My conclusions on VDI and other things

[superatticman]

Out of interest, what has been the level of support that IT helpdesks have performed on VDI installations?

 

Have you been expected to install, configure and support the whole lot?

 

Did you get specialists in to install and configure this and then handed over the day to day support to you with them available to troubleshoot any problems?

 

Or any other variation of the above?

 

My reason for asking this is that whatever your opinions on VDI are, have you not noticed that you have had to increase your skill set dramatically to support a VDI installation? Has this resulted in an increased salary or were you just expected to add it to your daily duties? Was any training provided to you?

[pcstru]

1) I don' t understand the arguments about stuff having to be physical. Why does SIMS have to be physical, DC physical etc? Is it performance? Do you really think in a school environment (we have 1450 students and 180 staff) that SIMS will be pushed to the absolute physical (excuse the pun) limit? I have sat and watched SIMS over many hours (years), IOPS, network, processor, memory and cannot figure out why it has to be physical? Please someone enlighten me!

I suspect it's mostly about support. When they are diagnosing performance problems (caused primarily by shoddy code and database design), they don't want to be debugging a dodgy VM set-up. So a physical server with it's own dedicated resources is the simplest way to decouple from that.

 

2) Why do people think Virtual infrastructures are a bottleneck and try to scale them to be able to run the equivalent of NASA? I read on here that someone is suggesting using 10Gb connections? What the hell for!!?

 

I'm putting in 10GB connections from core to edge because we need to replace cabling and I expect the investment to run 5-10 years. I don't expect to be using the available bandwidth any time soon, but nor can I easily forecast bandwidth use over those kind of time scales.

 

On VDI, I'm looking at it as a means of providing remote access for students rather than having thin clients replace cheap desktop PC's. I've steered clear of it so far because the promised cost savings just don't add up and if a consultancy can't even do basic arithmetic, what hope have they configuring a VDI solution!

 

Some great posts in this thread. Thanks to all participants.

[jayemm]

 

1) I don' t understand the arguments about stuff having to be physical. Why does SIMS have to be physical, DC physical etc? Is it performance? Do you really think in a school environment (we have 1450 students and 180 staff) that SIMS will be pushed to the absolute physical (excuse the pun) limit? I have sat and watched SIMS over many hours (years), IOPS, network, processor, memory and cannot figure out why it has to be physical? Please someone enlighten me!

 

There is no need (now) to have a physical anything. This is down to changes and advances in Virtualisation software/OSs and applications. Remember when it was advised that ISA server should be a physical server and that you should have a physical DC. As things change people or specifically IT techs are reluctant to change because of our 'KISS' and 'if its not broken dont fixt it' outlook on things. Hence if a physical SIMS and DC works... why change? I am replacing our SIMS server as we speak and that new one is going to be a physical server, but only because I want it to be, not because it has to be. Personal preference. Capita now accept that a Virtual SIMS server should have no issues, but they still have documents which say, and I agree, that a poorly configured virtual infrastructure can be worse than an older physical server.

 

As mentioned I think it all depends on IT tech expertise, skills, confidence and the equipment you have and how it is configured.

 

2) Why do people think Virtual infrastructures are a bottleneck and try to scale them to be able to run the equivalent of NASA? I read on here that someone is suggesting using 10Gb connections? What the hell for!!? Now I've got 8Gb fibre which I have come to realise is actually stupid, I could have saved money and stayed with 4Gb. It never uses more than 1.5Gb even when running Veeam backups over fibre. If this is for networking rather than attached storage (iScSI) then this is even more insane. As an example, I normally have 4 x 1Gb NICs teamed for host connectivity so I disconnected 3 of these from each host and ran them with essentially just one NIC each for a week just to see what would happen and you know what? no difference. The HP switch monitoring software never registered above 60% on any of the NIC ports at any time even with 720 workstations accessing work at the beginning (read) and at the end (write).

 

We have 10GB fibre and two ESX hosts each with 8 x 1GBNICs, maybe overkill, maybe just that we were upgrading every single switch in school installing a new managed wireless network and deploying over 1200 iPads and 100 macbooks ontop of our 800 desktops.

 

We dont use anywhere near the bandwith we have available and we stream media, have over 1200 ipads, use VoIP and everything else you can expect in a school environment but as always we like to provide resilience or over-spec for future proofing or sometimes buying the best we can now to mitigate not getting funding in the future. I am sure we are all guilty of this. I am contemplating upgrading to a 20GB backbone to protect against a single point of failure.

 

I feel it is now appropriate to say I work in the same LA as @Dave_O and know him very well. He is not always right and he will admit this, he is not always confrontational or arrogant as he may come across. I like to think of it as more critical friend like!

 

JayEmm

[Dave_O]

I am always right!

Don't mistake confrontation for someone who questions.

Arrogance is subjective, depending on perspective and personal understanding.

[jayemm]

Arrogance is subjective, depending on perspective and personal understanding.

 

Hence why I said "comes across" lol

[Arthur]

This is quite an interesting article on the subject of VDI performance...

