GeekSlate

var _gaq = _gaq || []; _gaq.push(['_setAccount', 'UA-24967127-1']); _gaq.push(['_trackPageview']); (function() { var ga = document.createElement('script'); ga.type = 'text/javascript'; ga.async = true; ga.src = ('https:' == document.location.protocol ? 'https://ssl' : 'http://www') + '.google-analytics.com/ga.js'; var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(ga, s); })();

[geek-slate]

[geek-slate]

Go on... Geek out.

[geek-slate]

Related the the RFP Link from yesterday.

[jstaffin]

[geek-slate]

I’ve been preaching this position for years, after coming from a customer and moving into the partner space it became immediately obvious to me that for most customers, an RFP process ensures they aren’t even getting close to the best possible solution for their requirements.

[geek-slate]

This is pushing the limits a bit, I'm not sure we have many customers that need 40GE anytime in the near future.  But then again I probably was telling myself that when we started implementing 1GE, and then 10GE

[geek-slate]

[geek-slate]

VMworld 2011 - Day 4

After waking up very early to follow up on some items left unfinished from the day before, I met up with Nirvanix to learn a little bit more about their product and how it might fit within our solution offerings. Nirvanix provides an enterprise grade cloud storage platform primarily targeted for use as a backup and archival destination. Both Commvault Simpana and Symantec Netbackup have direct built-in integration for sending protected data to the Nirvanix Cloud. Deployment looks simple and the cost structure is based solely on consumed GB/month.

This morning’s general session was a thought provoking discussion of the human mind and our perceptions of reality. From the Program guide:

The Brain, The Mind and Consciousness Is free choice an illusion? Can the essence of social networking be found in a child’s first words? Is a single type of brain cell responsible for all human civilization? Join VMware CMO Rick Jackson and groundbreaking researchers V.S. Ramachandran, David Eagleman and Deb Roy as they explore recent, remarkable advances in understanding the human mind. You will never think about how you think the same way again.

As usual the crowd thins out significantly in the morning of the fourth day of these conventions, this might be related to the previous evenings activities

Sessions

BCO2874 - vSphere High Availability 5.0 and SMP Fault Tolerance – Technical Overview an…  High Availability has been completely re-written for vSphere 5 and this session covers a lot of the technical detail on how and why it’s much improved over the version that shipped with 4.x. The greatest improvements are around resiliency and troubleshooting ability. There was a technology demo of a pre-release version of VMware FT which supports multi-vCPU VMs.

VSP2454 - Scaling Your Cloud with Cisco Nexus 1000V and VMware vCloud Director  I mentioned VXLAN in a previous post and this session provided details on the implementation of this new technology with the Cisco Nexus 1000v and integration with vCloud Director 1.5

LAS4001 - VMworld Labs User Automation and Workflow Architecture The VMware Hands-On Labs environment is an amazing piece of technological achievement and pushes the limits to the extreme of the hardware and software that it runs on. This environment this week provisioned and destroyed about 160,000 virtual machines used to provide isolated lab environments for show attendees. Discussions of the challenges of building the architecture and provisioning systems that make this work were very interesting and provided insight into the challenges of pulling something off at this scale.

[geek-slate]

VMworld 2011 - Day 3

After the third night in a row of less than 4 hours of sleep, this morning was a little rough. Despite being a bit slow to start, and that start being at 6 AM for project update meetings and planning calls, I spent most of the day in break-out sessions once again learning more details about what’s new with the vSphere 5 components. That is when I wasn’t being pulled away dealing with client issues.

I managed to get some time to break way and visit the VMware booth within the world of solutions expo floor. My particular interest was VXLAN which is a new technology announced here at VMworld 2001 which introduces a new way to provide for isolated virtual networks in a cloud environment. I discuss VXLAN in more detail here

Sessions

VSP2376 - Performance and Scalability Enhancements in VMware vStorage VMFS 5
  This session covered significant changes to the VMFS storage architecture introduces with vSphere 5. Among the greatest enhancements include support for up to 64TB volumes, support for larger than 2TB Raw Device Mappings and improved support for ATS array functions utilizing VAAI on supported arrays to reduce the performance impact on file locking operations

VSP2347 - What’s New with VMware vSphere 5.0  Given my heavy coverage of “What’s New” Sessions this year at VMworld, this session would have been better off early in my schedule. I actually left this session early to head into the world of solutions to dig deeper into VXLAN, which was mentioned off the cuff early on in the presentation.

