|
Thin provisioning rules. Use it for pretty much everything. The exceptions I would consider would be on OS volume VMDKs if I was worried about them growing out and locking things up and was de-duping the whole thing so I could reclaim the empty blocks anyhow, because the vendor is absolutely screaming for it, or because you have a VMDK encapsulated in its own volume that isn't shared with anything else.
BangersInMyKnickers fucked around with this message at 03:26 on Feb 7, 2013 |
#
?
Feb 7, 2013 03:24
|
|
|
|
| # ? Apr 27, 2024 04:01 |
|
|
Corvettefisher posted:You might want to take a look at some of the Dell T420's they will probably hit the price point you are looking for. You can use vSphere for your deployment however View does have some additional costs where Citrix, for this scenario, may be more cost effective. Unless however you are just pooling the desktops on the server and giving RDP access to them, which would probably be the cheapest way. e. I see what you mean about View. That's out of my budget but RDP is an option. Above Our Own fucked around with this message at 04:20 on Feb 7, 2013 |
|
#
?
Feb 7, 2013 04:18
|
|
|
Powdered Toast Man posted:I got into a bit of an, uh, tussle with the other guys working on my production cluster and I was curious what the prevailing opinion is here: My answer to this question is "What tool do you have to tell you that you're running out of space and do you actually pay attention to said tool?" When customers ask me if they should do it on the array or in VMware I almost always respond with that. If you've got good monitoring in place and you actually listen to it you could theoretically be just fine doing thin on thin. That said most people I talk to don't even notice the storage views in the vSphere client and haven't even bothered to setup alarm actions to alert on reasonable thresholds. Regarding VMware stunning a VM when a disk runs out growth room: it does that because it has no way of knowing if you care about the data on that vmdk. It's protecting you from potentially catastrophic data loss and you might come to appreciate that when some dickface out there reboots all the controllers on your storage array or fucks up your SAN zoning.
|
|
#
?
Feb 7, 2013 05:00
|
|
|
1000101 posted:Regarding VMware stunning a VM when a disk runs out growth room: it does that because it has no way of knowing if you care about the data on that vmdk. It's protecting you from potentially catastrophic data loss and you might come to appreciate that when some dickface out there reboots all the controllers on your storage array or fucks up your SAN zoning. If someone did that in my environment I'd rip his spine out and beat him to death with it. We monitor the datastores themselves directly, but then we also monitor the volumes on the NetApp filers where those datastores live, too. That being said, we've still had some situations where something causes a thin provisioned OS volume to grow rapidly enough that we can't respond to it before the VM gets suspended. Thin provisioning is probably ok as long as the volume your vmdks are on has enough room for all of them to grow to their full provisioned size. Ours, uh...generally don't. This is mostly due to it being a lovely environment where there has been a lack of discipline with storage and some things just keep getting allocated more and more space without even asking why it is needed. We could probably be in better shape if we had a few more shelves but we haven't been given the budget for that.
|
|
#
?
Feb 7, 2013 05:24
|
|
|
We thin provision everything. We also have multiple sets of eyes on our alerts, which are properly configured. Thin VMDK on top of thin luns on top of thin volumes.
|
|
#
?
Feb 7, 2013 05:28
|
|
|
Above Our Own posted:Thank you. Will a vSphere Essentials Kit license be all I would need to run my VDI? It will be for <10 users and I'm not worried about High Availability. If you just want to create RDP links to desktops you can get away with the free vSphere. This is a cheap rear end solution mind you, however for the number you are dealing with, it would probably work fine. The other solution is to look up Citrix's VDI in a box, which might be a better solution for you, that or Terminal Services or whatever 2013 calls it. Dilbert As FUCK fucked around with this message at 05:49 on Feb 7, 2013 |
|
#
?
