Original URL: http://www.theregister.co.uk/2009/05/02/it_power_stations/

IT utilities, the biggest game in town

Is your supplier a player or not?

By Chris Mellor

Posted in Servers, 2nd May 2009 00:20 GMT

Comment IT suppliers are lining up to play at the Data Centre Casino, the biggest IT game in the world, to stake a claim as a supplier of IT services as utilities. There are just two tables with limited seats; complete systems on one, essential components on the other. Are your IT suppliers in the game or not?

This is all about rewiring the data centre, Nicolas Carr Big Switch style, to deliver IT applications over the wire through a socket in the wall. With server virtualisation and cloud computing ideas in rapid development, IT vendors are pushing the public and private cloud concepts.

Public clouds provide raw IT platform services: pay-as-you-go compute, storage, and base application services from a global data centre infrastructure. There are very few players; for example, Amazon and Google with Microsoft building out its own infrastructure. These players build their own specialised infrastructures and represent a highly disruptive threat to existing business data centre suppliers.

It's probable that an exceedingly small number of suppliers have the will and financial resources to enter this market. They'll need annual revenue in the tens of billions and an existing global data centre infrastructure as a base. Cisco, IBM, and HP will be on that list - and that, amongst existing IT players, is probably that. The general public cloud IT supplier game is probably already closed to everybody else.

Suppliers like Mimecast, Parascale, Nirvanix, and others will have to develop niche public cloud offerings.

Private clouds

Private clouds will deliver applications as utility services inside a business to its users and partners. The developing mainstream concept is that these services will come from a data centre resource composed of unified and virtualised compute, storage, and networking components under the management of system software at data centre, hypervisor, and operating system levels.

This is where the bulk of the cloud IT market will be: re-equipping enterprise data centres to become IT power stations, generating applications as services, and gaining economies of scale from the high utilisation of commodity resources such as bladed x86 processors, scalable modular storage arrays, and an underlying Ethernet network.

Suppliers to this market can either supply complete systems or essential components for them, and there is an ongoing structural change in the industry as suppliers re-organise and acquire technologies in order to be a private cloud system or component supplier. Staking a claim to play at the complete-systems table requires the possession of existing major components, annual turnover of several billion dollars, determination, the cash resources to obtain the missing components, and a long-term committment.

At an overview level the major components of complete private cloud systems are pretty clear: scalable, powerful and cheap servers; virtualising server software; integrated, virtualised, and hugely scalable storage; a virtualised 10GbE network switching infrastructure; Fibre Channel over Ethernet capability to link servers and storage across Ethernet; disaster recovery capabilities; service-quality tracking and control; and management of a virtualised data centre. We could very broadly summarise these components as servers, virtualisation software, networking and storage. Complete-systems suppliers must have all four.

The main players are Cisco, HP, and IBM. Possibly Dell could join them, but there is a question over the scalability of its storage arrays and, for this writer, its ambition. Oracle/Sun might be a player, but chances are that Oracle would limit itself to the supply of complete private cloud stacks for its own applications and not be a general player. There may be scope for a Japanese player here as well, but - after these six - it seems apparent that the complete private cloud system supplier group is closed. The table stakes are just too huge for any more new entrants.

Take EMC for example. It has storage and virtualising software but no networking and no servers. It cannot be a complete-systems supplier.

Essential component suppliers

The essential components are the four categories above; servers, virtualisation software, networking (FCoE and 10GbE), and storage. How are suppliers doing so far?

Only Citrix/Red Hat, Microsoft, and VMware have the virtualising software needed, with VMWare furthest along the road to integrated virtualised server and data centre software with APIs for networking and storage. Microsoft's Hyper-V, positioned to be adopted by large numbers of existing Microsoft customers, will probably develop in the same way and we might expect some linkage between Hyper-V and Windows to enable a Hyper-V server to run more Windows and Linux virtual machines (VMs) than an ESX server. After all, there is much code in Windows - multi-tasking, for example - that can be junked in a single-app VM environment, and various Windows services could be sent upstairs to Hyper-V.

All other suppliers will use hypervisors and data centre-level software from these three sources. Oracle/Sun might try to propose a Solaris-based stack with Solaris containers instead. That will depend on Oracle's intentions - and we await those with interest.

None of the virtualising software vendors looks as if they could build a complete systems offering. EMC is furthest along this road, having both the leading virtualisation software supplier - VMware - and being the leading storage supplier to enterprise data centres. But it shows no signs of wanting to acquire a server supplier or a networking supplier.

Microsoft's traditional relationship with hardware vendors of the server, storage, and networking persuasion is to partner with them openly, favouring none. It has the financial resources to build a complete system, but to do so would be pretty much counter to its DNA.

Servers

Servers can come from Intel OEMs or AMD OEMs. It's entirely possible IBM will propose Power-based server complexes and even develop its mainframes into private cloud service utility machines, if it hasn't already done so, but the mainstream will be X86.

Blade and rack server suppliers such as Rackable and Verari could, do in fact, compete here, but will need compelling integrated storage, networking, and virtualisation software in a complete-system package that is differentiated from Cisco, HP, IBM, Oracle/Sun, and the potential Japanese supplier. They might look for private cloud niches to specialise in as a way of limiting the demand on their resources. Either or both might even get bought by a server supplier needing to strengthen its offering to the private cloud data centre market.

Cisco, HP, and IBM are confirmed players in the complete-systems market. To my mind Dell is not. It has the blade servers, already OEMs VMware and Microsoft software, and has both in-house (EqualLogic) and OEM'ed EMC Clariion storage. However, it doesn't have any more scalable storage than Clariion and, so far, has no expressed FCoE strategy. Clearly there is the potential for Dell to strengthen its storage offerings - but does it have the corporate will?