VSP2360 - VMware vCenter 5.0: What’s New, What’s Cool  There are plenty of things new with vCenter with this years release of 5.0, the item which will have the most visibility to our clients is the new Web Client. VMware has made it clear that while the current presentation of the new web client is not yet feature complete with the full installable client, this is where they are going and eventually the new Web client will be the only client. Other significant features include Auto-Deploy and a complete re-implementation of HA capabilities.

VMworld 2011 Party

Each year as part of the event, VMware throws a party for VMworld attendees. This year the party was held within the convention center and the headlining entertainment was The Killers, the opening act was a group called Recycled Percussion.

The Killers are one of my personal favorite bands and they did not disappoint. The show was loud, full of energy and the crowd seemed to be really energized. One of the reason’s I enjoy groups like The Killers is that they perform with so much emotion and enthusiasm live. There’s nothing worse than the disappointment of seeing one of your favorite bands play live, only to find out that they no longer care about the music they are playing.

After The Killers finished their set, the entire crowd was led by mermaids through the halls of the Sands Convention center, on through the Venetian, then the Palazzo casino and up the elevators to the pool deck where there was a relaxed pool party waiting.

While not as elaborate and diverse as some of the VMworld parties in the past, the Killers really made the party this time.

[geek-slate]

VXLAN - Networking for the Cloud Era

This weed at VMworld 2011, Cisco and VMware jointly announced VXLAN, a new virtual networking technology designed to address deploying scalable, access anywhere networking underneath private and public clouds.

VXLAN provides for the creation of multiple isolated network segments to support cloud usage scenarios. Each of these networks exists within the scope of a virtual distributed switch. The VXLAN networks are encapsulated and traffic between the hosts in the vDS is carried over the physical transport network as UDP packets. Multi-Destination frames utilize IP multicast for delivery to all hosts in the vDS.

VXLAN UDP Frame Format

VXLAN allows for layer 2 extension of virtual machines running on hosts, even when the hosts are connected to different layer 3 subnets. This de-couples the requirements of the application and virtual machines being deployed from the underlying network architecture of the datacenter.

Datacenter network architects have bristled at the request by the virtual infrastructure administrator that all layer 2 networks be spanned everywhere within the datacenter. This request allows any VM to exist on any host and provides for significant flexibility in deploying virtual machines. The problem with this approach is that it imposes performance and availability risk to the network.

With VXLAN, the network architect can design for reduced layer 2 domains again, just as they had prior to the advent of pervasive virtualization and allow VXLAN take care of providing the connectivity between the VMs

Example of VXLAN providing connectivity between hosts on different subnets

Cisco has submitted the VXLAN technology as a draft to the IEEE and there is a consortium of networking and software vendors working with them to move the standard forward towards ratification.

This new standardized and scalable approach to isolated cloud networking is refreshing to see given the scalability limitations of VMware’s previous approach to solving this problem within their vCloud Director product, vCD-NI

[geek-slate]

VMworld 2011 - Day 2

This pretty much sums up what my view was like for most of the day today.

My day was split between two competing forces, my scheduled sessions and scheduled meetings regarding client engagements unrelated to VMworld. This has left me with only limited opportunity to visit the world of solutions or hands-on labs. I did however have a moment to get a demo of CITIES by the Cisco Intelligent Automation product manager within the Cisco Booth. This is a solution that Cisco uses internally to provide self-service provisioning of private-cloud based resources to consumers within the Cisco Organization. Their primary use of the portal to-date is for a lab environment for their engineers and to provide a platform to run demo software for their sales organization. NewScale provides the front end web experience that the end-user interacts with and Tidal enterprise orchestrator provides the workflow behind it all to make it happen at the click of a button. The use cases presented align nearly perfectly with what we would like to provide internally to our engineering and sales organization and I will be looking into NewScale and Tidal further after the show.

I also had the opportunity to have a discussion with a vendor that provides a unique solution in the market that I believe has a place in some of our designs. We have had customers in the past very excited about VMware’s FT technology and were interested in protecting their critical applications with it, until the learned of the limitations. Stratus Technologies essentially provides a hardware solution that does what FT and vLockstep does at the VM level, but inhardware. This means you can have an entire ESX host protected by an FT like technology but without the scale limitations of doing it in software. One of the reasons I attend events like this is to get in front of technologies that I would not otherwise be exposed to, and this solution is a perfect example.

