Storage Peer Incite: Notes from Wikibon’s July 12, 2011 Research Meeting
Recorded audio from the Peer Incite:
On July 12 VMware made a "monster" announcement of a new version of its base software with a group of new features and an overall aim of automating infrastructure below the virtualization layer. The result is a much simpler and more unified virtualized environment. And this announcement is as much focused on making virtualization practical for SMBs as it is on bringing improved features and functions to large enterprise users. That afternoon Wikibon.org conducted a Peer Incite Meeting to discuss the announcement. This newsletter contains our conclusions.
Overall VMware 5.0 is a major step forward in the technology. Whether it is compelling or not, however, depends on your organization's situation and immediate plans. Overall, companies pursuing aggressive virtualization of their operational environments, or those waiting impatiently for a version of the technology that will allow them to begin such a program, should seriously consider upgrading soon. Those that are reasonably comfortable with their present virtualization environments have much less reason to do so and may find the cost of upgrading too high.
So is this the "software mainframe" that VMware CEO Paul Maritz promised in 2009? No. The present version still lacks important features that have long been part of a mainframe, particularly in the area of security, which is still problematical. Large VMware environments with large numbers of virtual machines will experience "noisy neighbor" and "nosy neighbor" issues. Also shifting applications between the internal and external clouds is still a much more complex procedure than is implied by VMware's marketing material.
However, this is a major step forward, and the simplification of management of the underlying infrastructure is particularly important, especially for medium-sized companies that are anxious to move forward with virtualization but who cannot afford the large specialized staff that large enterprises use to run their virtualized environments. And it clearly shows the direction of future development that we can expect from VMware over the next few years. G. Berton Latamore, Editor
VMware’s July 2011 “monster” announcement provides enterprise-class incremental improvements for customers who want to virtualize deeper into the application portfolio and bring in more business and even some mission-critical apps. Better storage integration will simplify I/O management and improve the lives of VMware admins considerably.
The announcement also provides excellent new capabilities for smaller customers who will make the platform more attractive for entry-level clients. Missing from the announcement are more complete capabilities that will satisfy the most demanding applications and CIO concerns about hybrid cloud viability. Specifically, while VMware made moves in the areas of security and management and has placed considerable emphasis on service provider partnerships, more work needs to be done to make the enterprise cloud a fully trusted, agile, and an economically viable solution.
The Software Mainframe
In 2009, at a financial analyst conference, VMware’s CEO Paul Maritz said that he had 2,000 engineers working on developing a “Software Mainframe.” At the time, the Wikibon community thought that was an outlandishly ambitious task and an enormous bet. After all, a mainframe is designed to be highly scalable, never go down, and provide a trusted and secure platform. Could a hypervisor company, run by a former Microsoft executive deliver on this vision? If so, we believed, VMware would become the next IT company to command a $100B valuation on the market.
Yesterday, on July 12, 2011, VMware unleashed a huge announcement spanning hypervisor to policy management engine and hundreds of capabilities in between. According to Steve Herrod, VMware’s CTO, developing these advances entailed:
- One million engineering person hours,
- Hundreds of announced functional capabilities,
- Two million hours of QA and testing.
Maritz said that this was VMware’s next step in the journey to bring automation to IT infrastructure. Maritz cited two major changes in the industry:
- A change in architecture – what many call “cloud” – in an effort to reduce complexity, simplify infrastructure, make systems more dependable, and automate IT; and
- A change in end user devices - specifically referring to the eclectic mix of consumer and digital devices that users have, often not even owned by the corporation.
Maritz sees these two worlds coming together and asserted that VMware was building a new platform that could service both legacy applications in the enterprise and a new, emerging breed of apps for consumers.
Characterizing the VMware offering, Maritz used the analogy of a Microsoft office suite. Maritz spent 15 years at Microsoft and said at the time, with Office, Microsoft was building a suite of productivity tools for white collar workers. Today, claimed Maritz, VMware is building the world’s first “cloud infrastructure suite.”
