vSphere Metro Storage Clusters

Lee Dilworth
Duncan Epping

HA protects against:
- host failure
- host network isolation
- permanent loss of datastore
- VM crashes
- guest OS crash / hang
HA is OS and application independent

HA Agent is now called Fault Domain Manager and moves to a single master, multiple slave model (from 5 master model).  In the event of loss of single master, slaves elect a new master.  Heartbeating is still over the network, but now uses the datastore as an additional heartbeat to monitor.  Network heartbeat is checked first, when this is lost the datastore heartbeat is checked.

vSphere Metro Storage Cluster (essentially a stretched storage cluster)
- hardware needs to be on the HCL
- HCLs for iSCSI, NFS and FC
- network is stretched across sites
- storage is stretched across sites
- essentially a federated HA solution
- single vCenter (which can be a VM in the stretched cluster of course, and this is a common use case)
- quicker than SRM to recover machines because it is essentially an HA activity
- max supported network latency is 10ms round trip time (RTT)
- vSphere Enterprise Plus licencing is required
- max supported storage replication is 5ms RTT
- good for relatively local failover across sync sites
- most storage hardware will support an async 3rd copy of the data
- uses DRS affinity groups per site to maintain multi-tier application performance using the "should" rules, not the "must" rules.

Capgemini Public and Hybrid Cloud - Service Brokering

Subtitle:  "Turning Days Into Hours"

 

Results of survey of 460 Execs:

- 45% of decisions to buy cloud services are in the business not in IT

- most claim to have a cloud strategy, most are being cautious adopters

- claiming 78% of new deployments are to cloud (CG did not clarify if this included internal cloud)

- want to move from CapEx to OpEx

- would like to move faster to cloud but secuirty, data sovereignty are main concerns

 

Service provision evolution:

- All internal IT services

- All IT services delivered by one external supplier outsourcing

- IT services delivered by multiple outsourcing

- Future will be micro sourcing of cloud services (according to CG)

Managing so many suppliers will be complex and not easy to control

 

 

Using a platform based on VMware vCD, vCAC, IT Business Manager (ITBM) and BMC software, CG are building what is essentially a brokerage service, or will be when they have completed their current work.  They are not building their own data centres, but partnering across mutiple service provider clouds.  GC are a member of the Open DC Alliance.  It will be pre-packaged "business solutions" catalogue across multiple application providers.  These will include intergration services across multiple IaaS and SaaS providers were necessary (and will, of course, mean lock-in to CG if no-one else is offering just the pre-packaged business service you need).

 

Customer will have choice of commercials and service levels driven by policy based provisioning.  Consolidated billing (with GC fees added of course) will be available across all the elements making up a business service.

 

It won't be quite brokerage yet, but CG see it moving that way, with trading of services like commodities.  At the moment there is no standard way of describing services and service level which makes commodity trading difficult until standards are reached. CG are using an "API factory" in India to build the interfaces to and from different service offerings.

The software in use will provide users with a dashboard of their services and costs and an administration view.  Also provides CG with their view of consumption, costs, margins, SLA performance etc.

 

 



Thoughts: Looks like a great idea to offer pre-packaged services.  From what we saw today it looks like there's still a little way to go in completing the development and getting a real service offering to market.  It has real potential, CG will need to focus on the largest market first to prove the model works, then they will have an extremely complex job building and maintaining all those interfaces and changes to those interfaces over time.  I hope they can make it work as this has potential to be game changing in the market.

Introduction to Network Virtualisation with NSX

Thomas Kraus
Eric Lopez

Straw poll of audience asking what type of cloud infrastructure is in place.  No VCD users in the room, but a few using OpenStack.

Service providers (internal or external) want to:
- deploy applications at scale with little pre-planning
- have mobility to move workloads across geographies and providers
- flexibility to create and add L3 - L7 services

Currently VMs are linked to physical network features such as ACLs, IP addresses, VLANs etc.  This means it is slow to make changes, placement of workloads is limited, there are hardware dependencies, operationally intensive and therefore expensive.

Treat NSX as though it is a network hypervisor equivalent to ESX - it places an abstraction layer between VMs and the physical network switches underneath.  Brings hardware independence, allows VMs to be moved around without changing the underlying physical host network config and brings the operational benefits of virtualisation.

In and NSX network every host server only has 1 IP and 1 MAC address per server.

The NSX controller planer is an HA cluster on x86 servers and is out of the data path.  Controls "thousands" of devices - but this was not quantified.

