. >You are talking about practical situations. I am as usual talking about theory. :-) Theoretically, xen shouldn''t enter the picture, but I can see in practice, it might be necessary. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
James Harper
2007-Jun-24 04:54 UTC
[Xen-users] RE: XEN and Windows Guests in critical environment(hospital)
> On Sat, Jun 23, 2007 at 10:44:15PM +0100, Nico Kadel-Garcia wrote: > > Ligesh wrote: > > > > > For the app vendor, it is irrelevant if you are running it insidexen> or > > > on baremetal. Even for windows it doesn''t matter, so for something > that > > > sits so high up in the application chain, why would you want to > > > complicate and confuse the easily confused people by bringing in > > > information about a software about which they probably haven''theard> about > > > > > Nonesense. I''ve *been* the vendor, and it makes a big differencewhat> > kernel you''re using. Lying to them just confuses the issues, andI''ve> > run into *that*, too. > > We are talking about application written in dotnet. That''s prettymuch 3> layers above the kernel. I had very clearly stated that if you arerunning> kernel level apps, then it is better you stick with real hardware, oreven> if you are migrating, you get the proper experts to do it. > > We are talking about pure user-level applications, like the one inthe> Original Post--a dotnet medical app --which would mostly be dealingwith> accounting and database. So obviously, it has nothing to do with Xenor> the hardware. If you are running apps that has device drivers, thenthe> situation is completely different, but that''s obvious.What you are saying is probably correct, assuming as you say that the applications in question are not data acquisition processes or anything requiring low level drivers. But the point you aren''t getting is about support, and what happens when something goes wrong. I made the assumption that the word ''critical'' used in the original post meant that lives could depend on it, eg if the application in question is responsible for scheduling medical test and sending results to the right person. In my experience the platform for ''critical'' applications like this is that they are sized and specified very tightly by the vendor, and if you have ever been in the role of support for such a thing you''ll understand why. So I agree with you that the application will almost certainly work and work perfectly, but everyone is very sensitive about legal responsibilities these days, and if someone died and it was determined that a computer failure caused the results of a medical test to be lost or delayed and that that was a contributing factor to their death, and it turns out that you used an unsupported system configuration to run the software on, then take a guess what happens next - the fact that it wasn''t Xen or VMWare''s fault will have nothing to do with the outcome of the inquiry... James _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Nico Kadel-Garcia
2007-Jun-24 06:01 UTC
[Xen-users] Re: XEN and Windows Guests in critical environment(hospital)
James Harper wrote:> > So I agree with you that the application will almost certainly work and > work perfectly, but everyone is very sensitive about legal > responsibilities these days, and if someone died and it was determined > that a computer failure caused the results of a medical test to be lost > or delayed and that that was a contributing factor to their death, and > it turns out that you used an unsupported system configuration to run > the software on, then take a guess what happens next - the fact that it > wasn''t Xen or VMWare''s fault will have nothing to do with the outcome of > the inquiry... >Heh. Yeah, been there. I used to design medical electronics for neural implants. All the designs had to be very belt-and-suspenders, and there was often a trade-off between keeping things the way they were, and keeping things supported or robust against the next potential threat. Since one of my first systems adminstration tasks was cleaning up after the Morris Worm, partly because I had specifically been forbidden from implementing the available system security patches to avoid any complicatons in a BSD 4.3 controlled neural stimulator, I got to see the booby traps of local configuration issues up close and personal. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Simon Capstick
2007-Jun-25 10:27 UTC
Re: [Xen-users] XEN and Windows Guests in critical environment (hospital)
Jordi Espasa Clofent wrote:> Hi all, > > One friend of mine are thinking about how to implement virtualization in > their critical job environment (a hospital). > The main "problem" is there are a lot of medical application builded in > .NET tecnology; so, I view three possible options: > > 1. Win server with VMware and win guests (IIS to support .NET). > 2. UNIX/Linux server with XEN (or XenEnterprise) and win guests > 3. UNIX/Linux server with XEN (or XenEnterprise) and UNIX/Linux guests > (Apache with mod_mono to support .NET) >... I don''t think virtualisation is necessarily going to help reliability although it may help in other ways. If money wasn''t an issue I would go for a redundant/mirrored iSCSI SAN with regular snapshots and remote backups and have bare metal Windows servers with HBAs booting off the SAN. Spare servers could be configured to take over a non-functioning one very quickly. You could still use virtualisation but you would need to couple it with something like heartbeat to make it more reliable than the bare metal servers. I would imagine if these applications are critical then the applications themselves should be written and be able to run in a redundant fashion anyway, otherwise the software is not really fit for purpose. You need to decide what amount of down time is acceptable. 1 second, 1 minute, 1 hour. 1 day etc.. Each has a technical solution, tending to be more expensive the more reliable it is. Simon _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users