|
GanjamonII posted:I have two hosts with 1gb NICs where we are configuring 10gb cards for iSCSI traffic. These are older 4.1 hosts and I can't upgrade them yet to a newer version for 'reasons' right now. If this is a bit basic I apologise, I am not really a vmware admin usually I just get to play with these things occasionally. I'm not understanding what your asking?
|
#
?
Mar 18, 2014 19:45
|
|
|
|
| # ? May 9, 2024 00:40 |
|
|
OK I didn't phrase that very well. Basically we have two 4.1 hosts with 1gb NICs. They have 10gb cards in them, but not configured. We are going to set up those cards to handle our ISCSI traffic, and what I am wondering how it will actually handle those additional paths to the storage. In my mind it show up as an additional path for each LUN, but I am not entirely sure if the 10gb vs 1gb thing matters. edit Second question - Anyone here used HP Power Protector with 5.5? Doesn't technically seem to be supported, just wondering if anyone has actually used it. GanjamonII fucked around with this message at 14:11 on Mar 19, 2014 |
|
#
?
Mar 19, 2014 14:07
|
|
|
Can you not just set the 10GbE path as active, and the 1Gb as standby? Or am I missing something here?
|
|
#
?
Mar 19, 2014 23:35
|
|
|
GanjamonII posted:OK I didn't phrase that very well. Basically we have two 4.1 hosts with 1gb NICs. They have 10gb cards in them, but not configured. It will show up as new paths so long as you mark them as such, just like your 1gb/s only now it will be 10gb/s. Or is this a HW based iscsi initiator?
|
|
#
?
Mar 20, 2014 00:12
|
|
|
Dilbert As gently caress posted:Or is this a HW based iscsi initiator? How many people actually use that? I had an old boss who always wanted to spring for the licenses to do the iSCSI offload with our Emulex cards, but I never saw a use in our environment.
|
|
#
?
Mar 20, 2014 02:20
|
|
|
Moey posted:How many people actually use that? I had an old boss who always wanted to spring for the licenses to do the iSCSI offload with our Emulex cards, but I never saw a use in our environment. Some banks that didn't want to go FC have for that slight performance gain on latency, it helps on VDI(I don't know why you would do VDI on ISCSI though), or CPU constrained environments. Or people who want to boot from san and not use auto deploy because %reasons%; then again very few people use auto deploy.
|
|
#
?
Mar 20, 2014 02:32
|
|
|
I am doing VDI with storage over iSCSI without issues? Our hosts are not CPU constrained though.
|
|
#
?
Mar 20, 2014 02:38
|
|
|
Moey posted:I am doing VDI with storage over iSCSI without issues? Our hosts are not CPU constrained though. You'll probably see better performance on storage with VDI on NFS, just an FYI. VDI works fine on ISCSI, it works better on NFS, ESPECIALLY if your array is VAAI aware. Recomposes and refreshes go much faster, as well as some features like disk reclamation work a bit better.
|
|
#
?
Mar 20, 2014 02:41
|
|
|
Dilbert As gently caress posted:Recomposes and refreshes go much faster Ahahahahaha. poo poo that pisses me off. Every user has a persistent desktop..... That is one of the things I need to fix once I get some time. Also we are running Nimble storage, so only iSCSI.
|
|
#
?
Mar 20, 2014 02:54
|
|
|
Moey posted:Ahahahahaha. Oh well that's uhh that's something... really nimble is iSCSI only? Did not know that.
|
|
#
?
Mar 20, 2014 02:57
|
|
|
We are just starting a deployment of 300TB of nimble. Nimble owns. iSCSI owns. gently caress nfs and emc and netapp and their bullshit.
|
|
#
?
Mar 20, 2014 03:04
|
|
|
Nitr0 posted:We are just starting a deployment of 300TB of nimble. I don't deny that I do tend to use iSCSI more for Server VM's but VDI on NFS is very very nice.
|
|
#
?