At the July 12, 2011 Peer Incite the Wikibon community broke down the announcement and evaluated the degree to which VMware is delivering on its commitment. The community discussed implications for users and advice for CIOs, CTOs, organizational structures, the ecosystem at large and asset management implications.
What was Announced?
VMware announced it’s Cloud Infrastructure Suite (short Wikibon video explanation) comprised of five elements, including:
- vSphere 5.0 – the hypervisor/cloud operating system with significant enhancements in scalability and storage functionality;
- vCenter SRM 5.0 – Site Recovery Manager to enable enterprise customers to automate failback and a native host-based replication product;
- vCenter Operations (not new, announced in February of 2011) – Looks at how organizations can operationalize hybrid cloud infrastructure – moving data from internal to external clouds;
- vShield 5.0 – bringing new security enhancements to VMware;
- vCloud Director 1.5 – allows for fast provisioning using a feature called Linked Clones and providing self-service and policy management for VMware customers.
Breaking Down the Announcement
Users should view this announcement in four broad segments:
- Offerings for smaller businesses: VMware is making the low end more cost-competitive by reducing the size of the kernel and providing a thinner, simpler version that competes more effectively with Hyper-V. The ability to consolidate and share server-attached storage is particularly attractive.
- Capabilities for existing VMware customers who want to drive VMware deeper into the application portfolio: VMware has accomplished this by increasing the number of resources supported (i.e. memory, storage, CPUs, etc.) and scaling the architecture. While it doesn’t resolve all the challenges of virtualizing mission critical apps such as Oracle, it does address many hardware constraints.
- A high-availability component with SRM 5.0 that automates not only system failover but also failback (one of the missing components of VMware’s integrated stack previously). This is critical for customers as often failing over to a new site injects substantial business and organizational disruption that can be minimized with failback.
- Security – a means of approaching security in logical versus physical terms.
Practitioners on the call, including Dave Welch of consultancy House of Brick, added that the suite is extremely impressive with vCloud Director’s Linked Clones being a highlight of the announcement contributing to faster deployment. SRM failback was also noted as an impressive capability and users especially liked the concept of a bundled enterprise offering that is comprehensive in nature.
The vision put forth by Maritz included the notion of ‘sliding’ applications from on-premise private clouds to external public clouds, while compelling is incomplete from a product perspective. The major gaps from a product standpoint, according to the Wikibon assessment, are in security and management.
Specifically, virtualization increases the number of entities that have to be protected from each other within the system (i.e. “nosy neighbors”). These include LUNs, memory resources, network ports, controllers, VMs, etc. The more entities you have, the more boundaries need monitoring and the more data is created to monitor boundaries. This dramatically increases complexity, and VMware security today simply hasn’t yet achieved mainframe-like status.
On the management side of the house, the tools for VMware still don’t provide application service level performance – management today is done at the VM and storage levels. If the number of apps being managed gets too large, users will get hot spots, which means you’ve got to constrain the number of apps under management. While this can be addressed by creating multiple vCenters with like apps (in separate silos), this brute force approach is less efficient with VMware than with a mainframe. As the number of choices and permutations increases, management overhead goes up, especially as it relates to the amount of code required to perform recovery. For example, in MVS, IBM’s mainframe OS, every module had two development teams, one for the functional code, one for recovery. That’s expensive – but effective.
Nonetheless, VMware is well on the way to achieving Maritz’ vision. Today, more virtual servers ship than physical machines. By the end of 2011, according to IDC and Gartner, 50% of all x/86 workloads will be virtualized. Less than 15 years after it’s founding, VMware runs nearly half of the workloads on the most popular platform (Intel x/86). Only the IBM 370 mainframe and the x/86 processor itself can claim such a rapid ascension in the world of the commercial enterprise.