Feb 7, 2013 05:42
|
|
|
Layering thin provisioning is A-OK, but you should be pretty knowledgeable about what a problem looks like when any of the layers are full. And yes, preemptive warning (and having someone able to intervene in time) is pretty key. For instance when the LUN layer or under is affected, the entire VMFS datastore becomes inaccessible with completely stopped access in some cases. Hooray! There might be performance implications as well. I'm oversimplifying it, but writes that induce a provisioning of new blocks while on thin VMDKs (just to start at that layer) tends to require at least triple the writes, each time. - Filesystem changes are accounted for. - A chunk of VMDK blocks are zeroed. - Writes requested from the Guest are performed. - Layers underneath this layer of the stack (LUN, etc) may interleave somewhere around VMDK block zeroing. Ultimately there's an increase in writes, which can be more expensive, then there's added perceived VM/Guest write latency, due to multiple layers needing to conduct these provisioning operations. A lot of the literature out there about performance issues from doing Thin provisioning (single or multiple layers) will probably be quite dated if they don't take into account good solid state or memory-based caching. And quite honestly, even with the above said, I don't think the performance impact is actually notable in practice. You're only going to provision the blocks once in most circumstances, and from thereon just write normally. Basically don't try to do write-based benchmarking on completely new thin disks and expect realistic performance until your follow-up runs.
|
|
#
?
Feb 7, 2013 09:40
|
|
|
Upgraded our main VM host last night 
|
|
#
?
Feb 7, 2013 16:18
|
|
|
scrub.
|
|
#
?
Feb 7, 2013 16:45
|
|
|
Nitr0 posted:scrub.  and not for the obvious reasonsSandy Bridge tops out at 1.5 TB for a 4-socket box. How did you end up at 1.25 TB?
|
|
#
?
Feb 7, 2013 16:53
|
|
|
8 cards, 8 slots on each, 2x32gb - 6x16gb on each card = 1280
|
|
#
?
Feb 7, 2013 17:00
|
|
|
oh yea x 3 machines.
|
|
#
?
Feb 7, 2013 17:02
|
|
|
I really with I had the screen cap of where I made my Lab machine use 100% of it's CPU resources when I rescanned my HBA's and my ISCSI box reset at that exact moment. Which made me have to shutdown a bunch of VM's and restart them via cli. Was a fun day Nitr0 posted:8 cards, 8 slots on each, 2x32gb - 6x16gb on each card = 1280 Now make a 500Gb Ram disk 
Dilbert As FUCK fucked around with this message at 17:16 on Feb 7, 2013 |
|
#
?
Feb 7, 2013 17:13
|
|
|
Nitr0 posted:scrub.  We're a small shop. I knew someone was going to blow my stats out of the water, and expected it really, but goddamn. Sadly I'm pretty sure that 32GB of ram maxes out this blade. The memory sites all say the blade supports 8GB sticks, so we might be able to do 64... but the actual manual for the blade says the most it supports is 8x4GB so we didn't want to risk it. I don't know how to figure out if we'd really work with 64 without blowing the money and trying it out. I'm afraid to call IBM because they'd probably just read the manual that I already read, and I'm afraid to trust the memory manufacturers because they just want to sell memory.  Meanwhile, I'm just glad we have plenty of room to spare now, and can install an Exchange 2010 VM without worrying about host swapping.
|
|
#
?
Feb 7, 2013 17:25
|
|
|
Nitr0 posted:scrub. Don't capitalize the 'W'!
|
|
#
?
Feb 7, 2013 19:24
|
|
|
Nitr0 posted:scrub. What processors are you using? Even our low-voltage Xeons start clocking in at at least 2ghz.
|
|
#
?
Feb 7, 2013 19:26
|
|
|
I work in a really small environment but I'm investigating virtualizing our entire setup w/VMWare. (As a side note, I'd love if somebody here wants to PM me for some general advice about the project and if it's even feasible). In any event, part of the virtualization would include our UTM, which has a virtual appliance image available. There would be three subnets, two of which that would reside solely in the VM server and one that would have to connect to the physical LAN. And WAN uplink of course. I'm trying to envision how exactly to accomplish this. If I understand correctly, I need to do the following: - Create 1 "internal" switch that hosts the VM-only subnets, no pNICs attached. - Create 1 "external" switches, with pNICs going out to WAN/LAN. - The UTM VM would have one vNIC in the internal switch, and two vNIC in the external switch (one LAN, one WAN). Am I on the right track here?
|
|
#
?