The NSX data plane is based on software switching in the host servers - vSwitch for VMware (NVS) and Open vSwitch for CloudStack and OpenStack (OVS).  Every host in the NSX environment needs a vSwitch of the correct flavour and works with vSphere, KVM and XenServer hypervisors.

NSX manager is a vAppliance

Physical servers are connected through an host running Open vSwitch NSX Gateway

NSX is a tunnelling overlay using STT, GRE and VXLAN.  It effectively adds another layer of L2 headers to each packet.  Jumbo frames are required.

High Power Graphics And Virtualization - Testing The Protocols

Alternative title:  How to Turn Water from Blue to Red

 

J. Gorzas

An excellent session on the vagaries of video protocols in a virtual environments.  Essentially the premise is the creation of a test environment that compares the various protocols against each other by capturing the results as a set of 4 videos playing next to each other.  Fascinating stuff that reveals that the answer to the question "which technology is best for remote high power graphics" is "it depends".  And it depends on many factors, not least of which is the applications you expect your users to be accessing most frequently.

 

Originally client side rendering was thought to perform best where all the info is sent to the client and the client needs to be able to interpret and render all possible formats, which was a problem for many Linux clients. 

Sender side rendering is when the rendering happens in the server (CPU or GPU, but always CPU in the early days) and the client only needs display the pixels.  This eases compatibility, but if there are many users then the CPU is quickly overwhelmed.

Historically VMware View prefers server side rendering, Citrix client side.  Now there is much more a blending of the 2, dependent on the format of the graphics, as pretty much all clients have at least H264 rendering capability.

 

Gorzas has built a testing protocol and rig to ensure testing consistency across different approaches. This has become what the industry relies upon and there is now a concern that the vendors are tuning their solutions to exceed in the tests rather than the real World.

 

Test rig:  Client is connected to server using an Apposite Link WAN simulator which allows the latency to be dialled up and down.  The client has Epiphian recording software installed and a very fast HDD is attached to the client via DVI or HDMI to capture the playback performance on the cleint device.  The outputs are then collated into a 4 image side-by-side comparison in one video capture to allow for the results to be compared.  Only production releases of software are used (no beta versions!).  A range of combinations of latency and packet loss are tested, up to 300ms / 1% packet loss. Example output:

The following types of GPU options are (or soon will be in the case of vDGA) be available for vSphere:

Throughout the discussion the terms VMware / PCoIP were interchanged freely, as were Microsoft / RDP and Citrix / HDX, with each solution being aligned to one vendor.

First to note is that RDP7 did not perform well under any circumstances other than a LAN.

For current formats (including RDP8), the results are VERY DEPENDENT ON THE PROTOCOLS AND CONNECTIVITY in use.  It's not really possible to say "x is better than y" because it very much depends on the application protocols in use.  Note that versions of protocols are important too. For example the latest version of HDX produces much better synchronisation between voice and video, but to achieve this it freezes occasionally to allow the catch up - so the video only element is now not as smooth as the previous version. 

On a LAN all the protocols will take as much bandwidth as they can get access to. (e.g. up to 80Mbps for PCoIP, but RDP more efficient at 50Mbps). 



As an example of the problems that can be encountered:

 

For a 2Mbps link with 50ms latency playing a QuickTime video:

- RDP7 is completely ineffective

- RDP8 is better

- HDX is smoother than PCoIP

 

For a 8Mbps WAN with 50ms latency playing Flash video:

- RDP7 fails completely

- RDP8 works reasonably

- HDX is better than PCoIP

 

For a 2Mbps WAN with 50ms latency playing Flash video:

- only HDX works, but only when the client has a Flash player

 

For a 2Mbps WAN with 50ms latency playing WPF:

- PCoIP steps in a controlled way

- HDX / RDP stutters

 

For a 2Mbps WAN with 50ms latency playing Fishbowl:

- PCoIP steps

- HDS is smooth

- RDP8 is very slow

- All formats change the water from blue to red:

 

For 2Mbps WAN with 50ms latency and running PowerPoint animation:

- PCoIP and HDX are good

- RDP8 is OK but not smooth

- RDP7 is very poor

 

For 2Mbps WAN with 50ms latency and delivering HD Video, best to worst peformance:

- HDX

- PCoIP

- RDP8

- RDP7

 

nVidia Cards:

Kepler K2 is 2 x high performance GPUs on one card - 1536 cpres per GPU

Kepler K1 is 4 x lower performance GPUs on one card

 

Recommendations:

- Designers and engineers need a dedicated GPU with pass through (vGDA)

- Power Users and Knowedge Users can use a shared virtualized GPU with API Intercept (vSGA)

- Task Users can use GPU emulation in CPU (sVGA)





 

 

Discussion:  Lync video testing - nothing to done to date, but looking into it.  Simulating users "is hard".

