Things do seem to go in cycles in the IT business. For example, wasn’t the mainframe the first integrated stack? It was a mysterious black box, incorporating hardware, storage and software, and it made your applications work, mostly without you having to worry too much about it. Now, integrated IT stacks are back, but in a different form with different names: Vblock, FlexPod, vStart, VirtualSystem, CloudBurst and Exadata, to name the main vendors’ offerings.
Open computing enabled companies to build their own IT architectures, bolting together hardware and software components at low cost and working out the kinks themselves.
Then came the emergence of virtual server infrastructures, when a myriad of virtual machines and their related storage made system and performance management much more complex. Integrated stacks are in part designed to solve that management problem with preconfigured collections of components, supported by software that automates much of the management overhead.
Businesses like integrated stacks because they shorten the virtual machine and storage provisioning cycle, said Gartner’s research director for storage technologies and strategies, Valdis Filks. “Businesses don’t want to wait three months for you to create a database,” he said, arguing that the provisioning time under an integrated stack can be shortened to hours.
In spite of these suggested operational benefits, Filks reports that integrated stacks overall have sold only “in the tens” in the UK.
But why? Filks argues there’s a fundamental mismatch at play and that integrated stacks merge server, networking and storage together while these functions in the IT department often remain separate, with budgets in these areas fiercely protected. “IT departments are hesitant, because this can force reorganisations,” Filks said. “People can lose their empires.”
Given how poorly they appear to be selling, there are a surprising number of integrated IT stacks to choose from, with a number of vendors either bolting their own products together or partnering with others.
IBM opts for the former approach. Stack integration is part of its culture, which perhaps explains why it arrived at a commercial product faster than most other vendors. It launched its CloudBurst appliance in 2009, offering an integrated set of server, storage, software and service management systems in one box. Even Big Blue, however, OEMs others’ networking components in this product.
HP, meanwhile, bought its own networking company—3Com—and is one of the few megavendors that can populate an integrated stack from products within its own stable. It launched its VirtualSystem in June this year. The VirtualSystem is a preconfigured selection of servers, HP network switches and storage from its 3PAR Fibre Channel SAN and LeftHand Networks iSCSI acquisitions. The software component of VirtualSystem supports Windows and Linux, with Citrix and VMware fleshing out the virtualisation side. The firm has added value to the stack with automated virtual machine management and security. This represents an upgrade path from the BladeSystem Matrix modular architecture that was previously positioned against other integrated stacks.
Dell, for its part, launched the vStart integrated stack in April. Aimed at smaller virtualisation implementations, in its first iteration this stack supports only the iSCSI arrays provided by Dell EqualLogic. Customers will have to wait for integration with Fibre Channel-capable products from Dell’s other storage acquisition, Compellent.
Cisco and EMC had to focus on partnerships to flesh out their more limited portfolios. The Virtual Computing Environment (VCE), the collaboration between Cisco, EMC and VMware that produced the Vblock, gets a lot of press, but it has its challenges.
“Sales [of Vblocks] are pathetically low. Really, surprisingly low; under 200, worldwide,” said one analyst who was privy to the information in a briefing but asked not to be named. “I think what’s going on there is that the pricing was off. It is higher than [the cost of separate components].”
Another problem could be user confusion. Filks points out that for all its bluster, VCE is still little more than a collection of three vendors that certify their stuff to work together. That’s further from the one-stop-shop experience than customers may want.
That, however, didn’t stop Julian Dyer, CTO at managed services provider Cobweb, from choosing Vblock at an early stage, when it didn’t even offer a unified fabric. The company was expanding into its third computer room in London’s Telehouse West and wanted a 10 Gigabit Ethernet (GbE) core to give it more performance than the 1GbE Nortel-powered core it had been using. Two Cisco Nexus 7000 switches powered that connection.
Cobweb took the SAN that came with the Vblock purchase and also connected three EMC Clariion CX480 arrays to the fabric.
The installation, including data centre room fitting, took six weeks and enabled the company to move its entire set of Microsoft Exchange mailboxes over to the new system. It gives Dyer more density, and he isn't worried about one of the biggest perceived problems with integrated stacks: vendor lock-in.
"As long as Cisco and Dell have some parity in price points, it doesn’t matter much. It’s all going to be obsolete in two years anyway."
Dyer could also have chosen FlexPod, the NetApp initiative with Cisco and VMware, which Forrester principal analyst Andrew Reichman calls a “me-too announcement.” FlexPod is effectively a Vblock but with NetApp storage bolted in instead of EMC. But when Dyer made his decision, EMC had just introduced a dual-density shelf for 2 TB SATA drives that enabled Cobweb to achieve higher capacity in the room than with most other vendors’ equipment.
There are, nevertheless, some technology advantages to the NetApp stack. The company’s hardware has some strong snapshot capabilities, and it does block-level primary deduplication well. It has also focused on thin provisioning technology, which has won it kudos in the private cloud/virtualisation space.
But, again, FlexPod is a three-horse team. Compare that with Oracle’s integrated stack in the form of Exadata, Oracle’s InfiniBand-based “database machine,” which incorporates hardware, storage, networking and software from a single vendor. It complements Exalogic, the company’s integrated grid platform, which supports a variety of platforms from COBOL through to Java and Linux.
Oracle’s advantage is that its Exadata stack capitalises on one of the most popular relational database systems in the world. It also includes significant under-the-hood optimisation in the form of Automated Storage Management (ASM), which rebalances data between disks as it is added, to enhance performance. Understanding the context of data, so that it knows what data should be transferred to an archive, for example, addresses a traditional pain point in the data centre. Communication between application and infrastructure teams has been a sticking point. Now, Oracle has automated it.
Our anonymous analyst cites figures under a nondisclosure agreement from Oracle that suggest sales of 1,000 Exadata units worldwide, although how many of these are free-of-charge, proof-of-concept shipments to customers is in question.
There are other converged architecture solutions closely related to the integrated stack concept. For example, Egenera offers stack management software designed to work with multiple vendors’ hardware and also sells BladeFrame, a chassis, processor and switch blade solution managed using its hardware. Dell uses Egenera management software. And startup Nutanix will offer an integrated stack solution that does away with SANs altogether, instead using direct-attached storage on scaled-out servers.
As the integrated stack wars heat up, we can expect to see more early-stage firms emerge from stealth mode with different perspectives on the problem. Keep watching.
This was first published in August 2011