New Licensing Causes Heartburn for Some
VMware announced new licensing as part of vSphere 5.0. It is a “pay-by-the-drink” licensing model that, according to VMware, shifts the emphasis in pricing off of physical resources (e.g. CPUs and cores) to a logical view (i.e. virtual pooled RAM – vRAM -consumed by VMs). Each vSphere 5.0 CPU license includes a specific amount of vRAM or memory configured to virtual machines, depending on edition – Standard, Enterprise or Enterprise Plus.
As Jeff Kelly writes this week in his CIO Actions research note:
Entitlements per CPU license in vSphere 5.0 Standard edition is 24 GB per vRam, with up to eight vRams, for a total pooled vRam capacity of 192 GB. Total pooled vRam can be spread across any number of servers, giving CIOs significantly more flexibility than the vSphere 4.x licensing model. At $995 per license, total costs in this scenario reach $7,960, slightly more than one would pay under the old licensing model.
Wikibon believes that in most cases the increase in price is worth the improved flexibility the new licensing model provides. However, if the numbers simply don’t add up, enterprises running vSphere 4.x should postpone upgrading to v5.0 until the economics make sense or negotiate heavily based on the organizational impact.
The bottom line is each client must do the math – some will be better off, some won’t but it doesn’t appear that VMware is attempting to make sweeping pricing increases across the board. Specifically, on balance, we believe that a move toward usage-based pricing, especially for those organizations with chargeback models in place, is the right move for VMware, as it’s the model for cloud computing.
Nonetheless, VMware did a poor job of communicating this change and created unnecessary angst in the user base. Users should insist that VMware provide tools to evaluate the impact comparing ‘as-is’ pricing with 5.0 fees.
Can VMware Compete with the Big Public Cloud Players?
Amazon EC2 and S3 changed the world of cloud computing. Google with App Engine and Microsoft Azure all offer resources for developers to leverage the cloud. Apple with its iCloud and Facebook are also big players. VMware is unique in that it is working with service providers such as CSC, Accenture, Verizon, and others so that customers can access compute and storage in a pay-as-you-go model.
VMware also is pushing the notion of hybrid clouds and specifically talks about ‘sliding’ applications from private to public clouds and back. The problem is that often data needs to go with applications, and moving large amounts of data across networks takes a long time. The other issue is that data centers have numerous applications that are interlinked—i.e. applications feed data to each other, which can add complexity and reduce feasibility for the hybrid cloud notion.
Users should ensure that applications that are ‘slid’ are appropriate for this use case – these include Web, some collaborative apps and certain less data-intensive apps. Over time this concept will catch on. However today, CIOs have many questions about the idea. Specifically, these concerns span from economic uncertainties (i.e. can hybrid clouds really cut costs) to organizational nuances (i.e. what roles and responsibilities will emerge) security and policy concerns, etc. CIOs should push service providers for answers to these questions and ensure proof points are in place before rushing forth to adoption.
Despite these concerns, VMware is unique in its positioning relative to the big cloud players, and its value proposition is a secure, enterprise-class, cost effective, high velocity experience where infrastructure is much simpler to manage and can meet the demands of enterprise customers. Users should proceed with caution with respect to VMware’s hybrid cloud vision as security and management are not fully baked to support the “software mainframe.”
Action item: VMware’s July 2011 announcement offers many enhancements for existing VMware customers, particularly in the area of hardware scaling and storage simplification. These features will support the continued adoption of more mission-critical apps within the data center. Additionally, SMBs will be well served by increased competition at the low end via VMware enhancements. Users should carefully evaluate new licensing impacts and in those cases where costs are reduced, organizations should aggressively adopt the new function in vSphere 5.0. In cases where licensing increases costs, users should carefully justify the business value of 5.0 as compared to staying on existing 4.X platforms.
VMware vSphere 5.0 provides significant storage management improvements for small- and medium-sized businesses with existing VMware installations. SMB CIOs who wish to increase the number of applications running on VMware to simplify data center management should migrate to vSphere 5.0.
