|
Dilbert As gently caress posted:Well you need to look at the business requirements; ESXi is a great common knowledge that doesn't require the knowledge transfer as libvirt. Sure some things out best vmware but when thinking for the business needs you gotta sometimes thing more along the continuity needs. What I'm saying is twofold: KVM isn't a product, and you don't "get your feet wet" with KVM unless you are a kernel developer working on it or follow trunk development and are happy running "qemu --slew-of-options" every time you want to run guest and are running an extremely trimmed system (but that's libvirt trunk, qemu communicates with KVM over... libvirt). But KVM is a kernel module. Nothing more or less. Libvirt handles everything. List of running guests? That's libvirt. Live migration? libvirt. "Business needs" and "continuity requirements" are meaningless terms here. Do I want to run just a few VMs on one host? virsh and virt-manager may be fine. Want HA VMs? Add corosync and pacemaker. Want a web interface? Archipel (or kimchi for one host). Want something which does it all and is broadly equivalent to vSphere//Center? oVirt or XCP. Want cloud stuff? Eucalyptus, OpenStack, or OpenNebula. Want something else? Roll your own infra out of the pieces these projects use. But all of these have pros and cons, and none are "learning KVM". That's like learning VMkernel. Saying "I'm mastered VMware so I'm gonna learn about KVM then Xen" is nonsensical because daily use of KVM and xen is 99% identical (libvirt), and the real difference happens at boot time until you get into "how do I map PCI devices to guests?" Running guests? Same tools, same interfaces. If you want to learn what some business uses, and you don't know what they actually use, you should broadly be familiar with libvirt, not KVM. It's not a homogenous ecosystem with addons like vWhatever. Each project has different goals and requirements with a lot of overlap. Platitudes about business needs aren't going to change the basics. E: continuity is bullshit which VMware isn't always great at either. virt-v2v lets you easily swap between guests, but companies aren't jumping from OpenStack today to oVirt tomorrow to Archipel next week. Continuity is fine inside product lines, and the knowledge is broadly cross-applicable. But oVirt shops are fine on oVrt, and Openstack on openstack, etc. Diablo -> Essex -> Folsom -> Grizzly -> Icehouse -> Havana works. RHEV has never broken compatibility despite completely rewriting the codebase and moving the web UI from .NET on Windows to Java on Tomcat on Linux. KVM doesn't break compatibility. But moving from RHEV to XCP would be about the same as GSX -> ESX. Use virt-v2v to move your guests, set up your new pools, and go. RHEV to oVirt would be ESX -> ESXi. VMware doesn't do this perfectly either. evol262 fucked around with this message at 16:46 on Feb 9, 2014 |
#
?
Feb 9, 2014 16:34
|
|
|
|
| # ? May 30, 2024 08:46 |
|
|
evol262 posted:This is nonsensical. Libvirt is a hypervisor abstraction layer which removes the need for knowledge transfer. "Business needs" aren't relevant in this sense because businesses use a product which abstracts this even further (rhev, xcp, whatever). But libvirt is how you generate storage pools equivalent to vmfs, etc. This sort of fragmentation is why 2004, 2005, 2006, 2007, 2008 and 2009 were the year of Linux on the desktop, and why KVM in the datacenter will suffer the same fate within generalist shops.
|
|
#
?
Feb 9, 2014 17:00
|
|
|
Misogynist posted:Everything you wrote is technically true, and is also a great list of reasons why so many people will happily pay money for vSphere instead of spending weeks getting GNU/LibviKVMacemakernux to run and fail over a single VM. I don't disagree that vSphere is preferable to plain KVM, and that trying to make live migration work from plain libvirt or virsh is frustrating starting from a base of zero knowledge. But I don't think anybody expects KVM to succeed on its own. It competes as a backend to oVirt and OpenStack in the same way as ESXi competes as a backend to vSphere/Center. Sane people don't use esxcli for management in the same way that sane people don't use virsh. KVM is not a competitor to VMware. RHEV is. XenServer was. OpenStack on KVM is when compared to OpenStack with nova+ESX. And people should learn a product. Not KVM. It just grinds my gears to see comparisons of "ESXi (meaning the vSphere Client, really) and KVM (meaning some muddled combination of libvirt, the module KVM actually is, some distribution of Linux, and a vague idea of how they might work together)".
|
|
#
?
Feb 9, 2014 17:21
|
|
|
This is all really great information and insight. Thanks for helping me to understand how KVM fits in to the greater picture of things. As you can tell I'm greatly uninformed as to how Esxi truly compares or doesn't compare to KVM based virtualization platforms.
|
|
#
?
