Virtualization, Cloud, Infrastructure and all that stuff in-between

I noted with interest the rumours today of Cisco ceasing a long-standing partner arrangement with HP (more here); essentially locking HP out of special pricing and insider product information and roadmaps.

For a long-time HP have offered Cisco components for their c-class blade chassis like the 3020 catalyst blade switch; to infrastructure people like myself this offered the best of both breeds (compute and networking) in a nicely integrated blade solution.

With HP’s recent acquisition on 3Com and their existing HP ProCurve range I would hazard a guess that they will stop selling Cisco blade switches in future – I also note from an email that all HP partners got this week that all Cisco manufactured blade switch components were facing supply issues, stoking the fires somewhat to resellers to push the HP product with some choice anti-Cisco FUD which I won’t repeat here.

Integrated is the key point for me, if you are building infrastructure solutions you want something that has been thoroughly tested together, even if bits come from different manufacturers and follow “industry standards” you can’t beat a pre-packaged solution with proven support processes and management toolsets.

To my mind this is why Apple have always been so much more successful in the consumer space than Microsoft have; aspiration brand arguments aside, Apple have always maintained a tight grip on the end-end quality of components that go into their products (software, firmware, hardware, accessories) whereas Microsoft were born out of the opposite*;

*let’s leave aside Microsoft’s later leverage and downright abuse of closed and proprietary formats and API’s to wield their market share weight – that’s a whole different argument..

Microsoft took the x86 PC out of the grip of it’s one-vendor supplier and inventor (IBM) in the 80’s and allowed it’s software to run on a wide variety of compatible, commodity hardware, providing API’s for thousands of independent software and accessory vendors to integrate their products with DOS and latterly the Windows OS.

Contrary to what many think (Apple included) Microsoft opened up the PC market to the masses by enabling competition – allowing the consumer to choose from devices that varied in quality and thus cost to match the needs and budget of the end-user – whilst Apple maintained a steady line of expensive/premium but excellent quality products.

In the longer-term Microsoft’s image suffered as a result of this “openness” because there was less (or no) quality control of what 3rd parties produced poor quality drivers, applications and hardware entered widespread use, and all that integration often ended up in a poor end-user experience; how many times have you been frustrated trying to get product X to work on PC Y with software from Z – you are relying on 3 different vendors, each with it’s own target market and user to all do their jobs properly, and to the standards they are given to work with (Win32 API etc.)

If you are a hardware company selling an optical mouse it’s a tight margin high-volume business so you don’t have a lot to spend on quality control or good coders – you want it “done” and shipped and get the profits in – or you may just be a fly by night company that wants to rip off consumers and make a quick buck, there is nothing really stopping you.

Microsoft took steps in recent years to address the poor-quality 3rd party stuff from tarnishing the mother OS’s image through the Windows Certified logo scheme, but they never took it as far as Apple; who refuse any code that is not blessed by their own quality control on devices like the iPhone – and look how it gets almost universal praise for it’s ease of use and reliability.

The ever increasing amount of crap-ware that PCs ship with is astounding – my new HP laptop ships with about 20 different 3rd party drivers and tools out of the HP box, Hawei 3G card, AuthenTec fingerprint scanner, ATI video card with it’s own software To boot, it would seem HP weren’t too diligent about their quality control either.

Now compare that to a MacBook Pro out of the box. Apple own the end-end process – you can only (legally) run THEIR OS X on it, sure they use things like Broadcom NICs and ATI video cards but because they control the OS as well as the underlying hardware they can ensure support is baked-in when it comes out of the box, as a result their test matricies are much simpler and the end-result is a much simpler and reliable end-user experience, they also reduce the bells and whistles to what people want (how many corporate IT departments really take advantage of the smartcard or biometrics support on modern PC hardware or the SD card readers, more often than not time is spent to disable such bells and whistles in the interest of support and not overwhelming end-users).

So, after that rather lengthy into back to my point – if you want something that is good quality and just works as a vendor you need a greater control of the whole “stack” – reduce the variance of components that make up each slice of the stack and the result is;

Simpler/smaller test matrix

Quicker time to market (less variance to test)

Simpler to support – less variance

Better scale of economies

smaller supplier portfolio to manage

There seems to be a fair bit of consolidation in the market place at the moment with the bigger players announcing grand partnerships or simply making massive acquisitions.

Maybe some of the open market, where you could run Linux, Microsoft or something else on your XYZ server is being replaced with a premium/preferred “stack” from a consortium of big-name vendors.

With all the talk of cloud computing this stack has gone well beyond the desktop, office or even corporate datacentre – You can buy the stack from a vendor consortium to run on-premise and offload some your infrastructure into their IaaS/PaaS/SaaS stack – with the promise of a more joined-up approach to support and services.

In the current landscape Microsoft are in an odd position; Ballmer is pushing them all the way into the cloud stack but at the lower levels they essentially have an open partner model, I would expect to see Microsoft partner with Dell (or even acquire, as has been rumoured for a while) to come up with a “premium”/integrated product line-up; but not entirely stop their traditional OEM arrangements with other hardware vendors.

Cisco would seem to have fired the first shot in this war at HP, maybe the gloves of co-operation will come off for the rest now that people don’t necessarily have to be seen to be playing nicely in the market.