The Big Party

As usual VMware had laid on a party on the Wednesday night with all attendees invited.

The format is now familiar - a very large room, 5 or 6 bars serving complementary drinks, pool tables, pinball machines, retro arcade games and then something a bit different, which this year was the option to go roller-blading alongside the hired dancers.  Great mix that, roller-blading and free beer!

The bars offered plenty of choice - including cocktails mixed to order.  I stuck to a few beers, being mindful of the need to get to the 9am sessions the next day.  A big improvement, I thought, was keeping the party on site at the conference venue.  There was a 90 minute gap between the end of the sessions and the start of the party - ideal to ensure that folks filled their time by walking the solutions exchange booths, and not wasting time on travelling between venues.  The party was from 7pm to 10pm so there was no excuse for staying up too late!  Well, not a VMWORLD excuse anyway.  Food was burgers / wraps, sushi and puddings.  Plenty of it too.

Entertainment was provided by dancers on roller skates, stilts or unusually dressed in interesting costumes or with ipads for faces, all backed by tunes from a DJ.  First act was an all girl group called High on Heels that comprised DJ, vocalist, sax, violin and percussion.  They did a really good job of covering classic and contemporary R&B, dance, soul etc., from Emile Sande to Black Box via Chaka Khan.  Good stuff, and, for me at least, somewhat more talented that the main act.  They also had a guy on stilts, covered in lighting and resembling Iron Man firing fire extinguishers just for the sake of it.  Better than it sounds!

I lost my specs to Taio Cruz.  Not sure if that's entirely true, but Cruz was up as the headline act.  He sang to a backing track provided by a DJ.  He did all 6 songs but I confess to not really spotting much difference between them.  The crowd were hyped up by his presence, but at the beginning of pretty much every song I thought he was launching into Dynamite, but I was wrong until the last track.  Still, I think I was in the minority as the crowd were very much enjoying themselves.  There was a great part of the set when large illuminated balls were passed out over the crowd (these were balls, not balloons) and it was great fun to bat them around overhead.  Except when one of them hit me on the back of the head, removed my specs and that was the last I saw of them as they were trampled underfoot by 2000 bouncing geeks.  Ho-hum, the hazards of partying!  Made the rest of my trip a bit tricky though, must remember to carry spares in the future.

DELL Caterham F1 Simulator

The simulator had queues pretty much all week.

VMware Virtual SAN

First fully attended session I've attended, and it is being repeated several times during the week.

Scale out storage in the hypervisor natively, not a virtual appliance
HDD, SSD, but requires SSD to perform well
Behaves just like shared storage with all the usual sphere features
Can apply policy based storage management
Distributed RAID, no RAID of physical disks required
Perf and capacity scales linearly as you add nodes to the cluster
Writes to multiple vmdk copies for HA

When deploying a VM:
- vsan checks where there is space for the number of images neede to be created in the array, keeps images separated according to policy
- writes go to all images simultaneously to the SSDs. All writes must complete before it is acknowledged
- data move to HDD a sync later
- if replication is required it is copied from SSD or HDD from the fastest available location

vMotion
- no change, VM moves around the cluster
- if a vMotion cases a performance issue then storage location automatically moved across vSAN whilst maintaining policies

Fault Tolerance
- loss of host - HA will restart vm as normal, waits 30 mins to see if host comes back up, if not vSAN will start creating new copies in the background whilst managing network traffic
Need min 3 servers in cluster.
Min 1 SSD and 1 HDD per host
Recommend 10Gb network
Assumption that virtual networking is in use
Server needs to be on HCL

<Suggested Workloads
- VDI
- Dev / test
- DR target

Performance
- claiming vSAN can deliver same VDI density as all SSD array

Thoughts:
- Great for workloads where compute : storage ratio is fairly constant, e.g. VDI
- if it performs as claimed it will be a game changer for some workloads

VCloud Hybrid Service

Claiming this is a big a change as esx was. Essentially VMware is now a cloud service provider. Using co-lo facilities.
Deploying a VM is now straightforward, but adding the network, storage, firewall etc creates delay.
Extract, pool, automate is the mantra of the week
VMware services in 3 US data centres
UK available in Q1 2014
Focussing on virtual private cloud
5GHz CPU
20GB RAM
$2 per hour
10Mbps standard connection over secure tunnel or 1Gbps for virtual private cloud 10Gbps for dedicated cloud are available as dedicated lines
Data protection is an add-on service
DR as a service is available
Portal to make services available to business, including dynamic, status reports.
Shopping cart approach to purchase- multiple different pricing models
Data migration at start up is normally via physical NAS transfer.
For the hosting only you can manage using a browser. If you have existing comprehensive internal VMware environment then it is claimed to be easy to migrate vms to VMware vCloud Hybrid Services.
Thoughts: I remember asking VMware folks if they would be providing their own cloud services when they announced vCD - they said no as all the service providers would be buying vCD - this could be VMware's recognition that the service providers have largely chosen other routes?