 

Comparing the CPU Performance of Physical and Virtual PCs (VDI) « Helge Klein

 

When you move users from a physical PC to a VDI environment you may find that they are not too happy with their new machine’s performance – it happened to me. To quantify things I took a series of measurements comparing the old PCs we migrated away from with both VDI machines and the new PCs available to some.

 

Closing Thoughts

When moving power users (a.k.a. knowledge workers) from a PC to a VDI machine CPU performance is a topic that needs as much attention as IO performance. In your project, do not rely on benchmarks alone – those are always synthetic and may or may not match what you see in reality – but have real humans test the applications they use in the way they use them. I have seen “harmless” Excel sheets turn out to be massive CPU hogs running for hours or even days. Differences in performance were noticed immediately by the users.

[seawolf]

 

2) Why do people think Virtual infrastructures are a bottleneck and try to scale them to be able to run the equivalent of NASA? I read on here that someone is suggesting using 10Gb connections? What the hell for!!? Now I've got 8Gb fibre which I have come to realise is actually stupid, I could have saved money and stayed with 4Gb. It never uses more than 1.5Gb even when running Veeam backups over fibre. If this is for networking rather than attached storage (iScSI) then this is even more insane. As an example, I normally have 4 x 1Gb NICs teamed for host connectivity so I disconnected 3 of these from each host and ran them with essentially just one NIC each for a week just to see what would happen and you know what? no difference. The HP switch monitoring software never registered above 60% on any of the NIC ports at any time even with 720 workstations accessing work at the beginning (read) and at the end (write)./QUOTE]

 

The reason for suggesting 10GbE for virtual infrastructure primarily has to do with iSCSI and the significantly reduced latencies of 10GbE compared to Gigabit. It's the matter of typical 2-10 microseconds latency vs 25-150 microseconds for Gigabit (http://storage.dpie.com/downloads/solarflare/Solarflare_10GbE_Value_Proposition.pdf).The increased throughput is secondary to the importance of reduced latencies in most (non-SSD storage) virtualised environments. This would affect your write and read latencies in particular and overall responsiveness. Pair 10GbE with an SSD or hybrid storage solution and you will have fantastic performance. Pair SSD with Gigabit and the performance is very good, but you will not be able to get anywhere close to maximising the capabilities (or value) of your backend storage.

 

Regarding VDI, my only experience with it has been with Sun VDI and SGD. These solutions work fantastically, and were always heads and shoulders above the rest being one of the most mature and best technologies (since 1999). However, Oracle have recently seen fit to kill off the Sun Ray product line, so I can't even recommend them anymore. But, unless someone buys the technology off of Oracle and resurrects it, I probably wouldn't touch any VDI solution with confidence. Certainly not VMware's solutions. They make very good server virtualisation products, but VDI is pretty rough...

 

Enjoyed your original comments and advice, much if it was spot on.

[seawolf]

"I'm going to play devil's advocate for a bit.."

Why would you have communication between SAN and host dependent on name resolution? That to me is a poor design decision. We had one or two power cuts early on and learnt this the hard way. Why is this a case for having a physical DC? With the advent of Server 2012 restoring a virtual DC is possible (no AD count number issues). We have the luxury of having a second virtual setup in a physically seperate location (another block) in which we host a third DC acts just like a physical one.

 

I have to admit that I do agree with the logic of having a physical DNS server around. Maybe I'm just paranoid, but I have gotten caught with my pants down finding out that systems that should have no problem with DNS being down actually having lots of problems with it being down! Lets just say that unless you test EVERYTHING whilst DNS is down then it is a risk. Small? Probably. One I'm willing to take again? Not a chance.

 

"VDI can be absolutely awesome if it's put in the right place."

 

So where is this place? Give me a use case that only VDI can fullfil?

 

1. Defence

2. Banks

3. Hospitals

4. Intel Agencies

 

None of these should really be using anything OTHER than a VDI solution. In fact, that's what many of them are using. Government agencies and the military are the biggest customers Oracle has. Many of them are now spitting their dummies over Oracle killing the Sun Ray.

 

Education is not on the above list because education isn't one of those use cases where VDI is the only answer. In most cases, it's not even the best answer in an Educational environment. VERY large deployments (10,000+) though - THEN it might be the only viable and cost effective solution.

[gshaw]

VDI up until now has really been hamstrung by the storage back-ends not being up to the job - hard disk based is horribly inefficient if you want to match the performance of any half-decent desktop. Seems like there's some new ways of doing things coming through using flash \ RAM now and if those work as well as suggested it might finally be possible to give a user experience that's more positive than the equivalent desktop

[PotNoodleTech]

Just replaced one IT suite of thin clients with Dell Optiplex 7010 fatties.

 

It will be the first of many over the next few years.

[seawolf]

Here is a short, but good explanation of why 10GbE isn't just about bandwidth and throughput. There are major differences between the 1GbE and 10GbE protocols themselves. Its not just about speed. But, actually I do "have a need...a need for speed!"

 

http://www.rtcmagazine.com/articles/view/100797