The following table shows my take on the major players and the offerings they have; I’m a realist and a lot of these are recent acquisitions or products so I don’t honestly think in all cases they really do have fully integrated end-end sales, services and support but it looks to be the goal.

I am pleased to announce that I’ve been asked to present again at this month’s VMware London user group meeting.

I’ll be giving a follow-up on my previous Home Lab : vT.A.R.D.I.S session for which I’ll be bringing a (quieter!) demo environment which I have dubbed vTARDIS:Nano Edition showing how you can build a complex 10 node clustered ESX environment with dvSwitch, iSCSI SAN and vMotion etc. on a single physical host for study and play.

I’ll also be presenting a case study for a project I worked on virtualizing a large Microsoft Windows 2003 Terminal Server estate and will be showing some interesting statistics that we gathered along the way.

I understand the event is now fully-booked which is a great turn out!

The agenda as as follows

1100 – 1200 (Optional) PowerCLI / Powershell workshop – Alan Renouf. Please bring your own curly brackets.
12:30 – 13:00 Arrive & Refreshments
13:00 – 13:20 Welcome & News – Alaric Davies
13:20 – 14:00 Sponsor Presentation – Chris Hammans, Pano Logic
Real world vSphere deployment experiences – Stuart Thompson
(Mostly) Zero downtime DC migration for Dummies – Jonathan Medd
15:00 – 15:20 Refreshment break
ESX home lab update, virtualizing Terminal Server workloads – Simon Gallagher
Thin provisioning and capacity planning in a virtual world – Chris Evans, ‘The Storage Architect’
Bringing the Cloud Down to Earth – Stuart Radnidge, vinternals
16:45 – 17:00 Close
17:00 – Pub

In future, to register your interest in attending, please send an email to londonvmug@yahoo.com with up to two named attendees from your organisation. If you do not receive a confirmation mail, please don’t just turn up since we will not be able to admit you to the meeting.

Content from the meetings will continue to be uploaded to www.box.net/londonug, NDA permitting.

Following on from Eric’s post on running VMs from SSD’s at this page, and my previous experiments at using SSD’s to run VMs I thought I would post up my initial (non-scientific) findings from the FusionIO card that I have been loaned.

FusionIO make solid state storage cards that come packaged as PCIe x8 format expansion cards, they make use of multi-level cell (MLC) NAND storage to create amazingly high speed direct-attach storage, the Duo640 device I am working with is the mid-range offering, at the higher end forthcoming versions can support up to 1TB/sec throughput. the Woz is also their chief scientist 🙂

In my test rig I am using it and the Starwind software on Windows 2008 R2 to share the FusionIO card over iSCSI to a couple of vSphere 4 hosts – in this initial test I’m just using a single GbE NIC in both the server and the vSphere client – as you’ll see from the screenshot below it can actually max out the GbE connection in these hosts if you push it with several concurrent VM cloning sessions – so there is plenty more performance to be had out of the card and high levels of concurrency, in this case the NIC was the bottleneck.

The FusionIO duo card comes with 2 banks of 320Gb of memory (640Gb in total), in my initial configurations it’s not configured to RAID across the 2 banks, but that is possible to improve performance and fault-tolerance.

The FusionIO card doesn’t yet have drivers for vSphere but they are working on them – so you can’t directly access it from an ESX host yet so I am connecting to it using the Starwind software iSCSI target software.

One issue I found with my Starwind configuration is that the Starwind software is’t able to see the FusionIO card as raw block storage like it can with normal direct attached storage (SSD/SATA HDD etc.) although it is visible to Windows disk manager as a normal disk. so to get it to work I had to format and mount the FusionIO “disks” as NTFS drives under Windows 2008 R2 Disk Manager and create a virtual disk files in these drives using the Virtual Disk feature of the Starwind software – this is then accessible both directly to the Windows 2008 host and to my ESX hosts via iSCSI.

So, based on the same software that Eric used in his post these are the out of the box numbers the FusionIO card gets – there is still significant scope for fine-tuning to increase performance – But it’s pretty impressive.

FusionIO (non-RAID configuration) inside a VM over iSCSI

FusionIO (non-RAID configuration) – direct attached to a Windows 2008 R2 host

Consumer-grade SSD – direct attached to a Windows 2008 R2 host

This is using the following SSD which I purchased last year for under £200.

In conclusion…

The FusionIO cards aren’t cheap storage; they are in the ‘000’s of £ price-range, but they are FAST! with solid state storage pricing coming down in 2010 and when combined with iSCSI target software like Starwind it’s an excellent way to build a very high performance solid state SAN using DAS technology without the enterprise SSD SAN price-range and FC/network interconnects.

When vSphere drivers become available I can see some excellent 2-node/p2p replicating cluster/vSAN configurations using either Starwind or HP Lefthand Networks vSAN VM appliances as shown below, removing the dependency on single shared storage is a great design goal.

Disclosure: FusionIO and their UK distributor have kindly lent me a 640Gb duo card to work with, I have received no financial compensation nor have they imposed any copy approval or conditions with regards to what I write about their device – it’s just great and I’m that impressed.