For enterprise CIO’s, today’s release also offers significant improvements in VMware manageability and scalability. These are two particularly important areas for enterprises looking to move large test and development systems currently running on Linux or x86 platforms into production. Wikibon recommends enterprises running on Linux or x86 architectures that want to move to vSphere 5.0 do so now. The storage improvements in vSphere 5.0, particularly the Auto-Deploy, Profile-Driven Storage and Storage DRS features, make it an attractive option for enterprises that want to simplify storage management.
One important issue causing some discord among VMware customers is vSphere 5.0’s new “pay-for-consumption” licensing model. According to VMware, the company is replacing the current physical entitlements of CPU cores and physical RAM per server with a single, virtualization-based entitlement of pooled virtual memory, or vRAM. Each vSphere 5.0 CPU license includes a capacity-specific amount of vRAM, or memory configured to virtual machines, depending on edition – Standard, Enterprise or Enterprise Plus.
Total pooled vRam can be spread across any number of servers, giving CIOs significantly more flexibility than the vSphere 4.x licensing model. However, the new licensing model could result in significantly higher licensing costs for those utilizing high-memory configurations such as those offered by IBM and Cisco.
Wikibon recommends CIOs use available sizing tools to determine the impact of the new vSphere licensing model on their specific environments and growth paths. If the numbers simply don’t add up, enterprises running vSphere 4.x should postpone upgrading to v5.0 until the economics make sense.
Action item: Enterprises intent on leveraging virtualization to improve data center efficiency and flexibility should move to vSphere 5.0 sooner rather than later. Organizations that are satisfied with their current level of virtualization, however, have no reason to migrate. Similarly, SMBs intent on further virtualizing their IT environments should seriously consider upgrading to vSphere 5.0. Those SMBs whose storage capabilities are adequate at the moment can afford to hold on upgrading. CIOs should also carefully examine the impact the new vSphere 5.0 licensing model will have on their costs before deciding to upgrade. If the new licensing model causes significant increase in costs, as it may at enterprises running just a handful of large machines with significant memory requirements, remaining with vSphere 4.x is advised. On the whole, however, Wikibon believes the new vSphere licensing model is a net-positive for customers, as it allows enterprises to "pay-for-what-you-use" as is typical of other cloud-related technologies.
VMware has three targets user groups for V5:
- Very low-end users, where it faces increasing competition from Microsoft’s Hyper-V virtualization (no charge) functionality and storage cost reduction from Virtual Storage/SAN Appliances from HP(LeftHand), FalconStor, Gluster, NetApp, StorMagic, and many others, that work with VMware or Hyper-V;
- The Kernel has been reduced to 100K to reduce the memory and server resources required to run VMware;
- vSphere Storage Appliance (VSA), which includes the capability of creating a very low cost/low administration RAID 10 Virtual SAN across the storage on up to three servers;
- High-end users with VMware memory, server and/or storage constraints:
- vRAM support has been increased from 256GB to 1TB per VM;
- vCPUs support for vSphere increased from 8 to 32;
- Improved VMFS largest extent from 2 terabytes to 60 terabytes and increased number of files (100,000);
- Bespoke storage partner plug-ins into the kernel have been replaced by APIs, which simplifies and improves the performance of VMware, and removes the overhead and availability concerns with kernel plug-ins. This also makes it significantly easier for more storage ecosystem constituents to develop advanced functionality.
- New block-storage VAAI functionality includes quiescing VMs if there is an ‘out-of-space’ condition (Thin Provisioning ‘Stun’), and deleted storage space reclamation;
- VMware vStorage APIs for Storage Awareness (VASA) is a new set of APIs that enable vCenter to see the capabilities of storage array; these replace the bespoke storage vendor plug-ins;
- Improved NFS-storage with support for full clones, lazy clones, extended statistics and space reservation (this is still not enterprise support of NFS)
- Improved HA availability, including disaster recovery;
- Site Recovery Manager (SRM) 5 provides a fail-back capability, as well as failover;
- Improved host-based replication (HBR) within SRM proves “out-of-the-box” solution to replicate VMs natively with vSphere 5.
