From thin clients to desktops… not the other way!
-
@Dashrender we have a three node vSphere environment (x3550 M4, 64 Gb of ram), that host ERP, fileserver etc. Mainly windows VM, some linux. We got not-good-as-real-desktop performance on gigabit network (cat6, SG500 switches) with Praim zero clients, so after trying many tweaking on the connection server, I think that maybe is a lot less complex and cost-effective to migrate towards a fat client environment… the core i5 of today are very capable (including integrate graphics) and z240 is IMHO a very nice machine to do office editing, ero and other not so heavy tasks. In addition, the AMT capability is really nice when you want to provision or tweak a desktop in a remote fashion. We already have veeam b&r in production, so the endpoint on the desktop seems like a very clean and effective choice to re-image a broken machine.
-
It's not about reimaging broken machines to me. The storage space for backing up endpoints seems like a waste to me. Why not maintain an image to use for fast restores?
-
@Dashrender veeam backup include domain join, machine name… I'm not that experienced with windows imaging, I'm more a Linux sysadmin.
-
@coliver said in From thin clients to desktops… not the other way!:
@Dashrender said in From thin clients to desktops… not the other way!:
@coliver said in From thin clients to desktops… not the other way!:
Good luck we've got about 100 virtual desktops deployed and are looking to do another 100 next summer. We've got a big IBM SAN and a half dozen hefty servers doing the processing.
How many compute nodes do you have? Coliver
Six right now. We're haven't gotten close to maxing then out yet. That being said these machines perform at the physical desktop level but they aren't persistent, yet, so every time someone logs in they get a new desktop.
@coliver do you use altro something like nvidia grid to boost graphical performance? Do you have SSD in your SAN? The price of the SSD for our SAN frighten me (IBM), I can buy a lot more PCIe storage for my server, enough to keep VMs replicated between two or maybe all three node. Intel enterprise PCIe SSD is now at 0,8€/Gb or less, and the performance are really top notch… not your usual two-way 8Gbit/s fibre channel!
-
@Francesco-Provino said in From thin clients to desktops… not the other way!:
@Dashrender veeam backup include domain join, machine name… I'm not that experienced with windows imaging, I'm more a Linux sysadmin.
Well sure. It's a bare metal restore. I suppose if you have the storage for endpoint backups. But still seems like a large amount of capital spend (potentially) for something that will rarely be used.
Dedupe would greatly reduce the amount of storage for endpoint backups.But you're looking to deploy win 10... How are planning to do that? Imaging would be the fastest way to deploy a unified type desktop.
-
@Dashrender said in From thin clients to desktops… not the other way!:
@Francesco-Provino said in From thin clients to desktops… not the other way!:
@Dashrender veeam backup include domain join, machine name… I'm not that experienced with windows imaging, I'm more a Linux sysadmin.
Well sure. It's a bare metal restore. I suppose if you have the storage for endpoint backups. But still seems like a large amount of capital spend (potentially) for something that will rarely be used.
Dedupe would greatly reduce the amount of storage for endpoint backups.But you're looking to deploy win 10... How are planning to do that? Imaging would be the fastest way to deploy a unified type desktop.
Regarding the storage: yes, our NAS has already plenty of storage, Veeam deduplication will save a lot of space.
Windows 10 is already deployed in the VDI, z240 will include it as OEM. -
@Dashrender said in From thin clients to desktops… not the other way!:
It's not about reimaging broken machines to me. The storage space for backing up endpoints seems like a waste to me. Why not maintain an image to use for fast restores?
@Dashrender do you know about some easy and simple tutorial to do an automated deploy of windows 10, AKA image it?
My plan is to eventually reset the workstation in an automated way via AMT, something like "fire that script and forget". -
@Francesco-Provino said in From thin clients to desktops… not the other way!:
@Dashrender said in From thin clients to desktops… not the other way!:
@Francesco-Provino said in From thin clients to desktops… not the other way!:
@Dashrender veeam backup include domain join, machine name… I'm not that experienced with windows imaging, I'm more a Linux sysadmin.
Well sure. It's a bare metal restore. I suppose if you have the storage for endpoint backups. But still seems like a large amount of capital spend (potentially) for something that will rarely be used.