Feb 7, 2013 19:34
|
|
|
BangersInMyKnickers posted:What processors are you using? Even our low-voltage Xeons start clocking in at at least 2ghz. Would have to find the guy that bought these I'm not sure 
|
|
#
?
Feb 7, 2013 20:06
|
|
|
How do you get the Performance charts in VMware able to change the date/times, rather than just the real-time chart of the last hour? Or is that yet another feature I can't use without vCenter?
|
|
#
?
Feb 7, 2013 20:26
|
|
|
Probably Xeon E7530
|
|
#
?
Feb 7, 2013 20:29
|
|
|
Frozen-Solid posted:How do you get the Performance charts in VMware able to change the date/times, rather than just the real-time chart of the last hour? Or is that yet another feature I can't use without vCenter? Yeah, you need vCenter for that. The data stream gets pushed to the server where it is aggregated and managed, otherwise your local hosts would be full of performance data being logged at 1 second intervals.
|
|
#
?
Feb 7, 2013 20:34
|
|
|
Frozen-Solid posted:How do you get the Performance charts in VMware able to change the date/times, rather than just the real-time chart of the last hour? Or is that yet another feature I can't use without vCenter? Yeah, you need vCenter. The hosts don't save historical performance data.
|
|
#
?
Feb 7, 2013 20:34
|
|
|
Frozen-Solid posted:How do you get the Performance charts in VMware able to change the date/times, rather than just the real-time chart of the last hour? Or is that yet another feature I can't use without vCenter? You can do this with vcenter because it stores that data over time. When it's just a standalone host there is nowhere to store the datapoints so you can't go back in history; just get realtime.
|
|
#
?
Feb 7, 2013 20:34
|
|
|
Yup, you'll need vCenter, Essentials kit is stupid cheap and can manage 3 hosts, you might want to look it up. PS: 1000101 Did you get my PM the other day?
|
|
#
?
Feb 7, 2013 20:41
|
|
|
Corvettefisher posted:Yup, you'll need vCenter, Essentials kit is stupid cheap and can manage 3 hosts, you might want to look it up. We've looked into it, and it's on our list of things we want, but we're a tiny company and all our VMware licenses and support contracts were setup before the Essentials pack was a thing. It's one of the wonderful things I learned when I took over as network administrator... my predecessors bought licenses for features we can't use without vCenter, without buying a vCenter license. No vMotion, HA, backups, etc for me. Plus we're looking to cut costs because the economy is just now starting to hit us. I doubt I'll get those toys any time soon. There are too many other upgrades that are higher on the priority list (like getting off of Groupwise). Unfortunately it seems there's no "cheap" way to convert our current 2 Host licenses to an essentials kit. We just have to buy vCenter outright 
Frozen Peach fucked around with this message at 20:51 on Feb 7, 2013 |
|
#
?
Feb 7, 2013 20:47
|
|
|
Nitr0 posted:8 cards, 8 slots on each, 2x32gb - 6x16gb on each card = 1280
|
|
#
?
Feb 7, 2013 20:51
|
|
|
Frozen-Solid posted:Unfortunately it seems there's no "cheap" way to convert our current 2 Host licenses to an essentials kit. We just have to buy vCenter outright Is there $500 available for the basic Essentials kit? I mean, it doesn't get you a LOT of features but at least you have vCenter for stuff like historical graphs and Update Manager. If your licensing is a total shitshow (like mine is) sometimes it's best to just eat it and start over, especially if you can do it all for a few hundred bucks. Essentials is still better than 2 standalone hosts.
|
|
#
?