Looks Like There's A Party Tonight

Now, where's my whistle and glowsticks?

Wednesday General Session

Much talk about the ability of vCAC and IT Business Management and its claimed ability to expose IT costs to the business in advance of consumption. Also claims that its heterogeneous, which remains to be seen what this means.

There are echoes of the previous 5 years of promises to move IT costs from 70% run to 70% investment. Virtualization was just the first step - as many of us knew at the time, vitualization solved a small part of the problem (space, power, hardware reduction, agility, hardware refresh etc.). Now all this can be automated. 

NSX
- in addition to vitalizing the network the intention is to make server to server traffic more direct and efficient
- can move existing VMs by using motion
- someone should advise US companies about how scripted conversations are far too obvious. Maybe it works in the US, but not very cool in EMEA

VSAN
- demo of how easy it is to deploy
- adding a host to a cluster adds the storage in that host
- policies allow copies of vmdks and writes go to all copies. No ,mention of performance impact
- for DR the data is replicated off site. No mention of recovery method - needs further investigated
- in beta now, no mention of GA date

User Self Service
- IT need control through policy, security and automation
- one click user access to apps regardless of how it is delivered including XenApp in Horizon workspace

VCops
- automation of responses to predicted events based on policies, eg add VMs to a service that is maxed out
- connections into many other vendors' management tools
- can move workloads across service levels using vMotion and Storage vMotion
- health of system exposed to business through vCAC

Thoughts: after a few years of more of the same and lots of acquisitions that sat alongside rather than as part of the suite, VMware seem to have started to make significant steps into new areas and better integration- . IF it delivers on the promises and IF the licensing makes sense. OpenStack probably has a better chance than Microsoft to respond quickly.
Personally, there are still a couple of elephants in the room - adding the automation tools adds more stuff to manage, change control and maintain, and is the infrastructure really the bottleneck in most IT projects? I can see that infrastructure can be a bottleneck on doing more or less of what is already being done, but for the development of genuinely new business IP implemented on IT, infrastructure is much less likely to be a project delay.

VRTX Stars On Greenbytes Booth

Greenbytes deliver a reference architecture for de-duped and accelerated IO and storage on Dell VRTX, aimed primarily at VDI

Dinner With QLogic and Co

DELL Data Centre Rack

High End Graphics VDI

Horizon 5.3 announced today delivers sVGA dedicated access to nVidia graphics cards.
nVidia GRID delivers gaming experience in the cloud.
Nvidia seem to be recommending pass thru for high end users.
SVGA allows sharing of the card across multiple users. Also available the RemoteFX. Limited by the generic vmware hypervisor graphics driver.
VGPU nVidia available soon for XenServer only.  Exposes the nVidia card to the OS so the standard driver can be used.

SANDISK Flashsoft

SANDISK demoing in server flash cache, accelerating reads by 2 to 4 times.

Demo runs on DELL R720 servers and PowerVault storage.

Making the Most of Hybrid Cloud

Will be interesting to see how much of this content admits to the existence of non-VMware clouds...
Starting with a level setting.

Making analogies with how containerisation of the shipping industry.
vApp is the standardized container.
These vApps run on vmware based clouds: Att, blue lock, colt, csc, loch, singhtel, t-systems are available providers with services across 24 countries

Application choice when moving to the cloude is important- data security and location to meet compliance, service levels required - crucial to provide right sizing for security and service level requirements as over specifying is an expensive mistake
the closer to commodity the application stack is, the easier it will be to move to the cloud, eg email, sharepoint, web, ecomm
Ensure splitting of components can still perform over a link to other components in the cloud
Costs to migrate
- understanding internal tco is complex
- need to add bandwidth costs
- migration effort
- easiest if you have vApps of course
- use feasibility analysis (vmware consultancy service) including internal costs for comparison

Business value?
- give devs quick access and less time building environments
- agility ( particularly in agile dev env or peaky workloads)
- potentially lower cost of ownership
- avoid the grey economy of cloud outside IT control (survey says somewhere between 37 and 58%, depending on who you ask suspect or know this happens in their business.