Feb 9, 2014 17:33
|
|
|
Misogynist posted:Everything you wrote is technically true, and is also a great list of reasons why so many people will happily pay money for vSphere instead of spending weeks getting GNU/LibviKVMacemakernux to run and fail over a single VM. this was the mind set I was going on when I replied.
|
|
#
?
Feb 9, 2014 21:18
|
|
|
Dilbert As gently caress posted:this was the mind set I was going on when I replied. Which is still irrelevant. Shops don't build infra on flat KVM, and VMware gets no wins here. RHEV, XenServer, XCP, and OpenStack are real live products with transferable skills and knowledge.
|
|
#
?
Feb 10, 2014 00:21
|
|
|
evol262 posted:Which is still irrelevant. Shops don't build infra on flat KVM, and VMware gets no wins here. RHEV, XenServer, XCP, and OpenStack are real live products with transferable skills and knowledge. I think we are speaking from two completely different points here. I'm not shunning anything you said, or debating the authenticity of other options. But when he mentioned "Is there anything out there that is even half as nice as Virtualbox, as far as GUI management of KVM?" that lead me to think he is starting out on something and may not be ready to dive into the twerks of setting up all the other stuff. Again what may be "just a simple set of commands to set something up" to someone learning concepts can be key, especially on a well supported platform. Which makes expanding to other platforms easier because you can relate processes and procedures to things you are familiar with. quote:Shops don't build infra on flat KVM, "Mah son whose good with line-x set this up he's in college going for an english degree now but you can fix this stuff right?" Dilbert As FUCK fucked around with this message at 00:46 on Feb 10, 2014 |
|
#
?
Feb 10, 2014 00:42
|
|
|
evol262 posted:Which is still irrelevant. Shops don't build infra on flat KVM, and VMware gets no wins here. RHEV, XenServer, XCP, and OpenStack are real live products with transferable skills and knowledge.
|
|
#
?
Feb 10, 2014 00:52
|
|
|
Misogynist posted:"Transferable" is a funny word, because you need another place actually running it to transfer those skills to. What's the market share of RHEV right now? I know a good amount of large enterprises that use it but there is a lot more of the SMB, and small-medium enterprises that don't. just gotta use the right tool for the job like anywhere
|
|
#
?
Feb 10, 2014 01:04
|
|
|
Misogynist posted:"Transferable" is a funny word, because you need another place actually running it to transfer those skills to. What's the market share of RHEV right now? Dilbert As gently caress posted:I think we are speaking from two completely different points here. I'm not shunning anything you said, or debating the authenticity of other options. But when he mentioned "Is there anything out there that is even half as nice as Virtualbox, as far as GUI management of KVM?" that lead me to think he is starting out on something and may not be ready to dive into the twerks of setting up all the other stuff. Again what may be "just a simple set of commands to set something up" to someone learning concepts can be key, especially on a well supported platform. Which makes expanding to other platforms easier because you can relate processes and procedures to things you are familiar with.  virsh is transferable. Every major virtualization suite on Xen or KVM will support libvirt and virsh.But Linux virt doesn't need those skills. There's no reason to touch libvirt on a real product which presents you with other interfaces (cinder/swift APIs, vdsm, xapi, whatever) to manage it, and you're only mucking with raw libvirt if you're a developer of one of those products or have a major problem. If you're familiar with vCenter, RHEV-M isn't very unfamiliar. "KVM", again, is like trying to use raw vmkernel. Learn products. Dilbert As gently caress posted:Also work for an MSP in the SMB space; you'll see some zonky poo poo.
|
|
#
?
Feb 10, 2014 01:22
|
|
|
4.0 to 5.0 upgrade is pretty much done. Still need to do the fourth host in the cluster at the co-lo, but that should only take around an hour + reconfiguring SRM. Noticed that even our idle CPU usage has dropped by about 10%. Screwed around with the first one for a little too long since I did a fresh install because of an upgrade, but I realized we're going to get new servers at the end of the year, plus 5.0 u3 is the highest version the hardware can run, so why install it 'right' so that we can remote upgrade later?
|
|
#
?
Feb 10, 2014 14:42
|
|
|
Bob, I don't get what question you are asking here? --- I also see 'Transferable' as 'Will it get me a job', at which point I'd really be questioning the value of investing time learning a suite with less than 20% marketshare unless I already had a job lined up for it. I mean sure, you need to be conversant with the tech if you want to head into that space, but there's no point digging into specific implementations until you know you're going to need them, especially if that involves skipping the market gorillas.
|
|
#
?
