|
KennyG posted:I have read the past 35 or so pages trying to get up to speed, but I have one specific question. Does VMWare support clustering at the guest level. It looks like what you're asking is if VMware can combine the resources of multiple hosts to power one VM. The answer is no.
|
#
?
Nov 20, 2012 21:34
|
|
|
|
| # ? May 9, 2024 00:09 |
|
|
Oh are you trying to do something like a beowulf cluster?
|
|
#
?
Nov 20, 2012 21:36
|
|
|
To be clear - we have some multi-threaded applications that do a batch process. I can't say what it is as it's proprietary but let's take the annalogy of video encoding because it's common and people understand it. Basically, you can encode a file F in X/T time where X is the encode time and T is the theads. However, if you have A) pair of single socket quad core machines and b) one machine with 8 cores, A) can produce two halves of an output in the same time as B) can produce 1 complete output set. Since re-combining the output into a complete set is a non-trivial task, having a machine that is as large as possible results in the smallest time for the desired compute task. The rub is that pricing of vCenter is effectively linear, by processor. If I install it on one 4CPU box it's the same as installing it on four single cpu boxes. However, Hosts are close to geometric by processor. The 4P tax is pretty nasty. You also can't even buy an 8 socket host. All of this means that it can be more cost effective to buy 8 x 1 socket systems than buy 2 x 4 socket systems. Thank you KS. That answers that - so it's better to eitehr get bigger hosts or re-code the application to be more modular. I can't really say which it will be but I'm guessing 4slot hosts are cheaper than re-development Thanks.
|
|
#
?
Nov 20, 2012 21:55
|
|
|
That or go to a worker server paradigm, but at that point you may as well go bare metal and yes I'm flogging myself for saying that.
|
|
#
?
Nov 20, 2012 22:02
|
|
|
The 4 socket tax is indeed nasty -- not only cost, but rack space-wise. With HP's Intel servers, I can get double the socket density per rack using 2 socket servers over 4 socket. I believe only Dell sells an Intel 2u 4-socket server that would level that playing field. You can definitely buy an 8 socket host, though, and if HP's discounts are still the same. the DL980 and DL580 are surprisingly close in price. However given the 32 vcpu limit in VMware 5.1 a 64-CPU box isn't going to help much unless you go bare metal.
|
|
#
?
Nov 20, 2012 22:48
|
|
|
Virtualization isn't always the only solution. For this specific workload, going bare-metal might just be the right call. To go even further afield, have you looked into AMD's offerings? They have some G34 socket cpus with 16 cores. Though they are not as fast per core (and per MHz, arguably) compared to intel, you can build a dual socket 32 core server without breaking the bank. That's pending on if your application benefits from a larger number of slower cores or not, though...
|
|
#
?
Nov 20, 2012 23:34
|
|
|
I doubt 32 will remain the limit, but at that time maybe VMWare will implement the feature too. Hardware is an expensive sunk cost, where on software you usually can get a sizable discount on full retail with the upgrade price. (I can't specifically speak from experience from VMWare but I would be shocked if it weren't true.) It's a tough sell to the senior leadership to ask for $$$,$$$.$$ in hardware and then turn around in 2 years and ask for it again. Our cycles tend to be more in the 5-7 range, even for servers. I have tried pointing out that we are paying more in power than to buy new hardware but they just think I want new toys. As to the AMD question, I have seen several people talk up the AMD's in CPU heavy virtualization and perhaps the 6300 series will be better but at this time it seems like a rather sizable political gamble on my part. KennyG fucked around with this message at 23:47 on Nov 20, 2012 |
|
#
?
Nov 20, 2012 23:40
|
|
|
http://www.anandtech.com/show/5058/amds-opteron-interlagos-6200 http://www.anandtech.com/show/5279/the-opteron-6276-a-closer-look Wouldn't get your hopes up too much for amd, More cores does not always mean better.
|
|
#
?