Sessions

CIM1264 - Private VMware vCloud Architecture Technical Deep Dive  On deck for the coming six months is a project to utilize a portion of our internal systems resources in a managed lab environment to provide engineers and sales resources with the capability to test and/or demo. vCloud director will likely end up being a part of the overall architecture needed to achieve that goal, along with other automation tools. This session provided a technical look at the components that make up a vCloud deployment.

VSP2122 - VMware vMotion in VMware vSphere 5.0: Architecture, Performance and Best Practices  The most significant change made with vSphere 5 in relation to vMotion is the way the hypervisor handles what is called pre-copy convergence failure, this is where the VM is updating RAM faster than it can be copied to the destination host in a vMotion session. Instead of blindly moving forward with the switch-over and expecting the destination host to catch up by requesting dirty pages of ram from the source host, the new version of vSphere actually throttles the VM activity down just enough to ensure that the transfer of RAM updates is happening faster than the VM is updating it’s ram. This ensures that there is never a pre-copy convergence failure and providing for a much better switch-over experience for the high-utilization VMs in your environment. There were a variety of other factors which provide for significantly improved vMotion performance discussed as well.

VSP3255 - VMware Storage vMotion Deep Dive and Best Practices  This session discussed the significant changes made to the way storage vMotion occurs on vSphere 5. In the past Storage vMotion was handled in much the same way as RAM was copied during a normal vMotion event. In the new version the hypervisor prepares the svMotion session and then begins by mirroring all writes to both destinations. Bulk copies of unmodified data happen while this mirrored IO is occurring and as soon as both source and destination are synchronized, the source is simply killed off. This provides for a much more deterministic performance impact to the VM during the move. Switchover times are on the order of 100ms even for very heavily loaded VMs. Other topics discussed were new support for linked clones and moving VMs with active snapshots.

CIM2452 - VMware vCenter Operations Technical Deepdive  This is the one product within the VMware portfolio that I have been meaning to look into in-depth and for some reason have not found the time to do so. This session provided an excellent detailed overview of the capabilities of vCenter Operations and how it does what it does (hint; it looks like magic, but it’s not) I see this being a very good fit in many of our client’s environments and I look forward to getting back into our environment and evaluating it.

After the sessions and meetings today it was good to relax a bit with a colleague and a client of ours who is also attending this year’s event

[geek-slate]

VMworld 2011 - Day 1

VMworld is being held at the Ventian Sands expo center in Las Vegas this week. I have attended VMworld twice before, both times in San Francisco, so this year it’s nice to have a bit of change. There are over 19,000 attendees on-site and some of the areas feel a little crowded, but overall this is a well run event.

A view of the main hallway looking down towards the registration hall

The “Hang Space” just outside of the expo floor

The blogger lounge outside of the expo floor

Today’s Sessions

VSP3305 - Upgrading to VMware ESXi 5.0

Provided a good overview of the upgrade process from ESX and ESXi to the new ESXi version 5. Considerations for what to do prior and after the upgrade itself were also discussed

VSP1823 - VMware Storage Distributed Resource Scheduler

This new feature in vSphere 5 will be a true game changer for our clients. VMware has removed most of the pain from the chore of managing the storage that a virtual infrastructure consumes. This session covers what Storage DRS is, how it works, and how to use it in your environment. At the end of the session there was a brief discussion on interoperability with other products and storage array features.

VSP3116 - VMware vSphere 5.0 Resource Management Deep Dive

This session didn’t provide much value to me personally as I have been helping our customers learn the ins and outs of this particular topic for years now, but if you are looking for a detailed overview of how to use DRS to manage the CPU and RAM resources within your vSphere cluster more efficiently then this is the session for you. I really can’t see why the 5.0 is even in the title here as I don’t think anything has changed since 4.0

VSP3299 - Using VMware vSphere Storage Appliance to Create Shared Storage from Local Storage

The new storage appliance product is exciting as it lowers the entry point for the very small customer and provides for a high-availability design at an even further reduced cost than we have had available in the past. This session covered an overview of what the product does, how it’s installed and configured. There was a good amount of time spent on the performance characteristics of the datastores provided as well.

Hands-On Labs

I have only had the opportunity to sit for a single lab so far, and the environment seemed to perform well enough for my session. I ran through the vSphere Auto-Deploy lab which includes configuration of the new PXE boot based stateless ESX provisioning method provided in vSphere 5. I’ve been told that the lab environment has been experiencing some periods of trouble throughout the day. For those who are not familiar, the VMworld hands-on lab environment is a 400+ seat lab with dynamic access to labs covering solutions and products from VMware’s product portfolio. Over the course of the coming week, this lab environment will provision and decommission over 200,000 virtual machines as part of the operations of this lab

Waiting in line for the Lab

General Session

First, the opening sequence of this presentation was mind-blowing, very impressive animation and motion across the entire width of the stage. Must’ve been expensive :)

Paul Maritz, CEO provided his vision for where computing is headed across the industry and spoke a bit about their effort in cloud, application technologies and even teased some stuff about vSphere 5.1 for next years event.