Thoughts - the need to work out what is right to provision via a cloud seems to be the obvious analysis to do before making a decision. I think what is lacking is a clear methodology for making these decisions. Need cloud centre of excellence to manage this, design solution, deliver solution and manage contracts.

Of course Dell Multi-Cloud Manager will help once you have decided what steps to take into the cloud.
And to answer my earlier question, apart from recognising that out of control cloud services exist, the solution is very VMware focussed. The principles of making decisions applies generically.

HPC Using Vitual

Originally HPC was on supercomputers
Now on commodity
Premise is that virtual is the next step

HPC preconceptions
- lowers performance
- slow and unwieldy
- etc

VSphere for HPC
- DRS and Storage DRS to balance workload
- direct IO
- in v5.5 flash cache
- center tops for insights into performance and health

VCD
- disperate workloads
- resources
- security

Bitbrains - NL Hosting Company
- using VCD to provide HPC cloud service - Microsoft HPC
- Dell servers
- IB connectivity
- IB connectivity for storage too
- fast storage
- CD gives customer flexibility pay as you go and bursting
- virtual private clouds in a public cloud
Subaru case study
- video rendering using hpc grids on virtual private cloud within 40 secomds
- nodes deployed as per job requirements, then closed down when they have been idle for a period of time agreed with customer
Insurance Case Study
- risk data
- dedicated lines to 3 data centres
- terminal servers
- distributed file servers
- vms for compute
- performance so good that the customer re-ran theasr 12 months of risk calcs in 2 weeks
Optimization
- perf bios settings
- disable interrupt coalescing on esx
- optimizing job scheduling and app

Benefits claimed
- no hw to buy
- don't need hpc infra skills
- pay as you go
- immediately available, much faster set up
- easy to scale down

You need
- collaboration with customer
- optimized apps
- fastest storage and connectivity available
- good capacity management

Debate
- still skepticism as based on Microsoft HPC - audience seem to think Linux is a more 'serious' HPC environment
- audience would liked to have seen evidence of direct comparison between physical and virtual performance

Thoughts: interesting in the use of cloud for the agility of delivering short term HPC to multiple customers.  Session was billed as proving HPC peforms better on virtual but this was not really addressed except to claim that cheaper cloud based HPC means you can add more nodes to add performance at low cost.

Solutions Exchange

VMware and DELL stands just gefore the solutions exchanged opened to attendees

Opening General Session

Main message is IT as a Service and serving the Global workforce.
More interestingly, compared to the messaging when VCD was released VMware's messaging has changed from multiple VMware clouds to just managing multiple clouds (but this isn't reflected in the creation of VMware Hybrid Cloud Services).

Sanjay Poonan on EUC
Claiming that VMware will reduce TCO.
VMware acquires Desktone to and portal management layer - mainly targetting service providers offering DaaS
Software Defined Data Center
Vitualize everything, not just servers
Automate processes
Compatible Hybrid Cloud
Bring program ability and policy-based approach to all of IT
Align storage to apps requirements
Virtualize network for agility and speed

CEO AtoS Thierry Breton
AtoS have adopted the SDDC
Manage largest number of DCs in EU
Intend to eradicate internal emails
Focus is now moving from cost reduction to value add in as and data. Infrastructure should just be a given.
Big Data will form a key part of value add.
TB now a member of EU Cloud advisory group helping Europe to bring standards to cloud services.

vsphere 5.5
More of everything - cores, memory, storage
GA at the end of this week
Leader in delivery of cloud foundry
Emphasising apps moving to scale out with less dependency on infrastructure resilience.
Software Defined Storage
VSAN distributed shared storage across servers for resilience and performance, simple mgmt
New backup capability - vshphere data protection advanced 5.5 much more compression and quicker recovery
Flash cache now available. Virt so now included in 5.5 functionality

Network
Programmatic provisioning
NSX released to GA today
Martine Casado VMware Networking
Networking follows compute
Equivalent to ESX on the network

Registration Time

Approach to the Conference Centre

 

Picked up the materials after registration.  The usual laptop rucksack with leaflets from various vendors.  Includes a chance to win Caterham F1 goodies on the simulator on the Dell booth #D208.

Neither the conference pass nor the materials provided include a physical copy of the agenda this year.  If you don't have a "device" you'll have to rely on the boards around the location.

 



Just as essential as the other kind of bar at a geekfest like VMWORLD