|
Sylink posted:I was under the impression you don't need to defrag under linux. Generally true. Filesystems will become fragmented under linux over time but it's generally so slight that it doesn't even matter. I've seen machines that have been running for years only have something like 1% fragmentation. I still like to run xfs_fsr on my xfs filesystems and e4defrag occasionally. Also tools like e4rat can use the defrag ioctls to migrate commonly used files to the front of the disk speeding up boot.
|
#
?
Jan 18, 2013 22:18
|
|
|
|
| # ? Apr 29, 2024 14:34 |
|
|
OnceIWasAnOstrich posted:Sorry if this is a dumb question. I'm trying to use ESXI 5.1 to virtualize a couple of machines we use in our research lab. I've got two physical NICs. I originally put the management interface on vmnic2, using the IP 10.1.185.4, and everything is great. I wanted to add a VMKernel port so I can hook this up to our NAS with ISCSI. I add a VMKernel port to vmnic1 and connect using the IP 10.0.185.2. Suddenly my client loses connection, and I can no longer connect to the management interface at 10.1.185.4...but I can connect to 10.0.185.2.
|
|
#
?
Jan 19, 2013 02:06
|
|
|
The other important piece is that fragmentation is only really important for certain I/O workloads. On a desktop, when you're one person doing one thing, the fragmentation overhead adds up because you turn a sequential workload into a random one. But when you have dozens or hundreds of users hitting a service on the same shared storage concurrently, all your sequential I/O turns random anyway as your OS tries to round-robin the requests.
|
|
#
?
Jan 19, 2013 06:10
|
|
|
Were having trouble getting authentication working right in a child domain. I ended up having to manually add the users from the parent domain to the vcenter server for anyone but enterprise admins to login to vcenter. I was hoping to add the local Administrators which contains the right nested groups in the web based SSO administration tool, but that thing makes no loving sense. Next week I will deploy this patch for the VMware tools spew, which is really loving annoying. http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=2036350
|
|
#
?
Jan 19, 2013 11:20
|
|
|
I have an interesting virutalization project coming up this year that I don't quite know how to implement. We're redoing our build system and we'd like to get as much of it running on VMs as possible. (Let's ignore the "is this a good idea or not" part for now...) Our big hurdle in the whole process revolves around CUDA. A part of one of our build tasks can take advantage of CUDA in a big way. It drops the task completion time down by a couple orders of magnitude so it is a requirement. I was hoping to find a way to present a CUDA GPU to a couple of VMs on this new cluster. I've read a lot of really terrifying things about using DirectPath and GPUs and it sounds like the Nvidia VIBs are going to be View only for the time being (and may not expose CUDA to the guest anyways). The GPUs will be either Telsa or Quadro cards, depending on the vendor and what they certify. Has anyone done anything like this before?
|
|
#
?
Jan 22, 2013 21:00
|
|
|
Number19 posted:I have an interesting virutalization project coming up this year that I don't quite know how to implement. We're redoing our build system and we'd like to get as much of it running on VMs as possible. (Let's ignore the "is this a good idea or not" part for now...) You won't be able to do that. If you use DirectPass to hand the card over to the VM, then that VM has exclusive control of the hardware. ESX wouldn't be able to share it across the VMs. If you have ESX control the GPU, then it shares it across the VMs. Unfortunately, it only supports limited capabilities to the guest. You get DirectX 9.0c and OpenGL 2.1 in the guest. You don't get CUDA.
|
|
#
?
Jan 22, 2013 21:26
|
|
|
Number19 posted:Has anyone done anything like this before? In this case I would recommend you keep the CUDA boxes as physical machines in order to get the best performance and also as DevNull stated, you can't do what you want to do.
|
|
#
?