Mar 20, 2014 03:08
|
|
|
Nitr0 posted:We are just starting a deployment of 300TB of nimble. gently caress yea Nimble. Are you using expansion shelves, individual units, or the new "scale out" poo poo?
|
|
#
?
Mar 20, 2014 03:12
|
|
|
expansion shelves + scale out poo poo. Just got version 2 certified on all the arrays and installed yesterday.Dilbert As gently caress posted:I don't deny that I do tend to use iSCSI more for Server VM's but VDI on NFS is very very nice. We do a 1000 head vdi on nimble. Works great as well.
|
|
#
?
Mar 20, 2014 03:15
|
|
|
Dilbert As gently caress posted:Some banks that didn't want to go FC have for that slight performance gain on latency, it helps on VDI(I don't know why you would do VDI on ISCSI though), or CPU constrained environments. Or people who want to boot from san and not use auto deploy because %reasons%; then again very few people use auto deploy. These people boot from LUNs with i/gPXE. Hardware HBAs are pretty much never worth it
|
|
#
?
Mar 20, 2014 07:32
|
|
|
Moey posted:Ahahahahaha. I feel you brother. My lot won't listen to any other option, and are now moving towards local raid storage instead of NFS because it's 'the cheapest option available'  Obviously this will be simple local volumes, with no clustering, because who needs operational downtime?
|
|
#
?
Mar 20, 2014 13:10
|
|
|
When doing performance monitoring of VMs you usually want to monitor at the hypervisor level because if you monitor within the VM itself in the information may not be accurate. For example, a VM might think it's using 8g of RAM but due to compression/page sharing/whatever-it's-called it's actually only using 6g of physical RAM. Is this the case for KVM? I know it's tightly integrated into the kernel so can I trust that performance tools within a Linux guest running KVM will be accurate, or should I monitor at the host level?
|
|
#
?
Mar 20, 2014 13:42
|
|
|
hackedaccount posted:When doing performance monitoring of VMs you usually want to monitor at the hypervisor level because if you monitor within the VM itself in the information may not be accurate. For example, a VM might think it's using 8g of RAM but due to compression/page sharing/whatever-it's-called it's actually only using 6g of physical RAM. Kernel same pages and memory ballooning mean that you should also monitor on the host on Linux.
|
|
#
?
Mar 20, 2014 15:03
|
|
|
Thanks dude. If I run stuff like top/free/sar/whatever within the VM will those be accurate or should I consider them to incorrect (like with Linux Hyper-V & VMware VMs)? I think what you're saying is they should be accurate but I should monitor at the host level too. Do you or does anyone have any good, all inclusive, resources for learning KVM? I know 90% of the concepts and plan to set up a little lab. I would prefer a recent book or ebook because it's a pain to learn the details of a product using a mismatch of blog tutorials, wikis, videos, vendor documentation, etc because I never know what I don't know, ya know? Maybe I read and watched all this stuff on the internet but none of them talk about Feature X where books are usually pretty good at mentioning 95% of stuff and I can Google to dig down as needed.
|
|
#
?
Mar 20, 2014 21:56
|
|
|
hackedaccount posted:Thanks dude. If I run stuff like top/free/sar/whatever within the VM will those be accurate or should I consider them to incorrect (like with Linux Hyper-V & VMware VMs)? I think what you're saying is they should be accurate but I should monitor at the host level too. What I'm saying is actually the opposite. They're accurate as far as the guest knows, but the guest has no idea whether it's sharing memory pages or is overcommitted. Check the host. There aren't any good, all-inclusive resources for learning KVM. You can easily set up a little lab and migrate machines and such with just libvirt (you don't even need shared storage if you're running a recent libvirt version), and the documentation as libvirt.org and linux-kvm.org is about as good as you'll get. But that's all basically libvirt stuff (including the stuff on linux-kvm), since KVM is just a driver. However, I'm not gonna go into that one again. Learn a product. oVirt (which is upstream for RHEV), XenCloudplatform, Virtuozzo, Archipel, Openstack (nova is the KVM bits), whatever. Or flat libvirt and virt-manager if it's a small environment. But it's not nearly as coherent as "buy a vmware book, read it". What's your use case?
|
|
#
?