Low-end installations that want to grow will have more options with vSphere 5. There is a change in license pricing, with the introduction of an emphasis on virtual storage. For entry VMware sites, this is unlikely to make a significant difference in licensing costs, and the cost reductions from additional function and hardware will more than compensate. Wikibon recommends these organizations start planning to move to version 5 VMware software and migrate at a convenient time (e.g., avoiding year-end if there is a significant peak) to the new environment.
Higher-end organizations that have a plan to virtualize production systems will find they can virtualize more systems with VMware 5. Previous planning for VMware environments have stressed very large memory servers with the latest high-performance processors as a way of reducing VMware licensing costs. The new vRAM-based pricing will mean that it may make sense to have a greater number of smaller or older servers and reduce the cost of the servers and overall costs. Wikibon recommends that organizations moving rapidly to virtualize production systems should migrate to VMware 5 quickly.
IT organizations should also take the opportunity to look at the RPO and RTO requirements of systems. VMware offers some very attractive HA capabilities, and many of these capabilities can run with or without traditional storage arrays. It makes little sense to migrate existing array-based solutions; it is difficult to actually achieve cost savings from conversions. However, the availability requirements of new applications or major upgrades should be reviewed closely, and the default HA environment should be native VMware and VMware appliance capabilities. The lines of business should need to clearly justify the business benefits of array-based storage solutions and clustered database solutions.
Action item: VMware V5 software offers significant additional capabilities to reduce IT operational costs and extend virtualization of production systems. Senior IT managers should focus on implementing V5 features that will reduce hardware and operational costs and move cautiously towards a private cloud with one or more virtualization technologies. Calls for strategic investments in so-called hybrid clouds should be ignored until virtualization and public cloud functionality is much more advanced and tested in data centers. The business case for private cloud and public clouds are very different and can be evaluated independent of each other.
When staffing IT functions, the choice of a generalist versus a more expensive specialist depends on the complexity and maturity of the solution and size of the company. SMB staffs often have a single person managing storage, networking, and server technologies, while enterprise accounts will have groups supporting each of these silos. The role of the virtualization administrator is new compared to these other roles but has been growing in importance. VMware and its ecosystem partners have been working on integrations such as VMware vCenter that support central management of the full data center infrastructure. These VMware administrators often fill the role of “IT generalist”, meaning that they touch on topics that traditionally were handled by individual database, server, networking, and storage administrators.
One of the fundamental reasons to move to server virtualization is the rapid pace of deployment compared to traditional infrastructure. One of the reasons for this is that a VMware administrator can spin up a VM without having to go to each of the physical infrastructure teams. As virtualization has spread beyond the low-hanging fruit of simple consolidation and into more areas including business-critical environments, virtual environments have not been immune to some of the process challenges that physical environments have. This includes VM sprawl and VM stall.
With the release of vSphere 5, VMware looks to mature the management of virtual environments to move from processes (where there are many administrator touch-points) to policy-based management (where administrators are alerted, but typically do not need to approve anything once the policies are set).
Additional features that simplify management from VMware's Cloud Infrastructure Suite announcement include:
- Linked Clones functionality in vCloud Director 1.5 provides fast provisioning that uses of templates and users no longer have to wait for a full copy each time they deploy a vSphere vApp.
- Site Recovery Manager (SRM) 5 supports automated failback, eliminating the need for scripting or administrator intervention for failback.
- Storage DRS automates the initial placement of data (see Duncan Epping’s post for many more cool ways to use this feature.)
The rise of IT generalists does not mean that specialists are not needed, but there will be less of them. Cross-training is important as the rise of virtualization and converged infrastructure solutions can make troubleshooting even more difficult than it was in siloed environments.
Action item: CIOs should look to flatten the organization to assist in busting the silos based on physical infrastructure components. More collaboration between IT groups is a necessity to efficiently leverage new VMware capabilities. By aligning organizational structures with new capabilities, IT staffs will be more likely to keep a productive staff at a flat headcount despite growth requirements.