Jan 22, 2013 21:31
|
|
|
Number19 posted:I have an interesting virutalization project coming up this year that I don't quite know how to implement. We're redoing our build system and we'd like to get as much of it running on VMs as possible. (Let's ignore the "is this a good idea or not" part for now...) View 5.2 is rumored to utilize Nvidia GPU's in a virtual environment, here is a sneak peak: http://www.simonlong.co.uk/blog/2012/10/25/vmware-view-3d-gaming-experience/ Pretty cool, but not out yet; However, I believe Citrix does GPU acceleration already. Yeah Citrix looks like it only supports a 1:1 mapping, View 5.2 should be able to divy it up similar to how they were doing it above Dilbert As FUCK fucked around with this message at 21:43 on Jan 22, 2013 |
|
#
?
Jan 22, 2013 21:39
|
|
|
1:1 mapping would be fine. I don't need to split the GPU up since only one task out a dozen actually uses the GPU. Having one VM per host with the GPU directly attached would work well. This is more about logically splitting up the CPUs than the GPUs. I've always been pretty sure that this won't work or it's just a bad idea in general but it's worth a shot. I'll probably use something like rCUDA (assuming that works) and just push a full virtual solution off until the next generation.
|
|
#
?
Jan 22, 2013 21:53
|
|
|
Corvettefisher posted:View 5.2 is rumored to utilize Nvidia GPU's in a virtual environment, here is a sneak peak: I didn't realize that the vib has not been released yet. The 3D functionality in no way depends on View. It is a feature of ESX 5.1 that builds upon the software rendering that was released with ESX5. I guess someone decided to hold off with the vib since the View team is really the group asking for 3D support the most. Or really at all. If you are familiar with 3D on Workstation and Fusion, this is the exact same thing. 3D has worked on those products for years because our backends can hook into the graphics stack on the hosted system. ESX5.0 ported over the backend and graphics stack to the ESX kernel. There were no drivers or hardware though, so the rendering was done with software. ESX5.1 added the ability for Nvidia drivers to be loaded into the kernel. This means the 3D rendering is now done with hardware. The capabilities in the guest are not limited by the physical device. They are limited by the virtual graphics card presented to the guest. Those capabilities are DX9c and OGL2.1 and are the same across all the recent products, hosted and ESX alike. Number19 posted:1:1 mapping would be fine. I don't need to split the GPU up since only one task out a dozen actually uses the GPU. Having one VM per host with the GPU directly attached would work well. This is more about logically splitting up the CPUs than the GPUs. I've always been pretty sure that this won't work or it's just a bad idea in general but it's worth a shot. Pass though would work for you then. Just keep in mind that you will lose a lot of other VM capabilities by having a pass though device.
|
|
#
?
Jan 22, 2013 21:56
|
|
|
Corvettefisher posted:View 5.2 is rumored to utilize Nvidia GPU's in a virtual environment, here is a sneak peak: I thought about using View for this. It only supports Windows 7 VMs but I could probably get away with that. I'm not sure if this method supports CUDA either though. The main reason behind all this is for disaster recovery. We need to deliver builds fairly often and being able to stand the whole build system up somwhere else reasonably quickly would be very useful. Obviously in the case of a full on disaster my recovery environment is unlikely to have CUDA available but in that case we'd live with the slow build times while I figured something out.
|
|
#
?
Jan 22, 2013 21:58
|
|
|
DevNull posted:I didn't realize that the vib has not been released yet. The 3D functionality in no way depends on View. It is a feature of ESX 5.1 that builds upon the software rendering that was released with ESX5. I guess someone decided to hold off with the vib since the View team is really the group asking for 3D support the most. Or really at all. That eliminates View for this which at least makes it clearer. To be honest, it feels (and always felt) like I'm trying to do something that's not going to work as well as I'd want it to. It'll have a lot of gotchas and other quirks. I'll just look into offloading the CUDA to a physical host for this generation.
|
|
#
?
Jan 22, 2013 22:03
|
|
|
Number19 posted:That eliminates View for this which at least makes it clearer. To be honest, it feels (and always felt) like I'm trying to do something that's not going to work as well as I'd want it to. It'll have a lot of gotchas and other quirks. I'll just look into offloading the CUDA to a physical host for this generation. The best you would be able to do is the pass though for the CUDA build. It would give you some benefits, but you would have to decide if it was worth the cost. You are the actually first person that I have ever heard of that wants CUDA in a VM. In theory, it could be virtualized. The problem is that not enough people use it to justify the work to do that. There is also the fight with OpenCL. Neither one of them is really used much. If one of them started to get widespread usage, then you would see an effort to virtualize it.
|
|
#
?