Mar 20, 2014 22:24
|
|
|
The RHCSA books on amazon cover a bit of KVM.
|
|
#
?
Mar 20, 2014 22:51
|
|
|
Dilbert As gently caress posted:The RHCSA books on amazon cover a bit of KVM. True, but not to a much greater level of detail than "right click a VM and choose Power On to have it power on" in the vSphere world.
|
|
#
?
Mar 20, 2014 22:58
|
|
|
evol262 posted:What I'm saying is actually the opposite. They're accurate as far as the guest knows, but the guest has no idea whether it's sharing memory pages or is overcommitted. Check the host. Got it. When it comes to Linux guests you should not trust the standard Linux utilities regardless of hypervisor (VMware, Hyper-V, KVM) and should watch at the host / hypervisor level instead. I know not to trust them for Linux on VMware and Hyper-V but was hoping there was some tight under the hood integration with KVM so I could trust em, but that isn't the case. No real use case. I know the basics of VMware and Hyper-V but never really dug into them because I don't want to be considered "the VMware guy" and have no desire to work with any part of a Microsoft stack after this gig is done. Linux on Hyper-V hasn't been bad but it means if I want to dig deeper I essentially have to learn Microsoft stuff (PowerShell, SCCM, etc) and I don't wanna do that. I want to move back to a full OSS / Linux type stack for my next job so it's essentially a decision between KVM and Xen and I think KVM will be the long-term winner. Once I get a good grasp of just plain KVM (and/or libvirt or whatever) then I'll move on to the platform but I want that fundamental KVM knowledge first. My gut tells me that eventually Platform X will go out of fashion but KVM won't so it's a long-term bet. I'll check out the docs on libvirt.org and linux-kvm.org. If anyone has any other suggestions toss em out there. Maybe it will end up being easier than I think.
|
|
#
?
Mar 21, 2014 07:53
|
|
|
hackedaccount posted:Got it. When it comes to Linux guests you should not trust the standard Linux utilities regardless of hypervisor (VMware, Hyper-V, KVM) and should watch at the host / hypervisor level instead. I know not to trust them for Linux on VMware and Hyper-V but was hoping there was some tight under the hood integration with KVM so I could trust em, but that isn't the case. You shouldn't trust the windows utilities regardless of hypervisor either. You can sort of trust the reporting on Xen paravirt guests. The upside for you is that "plain KVM" uses libvirt almost exclusively, as does Xen, so almost every concept will cross-apply. Because I don't think there's going to be a long-term winner between KVM and Xen (the projects have different goals). But I generally recommended learning a product because raw KVM isn't used by a lot of large shops. And fundamental KVM knowledge is abstracted away and literally useless in most products. I'm a RHEV developer. The entire stack is built on KVM. But you as an end user don't need to know anything about it, shouldn't touch it (we won't support you if you do), won't be able to find out any real useful information from it (openstack orchestrates through nova, rhev through vdsm), etc. By all means, learn it. Play with it. Read the docs. Figure out what it can do. But you're probably never going to apply any of that in a professional setting. You'll know the concepts and how live migration is implemented, but those concepts and implementations are pretty much the same across KVM, Hyper-V, vSphere, and KVM. It's essentially the networking argument again. A Cisco cert teaches you a lot about how Cisco does things and just enough actual networking knowledge to let you apply those concepts. VMware books teach you a lot about VMware and just enough actual virtualization to apply them. Reading about libvirt may teach you a lot about how virtualization works to understand what VMware/XCP/RHEV is doing under the hood, but it won't help you at all with actual problems you'll encounter at work, which will be 99.9% product related. I'm not trying to discourage you from learning about it, and you should do it if you're interested, but it's arcana. That said, if you really want to learn it front to back, you should install virt-manager and poke around a little. Check out the flags it's passing to qemu and how they're defined in the libvirt XML. Play around with doing things in the XML that virt-manager doesn't support yet (openvswitch bridges for guests are supported in libvirt but not virt-manager, as one example). Then try using virsh directly instead of virt-manager. Then qemu-kvm (which is generally a wrapper around qemu-system-${arch} accel=kvm) directly if you want to use KVM directly (qemu, since kvm_intel, kvm_amd, and kvm as kernel modules are actually KVM) features libvirt doesn't support. In order of least technical (still pretty technical) to most technical, read: libvirt-users qemu-discuss libvirt-devel
|
|
#
?