Included with VMware’s recent announcement of vSphere 5 is VMware’s first virtual storage appliance (VSA). Dubbed the vSphere Storage Appliance (VSA), it is arguably the biggest storage-related part of the announcement. This is not because it is a superior product. Indeed, it has oodles of limitations and barely competes with other products from many vendors who have offered VSAs for some time. But it does pool together internal direct attached storage (DAS) on two or three servers and allows this storage to be shared among VMs. Moreover, VMware touts it as “shared storage for everyone”.
With vSphere 5, VMware also announced host-based replication software. Again, it is limited in scope but works with all VMware’s HA, DRS, and SRM capabilities.
VMware positions these new components as targeted toward SMBs who do not have shared storage and cannot afford the cost of SAN hardware and the cost of external storage controllers with replication features.
Then there is VMware’s focus on the cloud. The entire vSphere 5 announcement was, according to VMware, about enabling infrastructure for the Cloud – private clouds in particular. VMware claims to be building a mainframe for the cloud. Such a mainframe must certainly provide a robust I/O subsystem, and with these enhancements we are beginning to see some of the underpinnings of this subsystem.
Adding to these underpinnings is a bevy of startups that plug directly into the kernel/hypervisor and provide high-performance, highly-available storage functions including remote replication. Examples include Tintri, Nimbus, Solidfire, FlashSoft, and IO Turbine to name just a few.
All of this directly threatens traditional storage vendors who offer SAN-based, two-controller external storage subsystems. These same vendors have spent the last decade shifting to industry standard (read Intel) platforms which is precisely what VMware runs on. And a major part of their target market is SMBs, which is precisely where VMware has targeted. These storage vendors are increasingly challenged to play well with VMware while continuing to enjoy high-margins on hardware and replication software. At the same time, the real innovation is being done by the startups. Doubtless we will see some of these startups quickly snapped up by the big system and storage vendors.
Action item: Traditional storage vendors must have virtual machine solutions that exploit new VMware capabilities in virtual machines. The old model of having huge margins at the array level is going away. While there's still money to be made by hopping on the VMware bandwagon by taking intellectual property and putting it into virtual machines as VSAs, the model is changing. Vendors must look to broaden their portfolios and realize that selling external arrays is not the only way to go. Specifically they must learn to sell to the VM administrator.
One of the most difficult issues for IT is that everything tends to be additive -- new technologies and new equipment rarely means that we can get rid of stuff. VMware server virtualization is a dual-edged sword on this topic. On the one hand, server virtualization has been a driving force in replacing underutilized older hardware with new hardware. On the other, virtualization allows applications on old OSes to be encapsulated and kept longer than might be possible in a physical environment.
VMware’s Cloud Infrastructure Suite focuses mainly on operational costs and simplifying the management of IT infrastructure. Simplification of the environment must translate into a reduction in labor costs. While IT organizations don’t like to talk about reducing headcount, the fact is that VMware’s goal of making infrastructure “invisible” (automated) should reduce the required headcount to manage and configure equipment. From a CIO’s perspective, this is a way to make internal IT as efficient as external service providers. This headcount reduction can be used to focus more on application value or cut labor cost, and organizations will do both as appropriate.
In the past, infrastructure was designed and tuned for each application. With virtualization as one of the enabling technologies of converged stacks, a growing number of applications can be deployed on common infrastructure. Standardization in infrastructure speeds up deployment time and reduces uncertainty in costs that are common when custom configurations are put together. While even fully-virtualized infrastructure are not yet as easy to deploy as an iPad, the trend is moving in the direction of allowing IT deliver self-serviceable, measurable (especially with chargeback), predictable solutions that can be compared in cost and utility to cloud alternatives.
Action item: CIOs need to figure out how to allocate the savings from the standardization and automation of infrastructure. For IT practitioners, it means that many must expand skills – by being able to support adjacent technologies such as virtualization or applications – to keep a healthy career track.