Jan 22, 2013 22:08
|
|
|
It just feels like I'm giving up most of the benefits of a VM by doing this. I do realize that I'm ahead of the curve on some of this stuff and that it's probably not something that a lot of companies are wanting to try. The money's in VDI right now so it's going to see the lion's share of the work. Overall no big deal...I just wanted to see if someone had done it or had heard of it being done. I'll put it lower in my options pile for now and re-evaluate it later on. Thanks for the input.
|
|
#
?
Jan 22, 2013 22:48
|
|
|
DevNull posted:I didn't realize that the vib has not been released yet. The 3D functionality in no way depends on View. It is a feature of ESX 5.1 that builds upon the software rendering that was released with ESX5. I guess someone decided to hold off with the vib since the View team is really the group asking for 3D support the most. Or really at all. Yes my guess based off what I have read is it requires 5.1's SR-IOV feature to help divy up the video cards resources the the corresponding requests, in conjunction with the VIB's is allows for acceptable hardware acceleration. VDI is really the driving factor for this. Number19 posted:That eliminates View for this which at least makes it clearer. To be honest, it feels (and always felt) like I'm trying to do something that's not going to work as well as I'd want it to. It'll have a lot of gotchas and other quirks. I'll just look into offloading the CUDA to a physical host for this generation. I actually think Cirtix will do supported GPU acceloration in a VDI sort of environment, so if you had a small team of engineers say 10 you could use the free licence for them, however I am not fully aware of all the limitations Citrix has for GPU mappings to Virtual Desktops Dilbert As FUCK fucked around with this message at 22:53 on Jan 22, 2013 |
|
#
?
Jan 22, 2013 22:50
|
|
|
Corvettefisher posted:Yes my guess based off what I have read is it requires 5.1's SR-IOV feature to help divy up the video cards resources the the corresponding requests, in conjunction with the VIB's is allows for acceptable hardware acceleration. No, it is not SR-IOV. It is a fullly emulated graphics device that is presented to the guest. If you look in any VMware guest, you will see the SVGA2 graphics device. Your guest will treat it exactly like a physical graphics card. It is completely virtualized. We manage the 3D state of the guests in the backend and we send commands to the physical device to do rendering. The guest will never see anything close to what is physically in the machine as far as graphics cards.
|
|
#
?
Jan 22, 2013 23:06
|
|
|
Yes I was talking about how View 5.2 will handle the GPU acceleration. I mistyped and put 5.1 but did not specify I was talking about ESXi 5.1, I am aware of how the current graphics work in the vmware environment. Sorry for the confusion, it's been a rough week.
|
|
#
?
Jan 22, 2013 23:58
|
|
|
Anyone have any experience with running ESXi from a SD Card on HP DL3xx G8 servers? I'm assuming it works well as long as you stick to supported cards. Any gotchas? We have a remote office in another country where hard drives are stupid expensive and could save quite a bit of money by not having the ESXi hosts with local drives. Boot from SAN is another option, but I'd like to explore the SD card first.
|
|
#
?
Jan 23, 2013 00:35
|
|
|
Corvettefisher posted:Yes I was talking about how View 5.2 will handle the GPU acceleration. View doesn't do anything for graphics. All they do is read back the frame from the SVGA2 device in the guest and remote it with PCoIP.
|
|
#
?
Jan 23, 2013 00:39
|
|
|
skipdogg posted:Anyone have any experience with running ESXi from a SD Card on HP DL3xx G8 servers?
|
|
#
?