Mar 21, 2014 17:39
|
|
|
Good stuff man. I agree: I think KVM itself will be pretty easy to learn and it's the products that cause the most problems but I still plan to start at the bottom and work my way up the stack. Linux in a general sense is solid and I don't need to know much about internals for what I do, but I do need to know how to configure and troubleshoot Apache because that's what causes the problems - not the Linux kernel or libc or whatever. I know my next job will use KVM but what sits on top of that is still a mystery so I'm not going to commit to learning a specific product just yet; I'll get the fundamentals (KVM) down and go from there. Also, part of the problem is I don't understand the terminology and low level components yet. I keep saying "KVM" but I should be saying "KVM and lib-virt". For example, I know that KVM somehow uses XML files to define the VMs but I didn't know those XML files are actually a component of lib-virt and not of KVM itself, ya know? It sounds like I should focus on both KVM and lib-virt and those will be my foundation to move onto a product. I'm pretty good at filtering out the arcana. Knowing "vxmend" back in the Veritas Volume Manager days would save your rear end if a server crashed in a weird way but I only had to actually use it once a year so I didn't make the effort to memorize the whole command. You definitely need to pick and choose what you take the time to learn in this profession because the amount of poo poo to learn is endless. PS: I still refuse to learn perl because if I do this will be me someday at some job. NEVERRR!!! http://devopsreactions.tumblr.com/post/80249709145/fixing-bugs-on-the-inherited-code-base
|
|
#
?
Mar 21, 2014 19:10
|
|
|
hackedaccount posted:Good stuff man. I know Perl, but nothing obligates me to write it or tell people. Basically, you keep saying KVM when you mean "virtualization on Linux", which is common. KVM is in the kernel. It's kvm_intel, kvm_amd, and kvm (which they both use, plus stuff like KVM on ARM) qemu is the program used to run virtual machines. It can be accelerated with kqemu (a kernel driver which nobody really uses these days), xen, or kvm, which reuses some of the old kqemu hooks to make it easy. qemu generates abstract bits for the guests (network drivers, storage, VNC, etc). libvirt provides a standardized way to pass options to qemu. This came up because Xen had their own config files (and still does), but they were finicky, and you had to use a LISP-ish format to do complicated stuff. And kvm in its early days required invoking qemu --litany --of --option --to --create --virtual --machines. Example: code:That said, you foundation for moving onto a product is the product. libvirt is a relatively small part of openstack. RHEV uses vdsm for orchestration, and tooling with VDSM and management server APIs is how you get by. Eucalyptus and Cloudstack communicate through a java app server. libvirt and qemu-kvm knowledge might be helpful if you get in the poo poo and you end up needing to read libvirt logs, but you'll probably open a ticket then anyway... If you end up at a shop that's rolled their own infrastructure and isn't using a product, it may be worthwhile. But it's worth saying that almost every Linux shop I've ever been in uses VMware.
|
|
#
?