Feb 10, 2014 15:05
|
|
|
Mausi posted:Bob, I don't get what question you are asking here?
|
|
#
?
Feb 10, 2014 15:11
|
|
|
Mausi posted:I also see 'Transferable' as 'Will it get me a job', at which point I'd really be questioning the value of investing time learning a suite with less than 20% marketshare unless I already had a job lined up for it. Whether you want to learn something which isn't the 300lb gorilla is another question, but "marketable" is "will this get me a job?" "Transferable" is "can I take these skills somewhere else relatively unchanged?" The open source world has very few market gorillas. You're always going to find people using new stuff, or smaller projects. But that aside, the question is "should I learn the underpinnings that I'll almost never use unless I'm at a job that requires a lot more Linux skills than I have, or should I just learn openstack or rhev, because those are the products I'm likely to encounter?"
|
|
#
?
Feb 10, 2014 15:17
|
|
|
Bob Morales posted:4.0 to 5.0 upgrade is pretty much done. Still need to do the fourth host in the cluster at the co-lo, but that should only take around an hour + reconfiguring SRM. Did you see your memory usage go up, too? evol262 posted:Whether you want to learn something which isn't the 300lb gorilla is another question, but "marketable" is "will this get me a job?" "Transferable" is "can I take these skills somewhere else relatively unchanged?" I hope they find some semblance of cohesive strategy, because hodge-podge bullshit is terrible to manage unless you're so big you can do whatever you want or you're so small who cares. I'd rather give VMware  than manage a bunch of stuff jammed together with bubblegum, if my goal was an efficient IT organization. That said, the latter is more fun!
|
|
#
?
Feb 10, 2014 15:23
|
|
|
three posted:Did you see your memory usage go up, too?
|
|
#
?
Feb 10, 2014 15:46
|
|
|
three posted:I hope they find some semblance of cohesive strategy, because hodge-podge bullshit is terrible to manage unless you're so big you can do whatever you want or you're so small who cares. The whole argument is that large open source shops who got into the virtualization game early jammed stuff together with bubblegum, but have the engineering talent to manage it (this is exactly what AWS and most of the cloud providers are/were). You probably want to know how libvirt works here, so you can fiddle with it. Other shops that don't want VMware or go with another provider for some reason end up with XenServer, Hyper-V, or RHEV, with some scattered around with oVirt (which is upstream for RHEV), XCP (open XenServer before XenServer was open), and a growing segment of "private cloud" shops running OpenStack, which is a whole different mess to set up. It's absolutely safe and reasonable to learn RHEV, XCP, Hyper-V, or whatever instead of learning "KVM" by learning libvirt when you may never use libvirt at all, and it requires learning a lot of other fiddly Linux stuff that the products don't make you learn. "Either I give VMware or I manage a bunch of stuff jammed together with bubblegum" is a false dichotomy. There's a third choice.RHEV, Hyper-V, and XenServer have smaller ecosystems than VMware, both because they're newer products and because they have less marketshare, and they're rougher around the edges. But they're coherent products that mostly work without messing with them. The "glued together with bubblegum" shops are extremely large and very invested in Linux engineering who decided that existing products don't fit their needs and are happy with what they have or tiny shops who won't give to any vendor so they cobble together something which kinda works for them with bubblegum. But you'd react to those the same way you would if you were trying to make it a VMware shop -- use a product (virt-v2v, XenConvert, whatever) to get them onto some platform which isn't glued together with bubble gum.
|
|
#
?
Feb 10, 2014 16:44
|
|
|
So if I understand you right here, you're saying that if you know VMware well, and want to branch out, but aren't a strong Linux admin then head for XenServer, Hyper-V, or RHEV. If you know your onions on Linux, then you could probably dig deeper into OpenStack, libvert, virsh etc, but then if you're that good you've probably already heard of them? This at least follows my intuition.
|
|
#
?
Feb 10, 2014 16:54
|
|
|
evol262 posted:The whole argument is that large open source shops who got into the virtualization game early jammed stuff together with bubblegum, but have the engineering talent to manage it (this is exactly what AWS and most of the cloud providers are/were). You probably want to know how libvirt works here, so you can fiddle with it. My complaint is more with the people that fall into the principle of "PULL YOUR ENVIRONMENT UP BY ITS BOOT STRAPS, WE CAN MAKE ANYTHING WORK" insanity. Obviously, this isn't everyone and there are plenty of logical setups. I just hate engineers that try to glue things together, and that tends to happen more in the "emerging markets" of the virtualization worlds. Just because someone can, doesn't mean someone should. If it takes several different unconnected, tiny projects to develop the base of the environment, it is broken.
|
|
#
?
