The media responded to Cisco's UCS announcement last month by rushing to categorize it as essentially a new type of blade server, albeit one with an advantage in performance capabilities. While UCS incorporates server blades, it goes well beyond the traditional idea of a server. UCS is an optimized hosting platform for virtual machines that unifies network, compute and storage access. UCS works hand in hand with VMware’s vSphere 4 to facilitate disruption of the traditional data center by accelerating widespread adoption of a completely virtualized architecture.
Let’s Get Virtual
A Goldman Sachs survey from late 2008 showed that 15% of servers now run as virtual machines, and the number will more than double by 2010. Servers, as we know them, are already dying – it’s just taking a bit of time for the carcasses to be hauled away. Gartner says that worldwide x86 server shipments dropped 11 per cent for the fourth quarter of 2008.
Declining server purchases are a very recent phenomenon. The first 32-bit server, the Compaq System Pro, shipped in 1989. Twenty years later, we have 50 million servers worldwide that are remarkably inefficient at delivering compute resources to applications. Servers are primarily expensive silos of CPU, RAM and sometimes storage resources that, on average, sit idle over 90% of the time. They not only consume huge amounts of rack space, but also expensive power to keep them running and an equivalent amount of electricity to cool them back down.
Server sprawl results from the software manufacturers’ traditional insistence that their programs operate in isolation from other applications. The underutilized servers must be purchased, configured, racked, stacked, cabled, connected, powered, cooled, protected, backed-up and maintained. Mission critical servers require redundant machines and expensive clustering software to help ensure against downtime from hardware failure. Tight coupling of servers with underlying operating systems makes fast recovery in the event of a data center disaster unlikely, if not impossible. As technology advances, the servers require costly, time-consuming and risky periodic upgrades.
Server administrators who have worked with virtual infrastructure are quick to sing the praises of the technology. The virtual machine is a software server replica that is freed from the liabilities, rigidity and high cost that plague its physical counterpart. Server downtime becomes things of the past. IT tasks such as server provisioning, troubleshooting, testing, patching and performance monitoring are all simplified and enhanced. Racks of physical servers are consolidated onto a much smaller number of host servers also enabling a reduction in switches, UPS devices, generators, PDUs, air conditioners, and other data center components. Disaster recovery and even desktops become components of an agile virtualized data center.
Over 130,000 organizations utilize VMware technology, but most midsized and larger companies have only converted a portion of their servers into virtual machines – often the low hanging fruit of test/development and other less mission critical servers. This partial approach to virtualization certainly beats the physical alternative, but fails to realize the huge economic savings and other benefits for the non-virtualized servers. VMware vSphere 4 changes the landscape by delivering the performance, reliability, security and management to enable 100% data center virtualization. Even the largest database servers can run faster as VMs on vSphere.
Virtual Machine Hosting
As organizations rapidly move toward 100% virtualization with vSphere, they will find limitations in the two existing VM hosting options: rack-mounted servers and blade servers. Both of these server models were designed for physical data center architectures; they not unexpectedly have weaknesses when it comes to hosting a virtual infrastructure. Large numbers of virtual machines consolidated onto a single VMware rack-mounted server, for instance, can create a cabling nightmare by necessitating multiple Ethernet and fibre channel connections. While blade servers can alleviate the cabling issues of traditional servers, they introduce other inefficiencies such an extra layer of virtualization that can degrade performance on both fibre and on Ethernet networks. Both server options fail to address new virtual infrastructure challenges such as how to effectively allocate network and storage resources in conjunction with provisioning new virtual machines.
VMware vSphere demands a new type of hosting platform to inspire IT professionals with the confidence to virtualize 100% of their production data centers. This platform needs to deliver both the compute resources and the high-capacity bandwidth required to facilitate large numbers of virtual machines contending for limited pipes. It needs to unify fibre channel and Ethernet storage platforms while also addressing the networking policy limitations of a virtual infrastructure.