Mar 21, 2014 19:37
|
|
|
So like with RHEV are you saying it doesn't use libvirt at all, or are you saying VDSM sits on top of libvirt (bottom to top: KVM -> libvirt -> VDSM)? If it's the latter it seems once I move above the libvirt layer that's when it's time to pick my product and I'm cool with that. I don't want to become an expert on KVM / libvirt but if you find them in almost every product then I'll need to get a good grasp of them so I can re-use the knowledge when the next orchestration layer comes into vogue. I agree with what you're saying about learning the product from a job seeker's prospective: It isn't my detailed knowledge of libc that will get me tons of job offers but my knowledge of "chmod" and Apache will. I was originally thinking about digging deeper into Linux internals and realized it's much better from a job seeker's perspective to learn the higher level stuff instead BUT I still want that fundamental, re-unsable knowledge. I don't mind VMware but there's just soooo much to learn and given the choice I prefer an OSS-type stack. I also believe that in the long term VMware and KVM/Hyper-V will be like Linux and UNIX: VMware for your mission critical systems (eg: about 10% of your servers) and KVM/Hyper-V will be the other 90% because why pay the premium when you just need basic virtualization? No need to drop the money on a SPARC box to run your little departmental web server, ya know?
|
|
#
?
Mar 21, 2014 21:09
|
|
|
hackedaccount posted:So like with RHEV are you saying it doesn't use libvirt at all, or are you saying VDSM sits on top of libvirt (bottom to top: KVM -> libvirt -> VDSM)? If it's the latter it seems once I move above the libvirt layer that's when it's time to pick my product and I'm cool with that. I don't want to become an expert on KVM / libvirt but if you find them in almost every product then I'll need to get a good grasp of them so I can re-use the knowledge when the next orchestration layer comes into vogue. Personally, I think that in the long run, VMware will just license vCenter under the same terms as Hyper-V (something like giving away Essentials Plus for free) and sell products (Horizon, whatever) on top of it, but eh. Still, I don't want to be a broken record. You should subscribe to the lists I gave earlier. libvirt is a fundamental part of most of the products, but it's so fundamental that you'll pretty much never see it as a user or an administrator of those products. libc or glib are actually apt analogies. Learning it can tell you a lot about how VMware works, though.
|
|
#
?
Mar 21, 2014 21:34
|
|
|
Moey posted:How many people actually use that? I had an old boss who always wanted to spring for the licenses to do the iSCSI offload with our Emulex cards, but I never saw a use in our environment. Tried to do hardware iscsi with QLE8242s and it was an unmitigated disaster. I think I posted about how bad their software suite was earlier in the thread, but in addition to the horrible GUI, we could never get them to survive a controller failover. Timeout values weren't adequately tweakable. Switching to software iscsi fixed it with no real loss in performance.
|
|
#
?
Mar 22, 2014 23:45
|
|
|
If you wanted to start a career with Virtualization and VMware's products, is there a set certification path to follow in hopes to maybe get a entry level job somewhere?
|
|
#
?
Mar 24, 2014 00:56
|
|
|
mutantbandgeek posted:If you wanted to start a career with Virtualization and VMware's products, is there a set certification path to follow in hopes to maybe get a entry level job somewhere? You can get the online VCA which is pretty easy. From there you go for your VCP certificate. This requires a class to be taken and a test. After that the higher level stuff is a deep dive and more experience based in datacenter deployments or focusing on VMware view [VDI]. Worth going for it I would say.
|
|
#
?
Mar 24, 2014 01:36
|
|
|
Cronus posted:You can get the online VCA which is pretty easy. From there you go for your VCP certificate. This requires a class to be taken and a test. After that the higher level stuff is a deep dive and more experience based in datacenter deployments or focusing on VMware view [VDI]. I saw the VCP certificate class offered in NYC for like $4k, so I just wanted to know where to start before I drop that kinda cash on a class I shouldnt be taking now, or in the wrong order. I'll work on the VCA this week. Heres the class I found on VMware's site: http://mylearn.vmware.com/mgrreg/courses.cfm?ui=www_edu&a=det&id_course=198883 edit: http://www.globalknowledge.com/training/promo/dates.asp?pid=GAME&courseid=21549&catid=513&country=United+States#top You get a free XBONE!
|
|
#
?