Solutions Exchange

The show floor was exceptionally crowded during this evening’s welcome reception. Otherwise, similar to other years. Plenty of beverages to drink, food to be eaten and swag to be acquired.

[geek-slate]

vSphere Storage Appliance lowers entry point for smaller customers

I’m preparing for the first day of VMworld 2011 and one of the sessions i’m most interested in for tomorrow discusses the newly announced vSphere Storage Appliance.

Given the ever increasing compute density of modern x86 servers, many of our clients fit into a scaled down architecture where two to three powerful hosts will provide them with more resource than they can reasonably expect to consume. For these clients the largest hurdle to enterprise grade virtualization is the requirement for a powerful and sometimes costly shared storage array.

The vSphere Storage Appliance is designed to provide high-availability storage to exactly this type of small-scale architecture. The storage appliance provides storage to virtual machines not from a shares storage array connected to vSphere hosts, but from disks installed locally within each host in the cluster.

Key Benefits

Five-Click Simplicity

High-Availability with no shared storage hardware required

World-Class datacenter capabilities for small environments

The VSA supports up to three hosts in a cluster and currently has some limitations including the inability to change the disk allocation and layout after initial setup, an issue which VMware states will be addressed in a future update.

If this product plays out as I expect within our designs, it will only serve to lower the entry point for our smaller clients and further remove any barriers that may be left for them to move towards virtualizing their environments.

[geek-slate]

VMworld 2011 Schedule

I will be attending VMworld this year in Las Vegas, August 29th through September 1st.  This is my current schedule

Monday

VSP3305 - Upgrading to VMware ESXi 5.0 VSP1823 - VMware Storage Distributed Resource Scheduler VSP3116 - VMware vSphere 5.0 Resource Management Deep Dive VSP3299 - Using VMware vSphere Storage Appliance to Create Shared Storage from Local Storage SPO3040 - Best Practices for Deploying VMware* vSphere 5.0 Using 10Gb Ethernet 

Tuesday

CIM1264 - Private VMware vCloud Architecture Technical Deep Dive VSP3067 - Mythbusters Goes Virtual VSP2122 - VMware vMotion in VMware vSphere 5.0: Architecture, Performance and Best Prac… VSP3255 - VMware Storage vMotion Deep Dive and Best Practices CIM2452 - VMware vCenter Operations Technical Deepdive VSP2884 - What’s New in Performance for VMware vSphere 5.0

Wednesday

VSP2376 - Performance and Scalability Enhancements in VMware vStorage VMFS 5 VSP2347 - What’s New with VMware vSphere 5.0 VSP1708 - VCDX Panel Defense Preparation VSP1700 - VMware vSphere 5.0 Storage Features VSP2360 - VMware vCenter 5.0: What’s New, What’s Cool VSP3868 - VMware vStorage Best Practices

Thursday

BCO2874 - vSphere High Availability 5.0 and SMP Fault Tolerance – Technical Overview an… VSP3866 - Performance Best Practices and Troubleshooting VSP2757 - A Deep Dive on Virtual Distributed Switching and Cisco Nexus 1000v

The complete content catalog is here, but you might not be able to access it if you aren’t registered for the event.  Let me know if you have any suggestions or things I should definitely be attending.

Also if you are going to be at VMworld this year, send me an e-mail and let me know so we can meet up.  I’ll be posting here and on twitter throughout the week.

[geek-slate]

Cisco UCS B230 Performance Issues due to C1E power State

A while back a client of mine was having sever performance issues with the newer, better Cisco UCS B230 M1 blade.  Through troubleshooting we determined that their VMs were stalling due to the CPUs on the host entering an enhanced power state called C1E. I have confirmed that this behavior also exists with the B230 M2 Blade

Within the VMware vSphere performance best practices guide VMware recommends disabling the C1E power state in the server BIOS

Disable C1E halt state in the BIOS if multi-threadedp performance and I/O latency are important considerations. Consider enabling C1E if “Turbo Mode” is present and single-threaded performance of certain workloads is more relevant.