Nov 20, 2012 23:54
|
|
|
Ok, another question I forgot to ask that came up when I was reading. We have an Oracle App (   ) One of the many features of Oracle is a flash cache space. This is essentially a scratch disk that is used like an "L2 cache" of sorts for things that would have aged out of the buffer cache and back to the hard disk. By design this is put on some extremely low latency storage space like a SSD or better yet something like a fusionIO IO drive. .01 ms IO latency vs 4-10ms That's a pretty good boost.I would love to utilize this on the host as I coud sub-divide a 800GB drive into 400GB for prod and 50-100gb for various environments and greatly improve flexibility. However can I do this with shared storage and v-motion? We have a NetApp iSCSI SAN and I obviously want to keep the benefits of shared storage, but since Oracle effectively uses the storage space as volatile storage (it does not count on that data being there when it reboots) I was wondering how I could expose it to the guest and make it work. Assume identical hosts in a 3 machine cluster. What buzzword/feature am I looking for? Oracle simply maps the space to a path in the guest OS filesystem.
|
|
#
?
Nov 21, 2012 00:44
|
|
|
Rhymenoserous posted:That or go to a worker server paradigm, but at that point you may as well go bare metal and yes I'm flogging myself for saying that.
|
|
#
?
Nov 21, 2012 07:04
|
|
|
KennyG posted:I would love to utilize this on the host as I coud sub-divide a 800GB drive into 400GB for prod and 50-100gb for various environments and greatly improve flexibility. However can I do this with shared storage and v-motion?
|
|
#
?
Nov 21, 2012 07:07
|
|
|
adorai posted:It's also not true. In this specific type of example, I might actually prefer to buy an 8 core box and create a 7 core VM that will run 100% than to go bare metal. Reason being that abstracting the hardware away from the application makes it easy to quickly upgrade the hardware with effectively no change to the guest except more speed. It really depends on what you're using, though. For VMware I'd have a really hard time justifying the licensing cost if there's no quantifiable cost savings from consolidation, especially with the performance hit. With KVM or whatever, not so much. Vulture Culture fucked around with this message at 08:40 on Nov 21, 2012 |
|
#
?
Nov 21, 2012 08:36
|
|
|
Misogynist posted:It really depends on what you're using, though. For VMware I'd have a really hard time justifying the licensing cost if there's no quantifiable cost savings from consolidation, especially with the performance hit. With KVM or whatever, not so much. As a XenServer and ESXi admin it isn't often that I recommend XenServer, but licensing costs are where it shines. I don't use free XenServer, but assuming all the hardware is the same in this case it may be a good option. The big advantage with XenServer free is you still get XenMotion Live Migration to move a VM live between phsycial hosts: http://www.citrix.com/products/xenserver/features/editions.html
|
|
#
?
Nov 21, 2012 16:08
|
|
|
I was reading about the new ARM A15s having virtualization extensions and was wondering if anyone here had heard anything about this and if there were any known plans to make use of these features in a user-facing way. IIRC KVM and Xen both support it, but the ARM server market is still pretty small so I can't say that really catches my interest yet. I think the business market would love the ability to virtualize on mobile devices, allowing a locked down primary OS to offer the user their own VM in which random apps can be installed and run while being kept entirely separate from any business data. On paper we could also see booting OSes that weren't designed for the devices as well, though in practice this will be mostly limited to running Android on iOS/Windows devices due to the legal issues.
|
|
#
?
Nov 21, 2012 18:48
|
|
|
I'm starting to get ready to upgrade to vSphere 5.1, and I think I'm going to replace my vCenter server (it was already upgraded from 4.1 to 5, and meh). The documentation says that SSO, Inventory, and vCenter can all run in separate VMs, or together, or whatever combination, but doesn't really provide guidance as to how to make that decision, other than "small" environments can use one VM and "larger" environments should split it up. Those terms are not quantified. We're super-small (4 hosts) and my initial thought was to use one VM so that it's easier to protect, but we use datacenter licensing for Windows, so there's no actual extra cost to using multiple VMs. Anyone have thoughts?
|
|
#
?
Nov 21, 2012 19:58
|
|
|
How many VM's are on those four hosts? Are you using a separate SQL server or the express? E: What licensing are you using Dilbert As FUCK fucked around with this message at 20:07 on Nov 21, 2012 |
|
#
?
Nov 21, 2012 20:04
|
|
|
Separate SQL. 85 VMs, but probably 1/3 of those are replica VMs from Veeam, or old powered off VMs.
|
|
#
?
Nov 21, 2012 20:07
|
|
|
In that case you would probably be fine with installing everything on one VM. vCenter might be a bit slow to come up in a HA event, or reboot. If anything, I would install VUM off the vCenter server, during updates it can get a bit grabby.
|
|
#
?