Feb 10, 2014 17:32
|
|
|
Mausi posted:So if I understand you right here, you're saying that if you know VMware well, and want to branch out, but aren't a strong Linux admin then head for XenServer, Hyper-V, or RHEV. If you know VMware well and you want to learn its competition, then head for XenServer, Hyper-V, or RHEV, none of which require you to know the gritty bits of how libvirt actually works, though XenServer and RHEV support those bits (though generally with another layer inbetween that handles it for you, and you shouldn't need to touch libvirt). If you know your onions on Linux and want to set up a very small lab environment to learn exactly how live migration and such is handled behind the scenes, you should use Ubuntu, Fedora, or CentOS flat libvirt+KVM, You should know that you'll probably never use this stuff on a production environment using a real product unless a support engineer asks you to. OpenStack doesn't have a unified product yet. Rackspace Private and RDO are moving into this space, and VMware might, but it's a big mess of Puppet scripts, command-line clients, and a web UI that might do what you need. It'll probably take a few years for someone to abstract this all away, and you should expect any OpenStack deployment you're working on to be sort of messy behind the curtain. You still don't need to know libvirt. OpenStack has its own APIs. three posted:My complaint is more with the people that fall into the principle of "PULL YOUR ENVIRONMENT UP BY ITS BOOT STRAPS, WE CAN MAKE ANYTHING WORK" insanity. Obviously, this isn't everyone and there are plenty of logical setups. I just hate engineers that try to glue things together, and that tends to happen more in the "emerging markets" of the virtualization worlds. I agree with this. I'm just trying to reiterate that the actual products in this space aren't pulling themselves up by their bootstraps. Yes, there's a bad habit of everyone who can rub two brain cells together starting their own project because project XYZ doesn't fit their exact idea of how it should be. But taking several unconnected tiny projects to develop the base of the environment is exactly how Linux works in all its forms. Even the most basic kernel+userland+SSH+vi environment is pulling from at least 4 disparate projects which have nothing to do with each other, and more when you consider that OpenSSH uses libraries maintained by the NetBSD project.
|
|
#
?
Feb 10, 2014 18:13
|
|
|
three posted:Just because someone can, doesn't mean someone should. If it takes several different unconnected, tiny projects to develop the base of the environment, it is broken. I would argue that this is BloodyStupid™, but they are paying me for it...
|
|
#
?
Feb 10, 2014 18:13
|
|
|
We all seem to be agreeing so nevermind. 
three fucked around with this message at 18:45 on Feb 10, 2014 |
|
#
?
Feb 10, 2014 18:42
|
|
|
Bob Morales posted:4.0 to 5.0 upgrade is pretty much done. Still need to do the fourth host in the cluster at the co-lo, but that should only take around an hour + reconfiguring SRM. You using the HP image as well? the G5's actually run 5.1 u2(so long as you use the HP image) fine but vmware won't support it.
|
|
#
?
Feb 10, 2014 19:16
|
|
|
Dilbert As gently caress posted:You using the HP image as well? I figured they would but I didn't want to have some issue down the line and then have support be an issue.
|
|
#
?
Feb 10, 2014 19:23
|
|
|
^nice Also anyone here at PEX? I have a few friends out there one taking his VCDX defense today in a hour or so, I made it a semi requirement that I go to VMworld and PEX for my new job so I shouldn't have trouble meeting any of you guys if ya'll go. Dilbert As FUCK fucked around with this message at 19:30 on Feb 10, 2014 |
|
#
?
Feb 10, 2014 19:24
|
|
|
Dilbert As gently caress posted:^nice Sitting in a PEX bootcamp for vCAC right now actually. The lunches this year are loving horrible.
|
|
#
?
Feb 10, 2014 21:33
|
|
|
1000101 posted:Sitting in a PEX bootcamp for vCAC right now actually. The lunches this year are loving horrible. Really? they are worse that the water taco's and fajita soup from last year?
|
|
#
?
Feb 10, 2014 21:46
|
|
|
Can someone explain OpenStack to me as if I were five years old? I've tried to read and comprehend the Wiki page for a few days now but each time I get a paragraph in my eyes just glaze over.
|
|
#
?