Dell also OEMs networking products from both Cisco and Brocade and has made no signs at all about distancing itself from Cisco or getting closer to Brocade. It's as if Dell is ruminating what to do. Can it afford, in the long term, not to be a complete-systems supplier to the private cloud market? Will there be any viable complete-systems suppliers to that market who OEM Dell servers?

Many people would answer these questions "No" and "No", in which case Dell has to assemble the components it needs and step up to the mark. How it will assemble the missing storage and network pieces is an interesting question.

HP and IBM have in-house storage products which they can integrate into their product offering. Cisco does not, hence its partnerships with EMC and NetApp. HP also has its in-house ProCurve networking products, an offering IBM lacks, and is probably the strongest complete-system private cloud vendor at present.

Oracle/Sun has in-house storage and networking, although the Sun networking products are not as well-positioned or as rounded as either Cisco's or HP's. Sun's Open Storage ideas are right on the money for building commoditised storage with commoditised software. But it all depends upon Oracle's intentions as to whether this will see the light of day.

Networking vendors

Only Cisco has the market penetration, resources, and ambition to go it alone and develop its own complete system with in-house servers and networking, OEM'ed virtualisation software, and certified storage from partners such as EMC and NetApp. All other networking players will need 10GbE switching and routing products plus partnerships to become an essential component supplier.

That's why Brocade bought Foundry and is establishing closer ties with HP and, more substantial already, IBM. It's also why Broadcom, an Ethernet switch vendor, is trying to buy FCoE IP-owner Emulex, and produce a product set encompassing server attach and offload for FCoE, TCP/IP, and iSCSI, and Ethernet switching that can deal with these three protocols. With Emulex having relationships with storage vendors, courtesy of its in-array switching and bridging technologies, Broadcom will be seeing opportunities there as well.

Other potential 10GbE-based networking players are Blade Network Technologies, Juniper - with an existing IBM relationship but no FCoE - and Extreme Networks. Their table stakes as a private cloud networking-component player have to include FCoE.

The only remaining FCoE-capable supplier is QLogic. Instead of pursuing an acquisition, merger, or takeover with a networking vendor, QLogic has chosen to buy NetXen and secure its own 10GbE technology.

This could imply that QLogic sees a viable future as a stand-alone supplier of server-edge network-attach products - adapters in either card or chip format, and that same technology stack used by switch and storage array vendors. On the other hand, it might be QLogic's way of making itself more desirable to a possible acquirer.

Storage vendors

Storage vendors need to decide if they want to supply the highly scalable, highly reliable, and modular storage closely integrated with VMware and Hyper-V and, to a lesser extent, with Citrix/Red Hat, or not. An absolute requirement is a roadmap to native FCoE support and, most probably, a multi-tiered storage offer embracing solid state drives (SSDs) or large controller flash caches - or both - for very fast-access data, Fibre Channel and SAS drives for fast access data, and SATA drives for bulk storage.

They will have to integrate with the virtualisation software for thin provisioning, disaster recovery, VM migration, and data protection. They will also have to integrate with complete-systems suppliers, either on an OEM basis or on a certified-partner basis.

EMC and NetApp have stepped up to the table. Who else is a candidate?

Hitachi Data Systems obviously. Like Dell there is a lack of public committment to the idea of getting involved with complete-system private cloud supply, though. HDS has to partner; it has no other choice if it is going to play.

There is also 3PAR, an existing enterprise data centre storage supplier with a utility storage idea that is already attractive to service providers, and a demonstrated will to integrate closely with VMware. It's easy to conceive of 3PAR positioning itself as a storage supplier to Cisco, and any other complete-system supplier lacking an in-house storage product line or wishing to broaden their overall storage offer.

Lastly there is Sun.

All other storage suppliers that do not have enterprise data centre products or customers are prevented from playing at this essential component table. That means Compellant, Copan, Infortrend, NEC, Nexsan, Pillar Data, Xiotech, and others.

Summary

The discussion above means that, in the public cloud market, only Amazon, Google, and Microsoft are confirmed players, with Cisco, HP, and IBM capable of entering if they decide to. For all others, entering the general public cloud IT market is not possible, leaving niche possibilities only.

In the private cloud market, only Cisco, HP, IBM, and possibly Dell, a Japanese vendor, and Oracle/Sun are viable complete-systems suppliers, having access to bladed servers, virtualising software, and integrated networking and storage. Rackable and Verari have outside chances of supplying such systems.

The only viable virtualising software suppliers are Citrix/Red Hat, Microsoft, and VMware.

The only viable networking suppliers are Broadcom (if it buys Emulex), Brocade, and Cisco (on an OEM basis). Blade Network Technologies, Extreme, and Juniper have to acquire FCoE technology or source it, with QLogic looking like the best potential source with it having NetXen technology for integrating FCoE, 10GbE, and protocols such as TCP/IP and iSCSI. They wouldn't source FCoE technology from Brocade willingly since they compete with Brocade in Ethernet switching.

This only viable storage suppliers are 3PAR, EMC, HDS, NetApp, and Sun.

That's it. We're looking at near closed shops. Six players could play the general public cloud supplier game. Only 20 suppliers are in or poised to be in the Data Centre Casino and playing the private cloud supplier game; six players or candidates at the complete systems table, plus two possibles, and the rest at the components table.

Everyone else is already excluded. It's now in the very early days, very early days, but already the stakes are pretty clear and it seems obvious that no one else qualifies at all; the game is already closed to new entrants. ®