Nov 21, 2012 20:10
|
|
|
VUM is already separate. Is there a quantitative reason behind what you said, or just a hunch? If it really comes down to "should be", then I might as well separate the roles.
|
|
#
?
Nov 21, 2012 20:15
|
|
|
If I was deploying it on that small of an environment, I would have it all as one, if SSO is unavailible you can't administer vCenter anyways. So for a small environment(which vmware usually associates the term small environment 10 hosts 100VM's). http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=2034918 quote:Why would I install SSO on a separate machine from vCenter Server? Really for that environment unless you have SLA's not stated for HA and an HA event breaks those SLA's installing SSO/vCenter/Inventory Services, just ups the resource overhead on your cluster, the administrative overhead, Tshooting steps required, and backup resources. I mean you can go with 2 SSO servers in HA and affinity to speparate the VM's on different hosts, then do heartbeated vCenter servers, and have complete protection; but it depends what your environment needs. I just really don't see a benefit with what you have stated to install it all on separate VM's for that environment, unless you need a highly available environment that exceeds what VMware's HA will do. have you looked into the vCenter Virtual appliance by chance? Dilbert As FUCK fucked around with this message at 23:40 on Nov 21, 2012 |
|
#
?
Nov 21, 2012 23:27
|
|
|
adorai posted:setup an openindiana or freebsd guest on the host with the fusionIO and share it out via iscsi or NFS. Add that disk to your guest OS. Beware: if that host goes down, the datastore will go with it, so I would recommend exporting from two seperate hosts and using software mirroring or something. At that point, it may not actually boost speed. Unfortunately 800gb of ram is about 4-5x the price of a enterprise IO drive. Given our storage /vm requirements I'd likely go local storage if I can't do hybrid shared storage unless I am misunderstanding what you are saying?
|
|
#
?
Nov 22, 2012 01:29
|
|
|
wolrah posted:I was reading about the new ARM A15s having virtualization extensions and was wondering if anyone here had heard anything about this and if there were any known plans to make use of these features in a user-facing way. IIRC KVM and Xen both support it, but the ARM server market is still pretty small so I can't say that really catches my interest yet. I think the mobile stuff is why they are doing it now, but it will filter in to the server market arm fairly quickly. Frankly the whole idea seems terrible to me. VMware has been making pushes in that direction for a while because they see the dollar signs. More than likely this will target Android and BlackBerry 10. It might work okay on BlackBerry where you have a locked down and well maintained host layer and then jump between the isolated sessions for work and personal (but considering their track record for the last 5 years I am doubtful). On Android you have a massive clusterfuck of unmaintained handsets because the vendors don't give a singular poo poo. Throwing in a virtualization layer is going to complicate things even more as it will be sold as a security feature that actually poses an additional security risk to the device. I do not see that working out well.
|
|
#
?
Nov 22, 2012 02:10
|
|
|
KennyG posted:How would this work. If I did this why would I share each through the hypervisor/network instead of deploying vms with scratch disks directly?
|
|
#
?
Nov 22, 2012 04:48
|
|
|
adorai posted:You would lose your ability to vmotion and leverage VMware HA. Doesn't sharing it out with nfs or iscsi through a guest across the LAN or vLAN take latency a from a few micro seconds to milliseconds?
|
|
#
?
Nov 24, 2012 19:22
|
|
|
http://gc.digitalriver.com/store/vmware/en_US/Content/pbPage.thanksgiving_2012?src=SocialMedia_eBIZ_Facebook_Pre-Announce_CyberMonday_112612 VMware just posted this on their Facebook, anyone looking for workstation might want to check it out Monday.
|
|
#
?
Nov 24, 2012 20:48
|
|
|
Corvettefisher posted:If I was deploying it on that small of an environment, I would have it all as one, if SSO is unavailible you can't administer vCenter anyways. So for a small environment(which vmware usually associates the term small environment 10 hosts 100VM's). Fair enough, sounds like all on one VM is the way to go. I have not considered the appliance since VDR and SRM replication both left a bad taste in my mouth (I try not to stray too far away from just ESXi + vCenter now unless it's 3rd-party software). I don't have any problems with running it on my own Windows VM that the appliance would solve (that I know of).
|
|
#
?