Feb 7, 2013 20:52
|
|
|
x3850 x5. There's the cpu too. xeon e7-4807
|
|
#
?
Feb 7, 2013 20:52
|
|
|
three posted:Don't capitalize the 'W'! Our cafeteria has is printed as VM Ware on the receipts. The spacing might have just been weird font, but the W was capitalized. Our own loving cafeteria can't even get this right.
|
|
#
?
Feb 7, 2013 20:57
|
|
|
Docjowles posted:Is there $500 available for the basic Essentials kit? I mean, it doesn't get you a LOT of features but at least you have vCenter for stuff like historical graphs and Update Manager. If your licensing is a total shitshow (like mine is) sometimes it's best to just eat it and start over, especially if you can do it all for a few hundred bucks. Essentials is still better than 2 standalone hosts. We're planning on getting vCenter, and I'm hoping it'll happen by the end of the year. Our licensing isn't bad, we just have 2 vSphere Standard licenses and we're only paying support on them now. One $1500 vCenter licenses should be all we need, but right now our primary focus is getting off of Groupwise and SQL Server 2000. Getting modern email and SQL is more important. That should be done within the next month or two, and then we can look at vCenter again. We're small enough that we can only really handle one migration at a time.
|
|
#
?
Feb 7, 2013 21:05
|
|
|
Nitr0 posted:scrub. drat YOU. I just bumped my hosts up from 64 to 128GB and I was feeling pretty good about myself.
|
|
#
?
Feb 7, 2013 23:14
|
|
|
Pff, triple-channel of bust you scrubs.
|
|
#
?
Feb 7, 2013 23:17
|
|
|
BangersInMyKnickers posted:Pff, triple-channel of bust you scrubs.
|
|
#
?
Feb 7, 2013 23:22
|
|
|
 Nehalem, actually. I probably won't get the chance to upgrade before we push all our vm's up to some kind of managed virtualization paltform and rip the datacenter out of this building entirely.
|
|
#
?
Feb 7, 2013 23:29
|
|
|
Whoa, why have I never heard of these E7 processors before? gently caress me, I want a couple of the 10-core chips.
|
|
#
?
Feb 8, 2013 02:14
|
|
|
Nitr0 posted:scrub. Who's gunna be the scrub when a host fails and takes out 700 VMs in one hit?  I started building our new cluster yesterday - 16x Dell M620 2xE5-2670 / 256GB RAM Customized ISO & kickstart scripts saved my life, the cluster was built in a day! Would rather gouge my eyes out than configure the same 20 portgroups on each host.
|
|
#
?
Feb 8, 2013 02:40
|
|
|
DevNull posted:Our cafeteria has is printed as VM Ware on the receipts. The spacing might have just been weird font, but the W was capitalized. Our own loving cafeteria can't even get this right.
|
|
#
?
Feb 8, 2013 02:59
|
|
|
Frozen-Solid posted:
What model is the blade?
|
|
#
?
Feb 8, 2013 03:02
|
|
|
A little setup for my question: - I just upgraded my "Mac" to two SSDs. One for Windows, one for MacOS. - MacOS is physically on one SSD, Windows is physically installed on the second SSD. - I am using (latest/greatest) VMware Fusion on MacOS to run the Windows partition as a "Boot Camp" image. Will VMware Fusion take advantage of TRIM on the second SSD? Should I be worried that the emulated hardware will not support TRIM? When I boot the Windows partition directly (not virtualized) I can verify that TRIM is up and running, however I'm not exactly certain how adding virtualization comes into play here. I'm going to crosspost this to the MacOS thread just in case.
|
|
#
?
Feb 8, 2013 03:33
|
|
|
|
| # ? Apr 27, 2024 04:01 |
|
|
GrandMaster posted:Who's gunna be the scrub when a host fails and takes out 700 VMs in one hit? That's why there's 3 of those hosts kiddo.
|
|
#
?
Feb 8, 2013 03:36
|
|