TechAgility

So as things slowdown in the lead up to the break I have had some time to get back and review some twitter posts and other blog articles that I have been meaning to catch up on. I must admit that I tend not to use Twitter that much but do find it quite useful as pointers to additional material for research and recently posted material on the topics that I’m interested in.

So one article that really caught my eye was this from Ruben Spruijt’s post at Brian Madden, because for a number of years now it is the Hard Drive that has not kept pace with advancements in technology for the rest of the PC/Laptop components, and in particular I have noted a number of people getting very enthusiastic about the performance improvements they have actually noticed when using Laptops with SSD’s

Runcore SSD’s – promises to boost computer performance vs. the traditional 1.8" HDD by more than 400% with read/write speeds up to 75/40MB/sec.
Experience with 128GB 1.8" ZIF in HP 2710p – Read/write performance for 4K random files has improved over 500% (even under Bitlocker). it just became the fastest computer I ever used… Word and Excel launch in one or two seconds (compared to 30-60 before)…

Just how fast does the storage component of VDI have to be?
So it stands to reason that if you are going to Virtualize your Desktops (and by definition centralize?) then you are going to need to give some serious throughput for the Disk I/O or IOPS? So for anyone who is embarking on this then I would seriously recommend giving this article a through read? Understanding how storage design has a big impact on your VDI!

Possibly the most interesting point that Ruben did bring to my attention with this is that in VDI implementation the Disk I/O is predominately Writes and not Reads – I know this sounds contradictory in some respects, but do review this for yourself – in particular is a good comment from Claudio Rodrigues regarding the nature of how the O/S’s deal with this, as well as confirmation from Dan Feller at Citrix who confirms that most of the numbers and math are in sync with his findings. Ruben has also supplied a Sizing Tool to get a good idea of what might be needed?

So it also starts to make sense that you make sure that you focus very sharply on the XP or Windows 7 Template VM’s in the first place to reduce the I/O required as much as possible?
Citrix has some great tips in this document - Best Practices – Citrix XenDesktop with Citrix Provisioning Server

So as this brought up some interesting thoughts I also found this article relating to sizing and best practices, VMware View sizing & best practices
which followed on from this original article Virtual Infrastructure best practices and in essence this confirms that the storage requirements will need to focused somewhere around an 20/80 split on the Read/Write (20% read and 80% write)

While this table gives a quick representation of what sort of sizing requirements are out there please make sure you research this subject thoroughly, and read the post in full? J

In all fairness I must also confess at this point that my weakest area in IT is storage, and as yet I have not had a functioning Lab running either Citrix’s XenDesktop or Vmware’s View to be able to play around with this and get my hands dirty with either Provisioning Server or Composer. From what I can see (based on Marketing so far J) both of these products have the ability to drastically minimize the storage needed for VDI and as a consequence I am assuming that this may indeed have a flow on effect to mitigating the IOPS issue? (Can anyone comment or set me straight on this?)

I then went on to find some more details around the costing and financial side of VDI implementations and found this article of Doug Brown’s regarding Cost Savings of VDI: Is It Possible? as the diagram below points out, it is complex, and as such it could be that the regular ROT/TCO calculations are not highlighting all the potential benefits and savings?

Changing the ROI/TCO Calculation?
So this then brings me to another update in my Twitter was from Tyler at LiquidwareLabs.com with a new addition to the ROI/TCO debate,
COP – Coefficient of Productivity in VDI the New Math , in which he points out that you should also factor in any additional productivity gains and use this to dilute the true cost of the CAPEX.

Now I wonder how many VDI proposals are brave enough to start assigning values to what has traditionally been seen as "additional" benefits related to an architectural change? Don’t get me wrong, I’m not against it, but I can see the debate might start to get a bit heated in the boardroom? and in some respects maybe that’s exactly what’s needed – by assigning values and assumptions and then extrapolating them across an enterprise the resultant *potential* savings become too hard for the board to ignore?