Nov 26, 2012 15:23
|
|
|
I've been trying to find ways to shrink the vmdk for a Virtual Appliance I maintain. The file size is around 24GB and I want to try and compact it a little. It's a single growable virtual disk with a max size of ~60GB, I did some reading and defragmented it with MyDefrag then ran sdelete -c to whipe the empty space. After doing this it grew to the max of 60GB! I've tried shrinking it but it's still at 60GB, even used vdiskmanager to "convert" it to create a new vmdk and it's still 60GB. I've still got the original unfucked vmdk at 26GB though. Anyone have any magic tricks for shrinking it? I'll be packing it into a 7zip archive (Ultra/LZMA2) for consumption so the smaller I can get it the better. It used to be smaller then this too, the previous version of the Virtual Appliance was about half the size. Ashex fucked around with this message at 21:49 on Nov 27, 2012 |
|
#
?
Nov 27, 2012 21:11
|
|
|
Yeah don't defrag thin disks ha. You should be able to Storage vMotion to a thin disk and reclaim the size, I think vmware tools also has a 'shrink' option however I think I am thinking of something else.
|
|
#
?
Nov 27, 2012 21:25
|
|
|
Ashex posted:sdelete -c There's your problem. You wrote 0s to all free space, and since you wrote something to all the space, the VMDK grew to accommodate. Like Corvettefisher said, svMotion or shrink should work, and since you defragged, it should shrink to as small as you can get it.
|
|
#
?
Nov 27, 2012 21:33
|
|
|
Erwin posted:There's your problem. You wrote 0s to all free space, and since you wrote something to all the space, the VMDK grew to accommodate. Like Corvettefisher said, svMotion or shrink should work, and since you defragged, it should shrink to as small as you can get it. I guess I should have known it would grow, but why wouldn't it shrink after running sdelete? That's what really confused me. I suppose I'll just do a thorough defrag and shrink it. Edit: This appliance will be run with either VMware Player 4 or Workstation 7+, can I convert it to thin provisioned and still have it work in both of those? Edit Edit: Multi-tasking is making me forget to type words. Ashex fucked around with this message at 22:22 on Nov 27, 2012 |
|
#
?
Nov 27, 2012 21:51
|
|
|
Ashex posted:I guess I should have known it would grow, but why wouldn't it shrink after running sdelete? That's what really confused me. I'm confused as to why you are defragging within VMDK's. You basically told the VM to write 0's in the free space, 0 and 1 both require space, e.g you wrote data in the form of 0's to where no data existed prior. You essentially made your disk a Thick Eager Zero disk. http://www.yellow-bricks.com/2008/11/01/defraging-a-vmdk/ Wow just realized that was for Fusion however defragging on a VM isn't the same as you would think in traditional hardware. quote:Edit: This appliance will be run with either VMware Player 4 or Workstation 7+, can I convert it to thin provisioned and still have it work in both of those? It might, however you run a risk of loving the virtual hardware over, which can be fixed but isn't fun. Dilbert As FUCK fucked around with this message at 22:02 on Nov 27, 2012 |
|
#
?
Nov 27, 2012 21:56
|
|
|
Only reason I'm messing with shrinking it is the final archive size is double that of the previous release. Historically we've just upgraded the previous VM for the latest release which I'm not a fan of. I changed our procedure so each major release is built from scratch, I'm using the same base VM as the previous release but the final archived final is double that of the previous. I added some other things but it only increased the actual usage by maybe 1GB (1.5 max).
|
|
#
?