Jan 23, 2013 01:01
|
|
|
skipdogg posted:Anyone have any experience with running ESXi from a SD Card on HP DL3xx G8 servers? I'm assuming it works well as long as you stick to supported cards. Any gotchas? We have a remote office in another country where hard drives are stupid expensive and could save quite a bit of money by not having the ESXi hosts with local drives. Boot from SAN is another option, but I'd like to explore the SD card first. I use embedded USB devices, really it seems like ESXI doesn't care what the USB is so long as the motherboard supports USB booting. These are primarily UCS servers. My supermicro server uses a off the shelf USB 8gb drive from walmart boots, runs, and patched fine. We actually have a few G7/G8's at the lab at school I can see about booting this week, should would, might have to download the hp iso though. DevNull posted:View doesn't do anything for graphics. All they do is read back the frame from the SVGA2 device in the guest and remote it with PCoIP. Yes I am speculating on view 5.2 which will use vSGA or Virtual Shared Graphics Acceleration. Which appearently will work with Nvidia Quadro cards to provide near real world GPU performance in the view environment https://www.youtube.com/watch?v=CrUGZGRbG6A Borederlands 2, which the SVGA 2 would poo poo the bed with. https://www.youtube.com/watch?v=FPPSS3TSsGQ Counterstrike, which the current might start and fail to play. Like so. Dilbert As FUCK fucked around with this message at 01:19 on Jan 23, 2013 |
|
#
?
Jan 23, 2013 01:13
|
|
|
Bleh both my remote options turned into dead ends. Looks like it'll be physical CUDA builders. I'll just spec them in a way that lets be convert to virtual if options open up down the road.
|
|
#
?
Jan 23, 2013 04:11
|
|
|
skipdogg posted:Anyone have any experience with running ESXi from a SD Card on HP DL3xx G8 servers? I'm assuming it works well as long as you stick to supported cards. Any gotchas? We have a remote office in another country where hard drives are stupid expensive and could save quite a bit of money by not having the ESXi hosts with local drives. Boot from SAN is another option, but I'd like to explore the SD card first. It was broke as hell in ESXi 4.1 / G6 servers, but works much better now in 5+ / G7+ environments. I like it.
|
|
#
?
Jan 23, 2013 05:13
|
|
|
Corvettefisher posted:Yes I am speculating on view 5.2 which will use vSGA or Virtual Shared Graphics Acceleration. Which appearently will work with Nvidia Quadro cards to provide near real world GPU performance in the view environment Dude, I work for VMware. On the graphics team. I am not speculating at all. I was the QA lead for the hardware acceleration project on ESX 5.1 back in 2011 before moving over to the dev team. I actually suggested some games and setups for Tim to demo in those labs at VMworld. They are in fact running on the SVGA2 virtualized device. Of course the device is fast enough. It is software, we can always make it faster. That is why virtualization is cool. vSGA is just marketing speak for hardware accelerated graphics as opposed to the software accelerated we had before. The guest doesn't know the difference between the two. The guest can't know the difference between the two. That is how we can vMotion from hardware to software acceleration and vise versa.
|
|
#
?
Jan 23, 2013 16:44
|
|
|
skipdogg posted:Anyone have any experience with running ESXi from a SD Card on HP DL3xx G8 servers? I chucked in a random SanDisk and it works fine on both my G7s and G8s
|
|
#
?
Jan 23, 2013 19:30
|
|
|
We're doing dual SD cards in our Dell PowerEdge R620 hosts with no internal storage. No issues.
|
|
#
?
Jan 23, 2013 19:32
|
|
|
three posted:We're doing dual SD cards in our Dell PowerEdge R620 hosts with no internal storage. No issues. All the Dell boxes I've got have only 1 SD card slot for the vFlash thing. Is this something new they're offering?
|
|
#
?
Jan 23, 2013 19:50
|
|
|
NullPtr4Lunch posted:All the Dell boxes I've got have only 1 SD card slot for the vFlash thing. Is this something new they're offering?
|
|
#
?
Jan 23, 2013 20:07
|
|
|
NullPtr4Lunch posted:All the Dell boxes I've got have only 1 SD card slot for the vFlash thing. Is this something new they're offering? Not sure how new it is, but it looks like this: http://www.dell.com/downloads/global/products/pedge/en/poweredge-idsdm-whitepaper-en.pdf
|
|
#
?