Enter Cisco’s Unified Computing System
Cisco’s Unified Computing System (UCS) is a data center solution that combines networking, server and virtualization technology into a package leveraging the best of what the industry has to offer. UCS is a lot more than the latest evolution of blade servers – it is a new type of architecture specifically built to optimize virtualized environments.
The UCS solution is comprised of five primary components: UCS 5100 blade server chassis, UCS 6100 Fabric Interconnect Module, UCS 2100 fabric extender, UCS B-series blade server, and UCS Manager.
The UCS 5100 Blade Server Chassis is built from the ground up to provide an optimized hosting platform for virtual machines leveraging engineering expertise garnered from years of developing high speed, passive fabric enabled interconnects.
The UCS 6100 Fabric Interconnect Module is essentially a Nexus 5000 switch designed to deliver line-rate, lossless 10 gb Ethernet, Cisco Data Center Etherenet (DCE) and Fibre channel over Ethernet (FCoE) across a single copper or fibre cable. The fabric interconnect forms the core networking component and achieves the “wire once” mantra of the Nexus and UCS technology line. It provides all of the network and storage connections necessary for complete consolidation.
The UCS 2100 Fabric Extender brings the I/O fabric from the 6100 directly into the backplane of the 5100 via four 10-Gbps connections. The Fabric Extender passes traffic into the chassis mid-plane where each half-slot blade server is connected to each of the two fabric extenders, providing each blade-server up to two 10-Gbps connections.
The UCS B-Series Blade Server utilizes the latest Intel processor technology, Nehalem, and can accommodate up to 384 Gb of memory.
The UCS Manager (UCSM) comprehensively manages the entire UCS solution and unifies compute, storage and network resources.
A UCS “cluster” can accommodate up to 40 chassis, each with 8 blades. It utilizes a patented memory capacity that enables far more memory per server than afforded by traditional two-socket blade server architectures, helping to alleviate the biggest bottleneck for hosting large numbers of virtual machines. As a result, a single fully populated UCS can host thousands of virtual machines.
UCS incorporates innovative virtualization technology that encapsulates a logical representation of the physical blade. This not only improves performance of virtual machines, but also improves performance of servers that must remain physical for compliance or political reasons by enabling dynamic resource assignment changes.
Cisco has worked closely with VMware to enable better network I/O performance for virtual machines. And because the hypervisor CPU is freed from network I/O responsibilities, more CPU cycles are available for hosting more virtual machines.
UCS Management Capabilities
UCS utilizes service profiles to define in software all of the characteristics, identities and configuration that make up a “server”, providing for a stateless architecture. These service profiles are created by the server, network, and storage administrators, and are then applied to the system against a pool of available resources. The role-based management allows provisioning of computer infrastructure with automatic coordination between the domains. The respective administrators maintain responsibility for their domain policies.
Policy-based management removes the burden of administrating server, storage and network resources as individual silos. Resources are allocated in alignment with network, security and compliance policies. UCS incorporates Virtual Network Link (VN-Link) technology that works in conjunction with VMware vSphere 4 to enable visibility into the individual machines, thereby ensuring appropriate resource availability even as the VMs VMotion or DRS across UCS blades.
UCS includes an XML-based API enabling third-party developers to potentially develop capabilities such as pooling resources from several blades. This grid-like functionality would give virtual machines access to resources far beyond anything economically feasible in the physical world, and would let UCS act even more like a mainframe in terms of scheduling huge processing jobs during off-hours.
Focusing on UCS as a Server is Myopic
The investor community has been wary of Cisco’s entry into what they perceive to be the low-margin server business. After all, Dell, HP and IBM certainly have far more server experience than Cisco. But this is a myopic view of UCS that fails to factor in the many innovations making it by far and away the most comprehensive hosting platform for virtual machines. The completely virtualized data center that Cisco and VMware enable together results in astounding savings for organizations along with benefits of high availability, reduced administration requirements, effective disaster recovery, superior desktop computing and improved performance and security. The unique value proposition of UCS is its ability to optimally support a virtualized data center through unification of server, storage and network resources.
Thanks to Steve Jones, Gary Lamb, Brad Hedlund and Scott Lowe for suggestions and editing.