Cisco has now provided a way to disable the C1E power state as of the 1.4(3i) release

VMs on a blade running VMware hypervisor will be as efficient or fast as expected when compared to other blades using a different Intel architecture. The ability to disable the C1E state in the BIOS is now exposed and may need adjustment to correct the performance problem. (CSCtq00382)

However they have not yet provided a way to set this BIOS setting through the GUI or a policy within UCS manager so even after you upgrade to the 1.4(3i) or later software you will need to manually boot into the server’s BIOS setup and disable the C1E feature.

The default setting in the BIOS is enabled, you will need to manually change it to disabled until Cisco provides a way to disable it via policy

[geek-slate]

vSphere 5 gets a D- on Networking?

Greg Ferro writing on his Blog Ethereal Mind says that even with the update to vSphere 5 VMware still does not provide “proper networking” in their virtualization hypervisor and grades the upgrade a D- on networking capabilities.

Colour me dubious but a 300% price hike in vSphere licensing for this ? They are definitely taking the mickey. Even with the backdown that delays the price hike to a couple of years in the future (when everyone has forgotten about it), we still don’t have proper networking in VMware.

I think the problem here is just one of perspective, VMware provides powerful easy to use networking features that server admins demand.  VMware is primarily selling their products to the server admin camp within enterprise environments.  They need to make sure that these groups can implement vSphere with as little fuss as possible and that includes imposing minimal impact on the upstream physical network.

The features added with vSphere 5 in regards to networking will be greatly appreciated and are starting to move the included networking functionality towards providing a more comprehensive solution to their customers. Both NetFlow and port mirroring support are welcome capabilities in my book.

For those network engineers that are looking for more complete network capabilities in VMware vSphere, they have built a pluggable architecture that lets vendors provide their own virtual switching infrastructure and provide an experience and feature set more familiar to network admins.  The Cisco Nexus 1000v is the solution for those customers right now, hopefully more vendors will provide their own solutions in the future.

This positioning allows VMware to provide easy to use, easy to implement networking features to the vast majority of their customers and provides a extensive highly capable solution to those that are looking for more.  

How is this a D-?  I’m not saying there is no room for improvement, but I would give it at least a solid B at the moment.

[geek-slate]

Signaling either Peak Oil or Peak Apple, Apple earlier today topped Exxon Mobil as the largest US company by market cap.  - Mark Gongloff for the Wall Street Journal

~

 Remember when Apple was still a small struggling company?

[geek-slate]

VMware quickly softens their stance on new vSphere 5 licensing

Just three weeks after the initial vSphere 5announcement VMware has updated their plans for the new licensing model.  The reaction from the customer and partner communities was excited to say the least, so it’s good to see that VMware is listening to us and loosening some of the restrictions.

In principle I think that providing a pooled entitlement for vRAM commitments is a good way for VMware to balance the licensing costs around a more realistic metric. However my initial reaction and the general consensus around the datacenter teams at work was that the vRAM entitlements were way too low.   

We have a long history of driving VM density within our clients and the previous commitment levels would have had a dramatic impact on past clients as they look to upgrade to vSphere 5, and would have also had an impact on how we design our solutions, as there would now be less of a benefit from a highly optimized dense system.

The low vRAM commitment levels seemed to fly in the face of their story of driving density higher to lower TCO and improve efficiency in the datacenter.  

The table below from the press release summarizes the new vRAM entitlements

These updated entitlements significantly reduce the negative impact I was expecting on our existing customers that are actively managing dense environments and our future designs. 

VMware also made two other significant changes to the licensing model. First the maximum amount of vRAM that a single virtual machine can consume is capped at 96GB.  For some this was going to be a deal breaker without the modification as they couldn’t see paying for multiple enterprise plus license for each of their monster VMs.  It put a very real, very large cost per monster VM that swung the TCO decision needle back towards physicalizing those workloads instead.

The second change involves how the licensing accounting is done.  Originally vCenter would throw an alarm instantaneously if you provisioned running VMs over the vRAM entitlement.  The update bases that alarm mechanism on a rolling 12 month average of vRAM usage.  This allows temporary or bursty environments to remain compliant with their licensing at a lower commitment level if their workloads are not year-round.

One last item added licensing tier for desktop workloads, this license doesn’t track vRAM commitment at all, keeping tradition with previous VDI licensing models which are based solely on user session count.

While some still take issue with the general structure of the new licensing model, I think that the changes just announced have allayed my original concerns and should be a much more acceptable fit for our customers.