Jan 23, 2013 20:11
|
|
|
Oh that's pretty sweet. I'll have to see if I can get that for the new hosts we're ordering. 
|
|
#
?
Jan 23, 2013 22:19
|
|
|
Dearest VEAAM: Thank you for the lovely present of 68 orphaned snapshots! It made my day so much brighter because I just love cleaning up after your mess. gently caress you. Sincerely, Goon Matchmaker.
|
|
#
?
Jan 23, 2013 23:11
|
|
|
Oh Veeam.... Are all your VMs in one job? Or do you have them each as their own job?
|
|
#
?
Jan 23, 2013 23:25
|
|
|
Hoping some people here can give me some guidance on Windows 8 Hyper-V: I just got done updating my BIOS, enabling Intel VT-x, installing 16GB of RAM and enabling Hyper-V. I want to get started doing malware analysis using the Sikorski & Honig book but their rudimentary setup guide only covers VMWare for about five lines. What kind of Virtual Network / Virtual Switch setup do I want to create so that I can do the following: 1) Update the guest OS (Windows 7) via Windows Update 2) Sandbox the guest so that I can have total control over what gets to/from the internet. Everything in the book I'm using is all benign so I'm not worried about VM escape (and hopefully once I am I'll have a machine at work). Am I asking for trouble trying to host Windows 7? I have XP licenses sitting idle if a 2600K and 16GB of RAM won't cut it.
|
|
#
?
Jan 24, 2013 01:22
|
|
|
Moey posted:Oh Veeam....
|
|
#
?
Jan 24, 2013 03:39
|
|
|
Speaking of running off of running off of SD card, our environment runs on SD rather than disc (all hosts read from SAN) but 3 of our hosts came with like 300GB in disc each. Is there a good use for these or is it basically wasted space outside of the host? The guy who setup most of our environment left but said that ESXi 5.1 finally made this useful but I don't remember what the solution had to deal with.
|
|
#
?
Jan 24, 2013 05:04
|
|
|
I'd pull them out and save them for something else you might actually use them in. They're only going to waste power and possibly fail.
|
|
#
?
Jan 24, 2013 06:17
|
|
|
BangersInMyKnickers posted:I'd pull them out and save them for something else you might actually use them in. They're only going to waste power and possibly fail. Ours our currently serving as VM scratch space, which means logs and VM swap. Also some datastores because we suck at SANs (thank god we sVmotioned those VMs onto the SAN).
|
|
#
?
Jan 24, 2013 06:53
|
|
|
The logs don't take up much space, why not just point them at a SAN datastore so they're actually accessible in the event of a failure? How does it handle things with the vswap running on local disk? I assumed that would cause a vmotion boundary.
|
|
#
?
Jan 24, 2013 16:45
|
|
|
I store my logs on local storage, but VM swap is stored with the VM iirc. Since our SAN is 2x1GbE we also use local storage for stuff we don't need to vMotion (test machines) or stuff that would be detrimental to SAN performance (Syslog for 100+ VMs)
|
|
#
?
Jan 24, 2013 16:48
|
|
|
|
| # ? Apr 29, 2024 14:34 |
|
|
So, what IS a decent virtualization aware backup solution? I see people complain about Veeam quite a bit, but it seems like that's all that everyone uses. We're primarily a windows shop but have a couple important linux servers. Right now we're in talks with some vendors to go almost total virtualization. We've been using virtualiztion for awhile with Hyper-V, but pretty much anything is on the table at this point from having a mix of VMWare and Hyper-V to all VMware. I've been trying to get people on board with licensing System Center 2012 as it would ease management of our environment greatly (despite some linux machines here and there, we are still primarily a windows shop), but one of the hooks for selling that to the people writing the checks is that DPM2012 would take over our backups from our somewhat shaky Backup Exec setup we have now. However, DPM doesn't have host level backup of VMware nor does it have a native client for linux. So, I'm not sure if I can really continue pushing this due to the linux servers we have and the potential for the switch from Hyper-V to VMWare.
|
|
#
?
Jan 24, 2013 17:23
|
|
