Wednesday, July 18, 2012

From Cloud 1.0, to Cloud 2.0, to Cloud 3.0

Meg Bear, Vice President, Oracle Cloud Social Platform, published an opinion blog article titled "Multi-Tenancy and Other Useless Discussions." Cited were two articles regarding "very strong opinions on either side of the multi-tenancy divide." The irony in this opinion article is that Meg does not seem to understand that Oracle has the technological lead among the giants in this arena, but are quickly losing their position. A subset of this post was submitted in a comment on Wednesday July 18th, a little after mid-day.

Oracle's Position
With the advance of massive ZFS storage capabilities from Oracle and hypervisors - there is tremendous opportunity for cloud based solutions to provide customers with incredible customization options.

Multi-tenancy offers the ability to offer consistent levels of service and features across multiple customers. Multi-tenancy also offers reduced overhead for the service provider to reduce costs in managing the solutions - which leads to greater application availability. Oracle has superior products which technologically corner this market.

A multi-tenant solution does not mean that every customer has to be on the same version of the software. New versions of the software can be rolled out in parallel and customers can "choose" which version they wish to be under... the former release or the newer release. Good multi-tenant solutions should never be monolythic, rather they should be modular and parallel.

The less capable an individual software solution is, the less isolation, scalability, and management features are offer. Less mature solution do not offer multi-tenancy. It is not necessarily an issue with "Cloud V1" vs "Cloud V2", as the Oracle CEO suggested, it is an issue with cloud solution maturity.

Let us illustrate:
The Oracle VM for SPARC hypervisor offers advanced Oracle ZFS filesystem capabilities with: massive dataset capability; encrypted datasets on a per-customer basis for storage, over-the-wire, and memory buffers for security; superior performance with hardware acceleration of compression (T5) and encryption (T1-T5); guaranteed data integrity & correction from the OS, through hypervisor, through memory, through HBA, over wire, to storage, and to the disk (at every layer of the stack); and much higher capacity & throughput via ZFS deduplication in memory, via hypervisor, over-the-wire, and on disk storage (and not just over the disk bus); and vastly superior visibility and analytics for live production systems with no application interruption with DTrace at every level of the hypervisor, to OS Kernel, through Java, to the application. Oracle owns all of this multi-tenant technology under "Cloud 1.0" - nearly everyone else just borrows some pieces to produce a popular (but technologically inferior) solution.

Why is this brought up?
These capabilities are notably missing from the Oracle VM for Intel hypervisor - the basis for "Cloud 2.0". Clouds based upon this solution are not as robust. Moving the ZFS to a dedicated storage infrastructure leaves gaping holes. From a cloud service delivery perspective, ZFS would need to run on the hypervisor or OS layer in order to: correct hidden data corruption introduced at and below the hypervisor layer (memory, backplane, HBA, wiring levels, etc.); provide throughput improvement via compression & deduplication (via memory, backplane, hypervisor, over-the-wire, storage subsystem); and provide massive storage capacity.

Security, data integrity, and performance drive multi-tenancy requirements. Applications run under "Cloud 1.0" and "Cloud 2.0" solutions. Oracle VM for Intel hypervisor needs to "get with the program" and gain some of these [nearly 10 year old] security, data integrity, and performance technologies leveraged by service providers in the traditional "Cloud 1.0" multi-tenant market as well as by some "Cloud 2.0" (Oracle-derived, but now independent) service providers.

Right now, there seems to be 3 technologies which completely satisfy market needs suggested by the CEO in "Cloud 2.0": Oracle VM for SPARC clouds (Oracle leaving this for other Multi-Tenant providers); Xen for Solaris Intel clouds (Oracle discarded); and KVM for OpenSolaris under Intel (Joyent produced SmartOS, based upon Illumos, based upon OpenSolaris.)

ZFS, Encryption, Compression, Deduplication, and Visibility are the technologies today. Filesystem clustering is tomorrow. Everything else is "just a cloud" floating by... based upon commodity Linux, without data integrity, without data security, without performance, without visibility, without clustering - but there is a thunderhead on the horizon.

Clustered filesystems in conjunction with ZFS fills a bigger hole, in what should be called the "Cloud 3.0" market. Honestly, a Cloud should use clustered filesystems with the data everywhere, with the data encrypted everywhere, with the data compressed everywhere, with the data deduplicated everywhere (and, of course, this means memory, the wire, and the storage.)

Intel x64, IBM POWER, and Oracle [any-architecture]
The HPC market is nothing more than a specialized cloud. Intel purchasing Whamcloud, for Oracle Lustre on Linus's Linux on Intel's x64, and IBM, layering Oracle Lustre on Oracle ZFS on Linus's Linux on IBM POWER, shows where the cloud market is moving towards. IBM pulling off what Oracle could not do, years after the purchase of Sun Microsystems. The market is clearly and proverbially "throwing down the gauntlet".

Oracle is the only vendor on the horizon who has the possibility of clearly layering Clustering on ZFS with full visibility, without sacrificing the performance of a Kernel implementation of ZFS and Lustre, under Oracle Solaris with a Solaris hypervisor, to produce (what I would refer to as) "Cloud 3.0".

If Oracle does not capitalize on "Cloud 3.0" by bundling Lustre, ZFS, and Solaris (on Intel with Xen, SPARC with LDom's) - Oracle "Cloud 1.0" multi-tenancy will still continue to be the 1000 pound gorilla in the room. Oracle (and nearly every other vendor, outside of Joyent) are "Cloud 2.0" Resus Monkies, lacking DTrace visibility and in-kernel clustering & ZFS. "Cloud 3.0" has the potential to dominate Oracle in technical capability, using derived Oracle technologies like ZFS, Lustre, and Xen (which both Oracle and Sun both participated in.)

With every technology Oracle ports to Oracle Linux, the more technology is "given away" to it's competitors, as it is absorbed "upstream" and sucked into Red Hat and Suse - who are recognized Linux vendors.
Network Management:
Network Management platforms used by Telecommunication Service Providers are dependent upon large multi-tenant software solutions. With Oracle Linux not showing up on Network Management software vendor list, the only serious options are SPARC Solaris, Intel Red Hat Linux, and Intel Windows.

The lack of an integrated clustered filesystem with ZFS under Solaris, has long dogged the telco providers. The drive towards Red Hat was relentless. The push by Oracle to "Cloud 2.0", without ever providing a technologically superior "Cloud 2.0" alternative under Intel (Windows, Linux, or Solaris) has left Oracle largely out of the picture.

The drive to "Cloud 3.0" appears strangely dim for Oracle, while it looks promising for OpenSolaris dependent vendors like Joyent. OpenSolaris based technologies (Solaris 11, SmartOS, Illumian, OpenIndiana, etc.) are superior to the dominate market players, but this could be pre-empted by Intel or IBM, as they fund the contribution of Oracle technologies to Linus and his Linux - to suck all value out of Oracle's "Cloud V1" and provide superior "Cloud V2" solution to anything that Oracle is supplying today.


No comments:

Post a Comment