Mar 24, 2014 01:48
|
|
|
mutantbandgeek posted:I saw the VCP certificate class offered in NYC for like $4k, so I just wanted to know where to start before I drop that kinda cash on a class I shouldnt be taking now, or in the wrong order. I'll work on the VCA this week. There's an online version of the class that only costs $185 that's offered by Stanly Community College - they usually have a long wait list, but they run 10 classes a quarter. If you sign up now there's a good chance you can get into one of the Summer classes (if you sign up now you will definitely get into one of the Fall classes). https://vmware.stanly.edu/waitlist.php I can confirm the class is legit and that it will count towards your VCP5-DCV requirements. After I took the class it showed up in my VMware transcript and I'm prepping to take the exam on May 5th.
|
|
#
?
Mar 24, 2014 02:19
|
|
|
I also got 75% off on the exam through my Community College. I'm testing on Wednesday!
|
|
#
?
Mar 24, 2014 03:04
|
|
|
So I guess now is as good of time as any to invest in the school and tech. So with that said...gonna need a lab! This will be my 4th setup, so here's what I have planned; let me know which you guys think is neat since I don't have any most of the hardware anymore. Time to start fresh and order some stuff from Amazon and enter credit card debit: #1: Basic ALL-IN-ONE: - $610
I still have the motherboard and one 8GB stick of ram, so I guess this is the cheapest option for me, but I'd need to re-do my file server since right now its using the MOBO and an G3220; plus a 32GB set since the one stick I have is 1333, and I don't know if the Xeon will play nice with it. #2: Whitebox Desktop: - $600-800?
#3: Lulz NUC Lab: - $850
Storage - $168 or $280 For storage I have a Synology 712+, which I can throw in x2 SSDs/Blacks for vMotion, iSCSI, etc. Network I want to also get my Cisco Certification, so I guess I'd need one of their switches. Read some good things about the Cisco SG300-10, which runs around $200. Maybe be better off getting a 16+ port HP Procurve instead? I have a 1810G, but its only 8 ports, and is pretty much full right now anyway. I'd want to keep the lab network separate from my full home network anyway, so a dedicated switch for JUST the lab would be a nice perk over hearing my household complain about "WHY IS FIREFOX DOWN," when I mess up a vLAN or something. Add in a few USB drives for ESXi install, and other silly stuff, like tax, so add $125. Really wanna keep everything around $1k said and done.
|
|
#
?
Mar 25, 2014 00:59
|
|
|
https://www.youtube.com/watch?v=Ntf0SBmWL1U Here's a talk about labs. Whitebox desktop is probably your best bet. For storage, I'd go DAS to an internal NAS, just pass up a raid controller.
|
|
#
?
Mar 25, 2014 03:13
|
|
|
Dilbert As gently caress posted:https://www.youtube.com/watch?v=Ntf0SBmWL1U Thanks I'll watch the video now. Internal NAS such as a Freenas VM and pass thru? Or a NAS connected over cross cable?
|
|
#
?
Mar 25, 2014 03:29
|
|
|
mutantbandgeek posted:Thanks I'll watch the video now. Internal NAS such as a Freenas VM and pass thru? Or a NAS connected over cross cable? Do what? Yeah I mean I run my freenas appliance as a VM, It works amazing if you have a controller that you can pass upto a VM. That way you don't have to worry about switching.
|
|
#
?
Mar 25, 2014 04:55
|
|
|
Is it worth to go for a Supermicro X10SL7-F with its internal LSI SAS2308 controller instead of a dedicated card? Can it be passed through as well?
|
|
#
?
Mar 25, 2014 06:52
|
|
|
|
| # ? May 9, 2024 00:40 |
|
|
yomisei posted:Is it worth to go for a Supermicro X10SL7-F with its internal LSI SAS2308 controller instead of a dedicated card? Can it be passed through as well? Should work, I don't know if it will work. I think rosewill has a 4 port adapter that freenas can use for like 50 bucks.
|
|
#
?
Mar 25, 2014 07:05
|
|