Feb 11, 2014 02:59
|
|
|
I have a question about VMWare and how it handles CPU allotment on the VM itself. I don't know quite how to phrase it though so I'll just ask. Can a hypervisor pool and/or split physical resources to make a virtual machine think it is being hosted on more or less processors than it actually is using? In other words, can an administrator split one physical CPU so that the virtual machine "thinks" it's on a two core machine but really it's just using one physical core that's handling the demands of two processors? And likewise, can an administrator set a virtual machine to use multiple physical processors to make the virtual machine think it's being run on a machine with fewer but more powerful processors? So like can I use a physical quad core processor as a host for one virtual machine that thinks it's running on a dual core machine, then devote two physical cores per single "virtual" core so the virtual machine just thinks it's on a really, really fast dual core machine? GreatGreen fucked around with this message at 04:47 on Feb 11, 2014 |
|
#
?
Feb 11, 2014 04:45
|
|
|
GreatGreen posted:I have a question about VMWare and how it handles CPU allotment on the VM itself. I don't know quite how to phrase it though so I'll just ask.
|
|
#
?
Feb 11, 2014 05:20
|
|
|
Cool, thanks! Finally, if I'm running VMware workstation and have two servers running, both with with dual core processors, can I allocate 4 processors to a single virtual machine?
|
|
#
?
Feb 11, 2014 05:38
|
|
|
GreatGreen posted:Cool, thanks! only allocate what is needed I do a talk on some of this here, https://www.youtube.com/watch?v=Ntf0SBmWL1U
|
|
#
?
Feb 11, 2014 05:47
|
|
|
Martytoof posted:Can someone explain OpenStack to me as if I were five years old? I've tried to read and comprehend the Wiki page for a few days now but each time I get a paragraph in my eyes just glaze over. The 5 year old version is "a software suite for building your own version of Amazon AWS". It gets confusing quickly because it is trying to be all things to all people to some extent, and open source groups tend to be awful at or not care about marketing. Or, you know, writing documentation 
|
|
#
?
Feb 11, 2014 06:21
|
|
|
Docjowles posted:The 5 year old version is "a software suite for building your own version of Amazon AWS". It gets confusing quickly because it is trying to be all things to all people to some extent, and open source groups tend to be awful at or not care about marketing. Or, you know, writing documentation Andrew Clay Shafer wrote a terrific rant on everything wrong with OpenStack's community process back in November.
|
|
#
?
Feb 11, 2014 06:40
|
|
|
Martytoof posted:Can someone explain OpenStack to me as if I were five years old? I've tried to read and comprehend the Wiki page for a few days now but each time I get a paragraph in my eyes just glaze over. This is some months old and doesn't include Heat or Ceilometer (which are for orchestration and billing, respectively), but it's still broadly accurate. Misogynist posted:OpenStack has the opposite problem. Half the product is marketing-driven by vendors who half-bake OpenStack integration so they can tick a box on their features list. The other half is projects that never should have existed in the first place. Misogynist posted:Andrew Clay Shafer wrote a terrific rant on everything wrong with OpenStack's community process back in November. The community process isn't ideal and OpenStack can be a confusing hodgepodge of seemingly rudderless projects sometimes, but that rant is scattershot at best and says nothing so much as "I was way too involved in the politics of this project".
|
|
#
?
Feb 11, 2014 15:55
|
|
|
Who all here has complained about the remote console performance connecting to ESX? I need some help convincing people that we need to improve it. Send me a PM.
|
|
#
?
Feb 11, 2014 18:48
|
|
|
Strangely enough I haven't had any problems with it. I'm in Texas and regularly have to connect to ESX hosts around the world, even to a site in Manaus, Brazil which is literally in the middle of the Amazon Rainforest and has a super lovely WAN link and it's always been serviceable.
|
|
#
?
Feb 11, 2014 20:12
|
|
|
skipdogg posted:Strangely enough I haven't had any problems with it. I'm in Texas and regularly have to connect to ESX hosts around the world, even to a site in Manaus, Brazil which is literally in the middle of the Amazon Rainforest and has a super lovely WAN link and it's always been serviceable. drat it, that isn't what I want to hear.  Are you just using a low standard for what kind of performance to expect? I am trying to get an improvement that would let you watch smooth video in your VMs if you so desired.
|
|
#
?
Feb 11, 2014 20:18
|
|
|
Oh. If I'm connecting directly to a VM like that I'm either doing the initial install or troubleshooting something. My expectations are super low, just need to be in it long enough to get it joined to the domain so I can RDP, or fix it so I can RDP.
|
|
#
?
Feb 11, 2014 20:20
|
|
|
|
| # ? May 30, 2024 08:46 |
|
|
DevNull posted:drat it, that isn't what I want to hear. I think the complaints were mostly the Web/flash stuff; but I've never had any real issues to report.
|
|
#
?
Feb 11, 2014 20:20
|
|