Dedupe would greatly reduce the amount of storage for endpoint backups.But you're looking to deploy win 10... How are planning to do that? Imaging would be the fastest way to deploy a unified type desktop.
Regarding the storage: yes, our NAS has already plenty of storage, Veeam deduplication will save a lot of space.
Windows 10 is already deployed in the VDI, z240 will include it as OEM.I did not think Veeam endpoint did dedupe
-
@Francesco-Provino said in From thin clients to desktops… not the other way!:
@coliver said in From thin clients to desktops… not the other way!:
@Dashrender said in From thin clients to desktops… not the other way!:
@coliver said in From thin clients to desktops… not the other way!:
Good luck we've got about 100 virtual desktops deployed and are looking to do another 100 next summer. We've got a big IBM SAN and a half dozen hefty servers doing the processing.
How many compute nodes do you have? Coliver
Six right now. We're haven't gotten close to maxing then out yet. That being said these machines perform at the physical desktop level but they aren't persistent, yet, so every time someone logs in they get a new desktop.
@coliver do you use altro something like nvidia grid to boost graphical performance? Do you have SSD in your SAN? The price of the SSD for our SAN frighten me (IBM), I can buy a lot more PCIe storage for my server, enough to keep VMs replicated between two or maybe all three node. Intel enterprise PCIe SSD is now at 0,8€/Gb or less, and the performance are really top notch… not your usual two-way 8Gbit/s fibre channel!
Nope, we don't do anything graphically intensive. The machines that do cad or graphics work at still traditional desktops. We do have 15K drives and use SSDs.
-
I was wondering where the bottle neck for the OP is? Is it RAM, disk IO, network bandwidth.
Not sure one can suggest that you bail on VDI if you don't know the reason your VDI is slow.
For example. If your switch is bad, over saturated, then new PCs won't help everything.
-
@Dashrender said in From thin clients to desktops… not the other way!:
I was wondering where the bottle neck for the OP is? Is it RAM, disk IO, network bandwidth.
Not sure one can suggest that you bail on VDI if you don't know the reason your VDI is slow.
For example. If your switch is bad, over saturated, then new PCs won't help everything.
I don't think our network is that bad, we have cisco sg500 and the bandwidth requirement for PCOIP is very low… maybe the storage can be the limiting resource, but sometimes… it just feel like the scheduler of the hypervisor have something better to do than serve our desktop VMs, and I haven't figured out how to obtain the performance of a real desktop. Maybe, just because the performance of an recent core i5 with plenty of ram and ssd are REALLY high by any (desktop computing) standards, so its hard to come close to that benchmark with a VDI.
-
I wonder if running DPACK for a month on both your VDI servers and say one desktop would allow you to see the differences.
-
@Francesco-Provino said in From thin clients to desktops… not the other way!:
Maybe, just because the performance of an recent core i5 with plenty of ram and ssd are REALLY high by any (desktop computing) standards, so its hard to come close to that benchmark with a VDI.
Well you can't really compare 4 year old servers with HDDs to a new PC with an SSD, that's just not a fair test.
How long have you been doing VDI?
Test for you - try connecting to the VDI from a PC instead of your Zero Clients. When I ran thin clients back in 2000 they were horrible! This was with Terminal Services instead of VDI, but the concept was similar - shared resources and all. The thin clients would flash a white screen whenever the session was switching between pages on a Java or Flash website. This made Lotus Domino Web Access (don't recall real name) nearly unusable for the client.
When remoting into a Terminal Server (non Citrix) from a P2, 256 MB RAM XP machine was an awesome experience for the end user, but the thin client was nearly useless, as I said. -
@Dashrender said in From thin clients to desktops… not the other way!:
@Francesco-Provino said in From thin clients to desktops… not the other way!:
Maybe, just because the performance of an recent core i5 with plenty of ram and ssd are REALLY high by any (desktop computing) standards, so its hard to come close to that benchmark with a VDI.
Well you can't really compare 4 year old servers with HDDs to a new PC with an SSD, that's just not a fair test.
How long have you been doing VDI?
Test for you - try connecting to the VDI from a PC instead of your Zero Clients. When I ran thin clients back in 2000 they were horrible! This was with Terminal Services instead of VDI, but the concept was similar - shared resources and all. The thin clients would flash a white screen whenever the session was switching between pages on a Java or Flash website. This made Lotus Domino Web Access (don't recall real name) nearly unusable for the client.