Nov 27, 2012 22:19
|
|
|
When there's thin provisioning of any kind, unallocated blocks are null. If you write a 1 or a 0 to the block, it's still a write. When something wants to read from blocks that are unallocated, a zero is returned. Secure-deleting writes zeroes over top of actual data that consists of ones and zeroes. This also, however causes all the previously unallocated blocks to also be written as zeroes. Instead of null, unallocated blocks, these are now legitimate blocks full of zeroes. Thin-provisioned disks are supported in even earlier versions of Player and Workstation, so you'll be fine whenever you sort out the defrag/re-allocation a second time around. Shrinking to me means you are reducing the partition size in the Guest disk, adjusting the file system accordingly, and then physically shrinking the whole virtual disk to the size of the partition (or larger). Defragging is especially necessary here. To shrink disks, it's not so much a necessity to defrag, but it'll help you free up even more space during the "re-thinning" effort during your second conversion or migration to Thin Disks again. Also thin disks are Copy-On-Write and are susceptible to additional fragmentation. Both thin disks and lazy zeroed-thick disks (default for ESXi VM disks) have some additional "write-cost" when allocating zeroes for the first time. I am pretty sure that Hosted (Server/Player/Workstation/Fusion) will fully-allocate disks with zeroes on creation if you don't elect them to be thin. For an unallocated block to be written-to for the first time, the following process is performed: Write command request -> Block is unallocated, so write Zeroes to a bunch of blocks sequentially -> Write data to one or more blocks as requested. Repeat as needed if it's a huge file. Keep that stuff in mind next time you try to benchmark writes and such for the first time. If you were to make a thin disk at first and later fill it with zeroes, it becomes an eager-zeroed thick disk.
|
|
#
?
Nov 27, 2012 22:21
|
|
|
Don't defragment VMs. It doesn't make sense. The defragmentation tool is totally unaware of how the VM is laid out on the SAN, and it's going to cause issues with storage auto-tiering, changed block tracking, etc.http://blogs.vmware.com/vsphere/2011/09/should-i-defrag-my-guest-os.html posted:Internally at VMware, we have not observed any noticeable improvement in performance after a defragmentation of Guest OSes residing on SAN or NAS based datastores.
|
|
#
?
Nov 27, 2012 22:22
|
|
|
Allow me to clarify, this isn't on a SAN rather a desktop. This is a virtual appliance that we use internally, everyone is running it on their laptops so being able to compact the initial VM as much as possible is ideal. I'm not defragmenting for performance purposes, purely so I can shrink the vmdk as much as possible.
|
|
#
?
Nov 27, 2012 22:52
|
|
|
 B-but what? Defragmenting doesn't save disk space or shrink existing files, it just rearranges I am seriously confused here. Dilbert As FUCK fucked around with this message at 23:01 on Nov 27, 2012 |
|
#
?
Nov 27, 2012 22:57
|
|
|
Any mentioning in my post about defragmentation would solely be for the rearrangement of blocks so the file system/partition can be safely shrunk (as in, instead of 20GB, it can be reduced to 15 if you move the last 5GB of content inward on the disk). This is more informational than anything - I am pretty sure that there isn't going to be any partition/filesystem shrinking going on; just re-thinning to reclaim zeroed space. The SAN equivalent is "Zero-Block Reclaim." On mechanical hosted disks like in this case, defragging does help with sequential performance, I suppose. For re-thinning I don't really know if there is much benefit from a defrag, but it can't hurt in this particular case before finally re-thinning the disk, for sure. Edit: Actually, defragging before re-thinning is probably best. If you somehow end up needing to defrag later, it might just end up bloating up the virtual disk. This is from a third-party or Windows-based defragger perspective, though. I don't really know how the Hosted VMware disk defragger behaves. It might not be filesystem-aware (i.e. does not care about what is inside the VMDK). Kachunkachunk fucked around with this message at 23:29 on Nov 27, 2012 |
|
#
?
Nov 27, 2012 23:24
|
|
|
Yep, just reclaiming zeroed space is all I'm trying to do. I'm very aware this sounds backwards and kinda silly but it keeps people from complaining to me about how small their laptop drive is. I'll try defragging then re-thinning, I've already shipped the Virtual Appliance so this is more practice for next release than anything else.
|
|
#
?
Nov 27, 2012 23:54
|
|
|
Ashex posted:Yep, just reclaiming zeroed space is all I'm trying to do. I'm very aware this sounds backwards and kinda silly but it keeps people from complaining to me about how small their laptop drive is. No offense, but how in the hell did you get this job?
|
|
#
?
Nov 28, 2012 00:21
|
|
|
|
| # ? May 9, 2024 00:09 |
|
|
madmaan posted:No offense, but how in the hell did you get this job? None taken! I do DevOPs, this is just one small thing out of the many that I do. The virtual appliance is our entire application stack that pre-sales/training/partners use for demo purposes. I'm in the middle of setting up a couple environments from scratch and upgrading another  .
|
|
#
?
Nov 28, 2012 00:58
|
|