Sure there will be debate, and rightly so, but in following this sort of process it should quickly become apparent that there is a technological shift under way where the concept of "My Desktop" still being something that can be pointed to in the Office sitting on a desk somewhere is rapidly becoming redundant?

Now having said all that – it’s interesting to note that Gartner’s Hype Cycle Report for 2009 seems not to have VDI listed? Maybe it’s hiding under something else here?

So will VDI make it to "mainstream"? AMD seems to think so, Taking Desktop Virtualization Mainstream, but also a lot of it will depend on a number of factors.

My main thoughts on this are:

To truly succeed VDI needs to be able to deliver a *High Fidelity* User experience equal to or better than what users have today – if what you are designing is unable to provide this then you should stop and re-evaluate now

If you can build a system that is able to provide this then users will want (no, Demand!) that they have it ASAP – you will know that you have succeeded when demand outstrips supply

To provide this *High Fidelity* User experience I’d suggest focusing on two key areas that may/or may not apply depending on your particular focus and what is considered in or out of scope for your project?

• Graphics
  
  • Bottom line – Graphics performance in this design CANNOT be measured, it has no metrics, or Best Practices. One persons "that’s fine" is another persons "there is absolutely no way I’ll put up with that", etc. This will always be a completely subjective assessment based on whoever is rendering the judgement.
  • You may think that you don’t need to focus too much on Graphics because "there isn’t much need or requirement for Video from the business" but I can almost bet that you would be wrong – if you are operating on this assumption be sure that you double-check with Stakeholders and get their agreement in writing?
  • So much content on the Internet is done in Flash today, it’s not just watching a CEO or Analyst briefing via YouTube, a lot of Web Based Training modules use the same or similar technologies, so even if you currently block YouTube and Facebook today it’s still quite likely that you need to factor this in to your design
  • Is RDP good enough? Check this assumption by giving targeted business users a Thin Client and removing their desktop to the lab and forcing them to access via the Thin Client – this is a simple but effective test that validates whether they truly need/demand High Fidelity, or will OK Fidelity do?
  • Do you need Citrix HDX? PCoIP? RGS? While I’m making the point about Remote Protocols – don’t forget to gather details of connecting accessories? Do you need to support serial or parallel?
• HDD or IOPS
  
  • Make sure the VM performs as well as possible while measuring the I/O on the VM itself? Check the disk queue, and ensure that it’s not the disk performance that’s holding this back – and make sure you have enough throughput to allow for Login’s etc.

So where does that leave us?
In the Graphics space there is already a number of solutions for VDI that have been out for some time (Citrix’s HDX) or have recently been released (Teradici’s software version of PCoIP). In addition to that there are other approaches if you need to support High Fidelity Users at the top end with HDX-3D taking advantage of Nvidias GPU’s that almost bring TeraFLOP performance to the Desktop.

A good introduction to CUDA – Just how powerful can GPU’s be? Back in mid-2008 the GT200 was able to deliver 1 TeraFLOP performance, the G80 supports 768 Threads per Core – on 128 Cores… Want a Personal Supercomputer? What would you use it for?

So I suspect that we are not far away from seeing vGPU’s being available to VM’s that will be able to deliver the performance on demand – the main thing is that you make sure you are either using a Remote Protocol stack that can provide this functionality – OR make sure you are using a Broker that supports a shift to one?

At the HDD level I’d suspect that as SSD prices come down and SSD longevity increases, along with improved methods of utilising this in some kind of shared caching mechanism in a VDI deployment we may well see some serious advances that enable VDI to overcome this IOPS roadblock?

With the monotonous regularity of Moore’s Law driving CPU performance up it should be possible to provide way more CPU power to users than they really need to enable this – but again – it’s quite possible that the Disk I/O that has the potential to spoil this party?

Please feel free to correct any of my ramblings?

written by dcaddick