When remoting into a Terminal Server (non Citrix) from a P2, 256 MB RAM XP machine was an awesome experience for the end user, but the thin client was nearly useless, as I said.Hi @Dashrender, I've tried what you say, and there's no advantage in our configuration… we used praim p9002 as zero client, highly specialized devices with teradici processor. I think they deliver the best experience reganding VDI…
-
Are you sharing compute nodes with Servers? We don't do that and it goes against the best practices that VMWare has set out.
I meant to say earlier that we use 15K drives for the VM Image and SSDs for the replica image. We have bonded 40Gb (4 10Gb ports) to each compute node and redundant 16Gb/s FC to our storage. So our system probably is a bit over built for what we are doing.
-
@coliver said in From thin clients to desktops… not the other way!:
Are you sharing compute nodes with Servers? We don't do that and it goes against the best practices that VMWare has set out.
I meant to say earlier that we use 15K drives for the VM Image and SSDs for the replica image. We have bonded 40Gb (4 10Gb ports) to each compute node and redundant 16Gb/s FC to our storage. So our system probably is a bit over built for what we are doing.
No wonder you don't have performance issues
-
@coliver said in From thin clients to desktops… not the other way!:
@scottalanmiller said in From thin clients to desktops… not the other way!:
@coliver said in From thin clients to desktops… not the other way!:
@Dashrender said in From thin clients to desktops… not the other way!:
@coliver said in From thin clients to desktops… not the other way!:
Good luck we've got about 100 virtual desktops deployed and are looking to do another 100 next summer. We've got a big IBM SAN and a half dozen hefty servers doing the processing.
How many compute nodes do you have? Coliver
Six right now. We're haven't gotten close to maxing then out yet. That being said these machines perform at the physical desktop level but they aren't persistent, yet, so every time someone logs in they get a new desktop.
That's often what you want with VDI. Pristine desktops every time.
Yes, we're working on getting persistent storage, not VMs, setup for staff. Then we're looking at doing OneDrive on student machines mounted via WebDAV.
If you are able to figure out how to mount OneDrive via Webdav automagically, I'd be interested in knowing how you do it.
-
@Dashrender said in From thin clients to desktops… not the other way!:
@coliver said in From thin clients to desktops… not the other way!:
Are you sharing compute nodes with Servers? We don't do that and it goes against the best practices that VMWare has set out.
I meant to say earlier that we use 15K drives for the VM Image and SSDs for the replica image. We have bonded 40Gb (4 10Gb ports) to each compute node and redundant 16Gb/s FC to our storage. So our system probably is a bit over built for what we are doing.
No wonder you don't have performance issues
The students haven't started banging away on it yet, lol.
-
@dafyre said in From thin clients to desktops… not the other way!:
@Dashrender said in From thin clients to desktops… not the other way!:
@coliver said in From thin clients to desktops… not the other way!:
Are you sharing compute nodes with Servers? We don't do that and it goes against the best practices that VMWare has set out.
I meant to say earlier that we use 15K drives for the VM Image and SSDs for the replica image. We have bonded 40Gb (4 10Gb ports) to each compute node and redundant 16Gb/s FC to our storage. So our system probably is a bit over built for what we are doing.
No wonder you don't have performance issues
The students haven't started banging away on it yet, lol.
Sure they have, we've been in production for a few years now. We have three different labs already on it.
-
@coliver said in From thin clients to desktops… not the other way!:
@dafyre said in From thin clients to desktops… not the other way!:
@Dashrender said in From thin clients to desktops… not the other way!:
@coliver said in From thin clients to desktops… not the other way!:
Are you sharing compute nodes with Servers? We don't do that and it goes against the best practices that VMWare has set out.
I meant to say earlier that we use 15K drives for the VM Image and SSDs for the replica image. We have bonded 40Gb (4 10Gb ports) to each compute node and redundant 16Gb/s FC to our storage. So our system probably is a bit over built for what we are doing.
No wonder you don't have performance issues
The students haven't started banging away on it yet, lol.
Sure they have, we've been in production for a few years now. We have three different labs already on it.
I misread that, then. *-)
