A blog for storage administrators and data managers.

We just completed great editions of our “New Rules:  Backup and Data Protection” seminar in Atlanta and Dallas last week.  (Thank you, Trish, from TechTarget for your great logistics and on-site management, and thanks to all who attended the events.)  At the Dallas event, I was struck by a funny happenstance…

Dallas was the first city, I believe, in which TechTarget loosened its rules and allowed sponsors to sit in the back seats and view the seminar.  Prior to this, they had to stay out of the room and listen with a glass through the wall or something.

Anyway, TechTarget asked me if I would mind letting them in, and I courageously agreed.  Truth be told:  I routinely piss off vendors (well, their marketing droids more than their engineering staff) with what I write; so, why not let them hear the words as they flow “trippingly from my tongue?”  It didn’t dawn on me until after the second hour of my three hour talk, when everyone adjourned to the vendor display area for lunch, that I had said critical things about cloud storage and cloud backup at an event where three of the four vendor sponsors were promoting cloud storage or cloud backup.  Honestly, I don’t naturally think of cloud when I think of IBM, or EMC, or Dell, or whatever.  However, this cloud verbiage was showing up on their signage.

Let me state my current thinking about clouds.

First, the term “cloud” has zero probative value and is a marketing term only — like software-defined networks, software-defined storage, storage area network, enterprise storage, tier zero, etc.  If I had my way, it would be stricken from the vocabulary of technology.

Second, private cloud appears to be a synonym for “managed infrastructure” — at least as it applies to idiotic terms like IaaS (infrastructure as a service), PaaS (no, not the egg coloring kit for Easter, but platform as a service), SaaS (storage, rather than software, as a service — I guess the software guys need another acronym; maybe AaaS).  The “as a service” thing suggests that we provide resources on a contract basis and in some sort of atomic unit of compute with an associated cost.  So, if I want to roll out 50 virtual desktops, I might obtain one unit of DataCore Software’s new virtualization appliance kit; 2 if I want 100, etc.  DataCore provides the software to manage the delivery of performance, capacity and protection in a predictable and manageable way in a private LAN/SAN setting.  To paraphrase Geico, provisioning virtual desktops with their technology is so simple even a neanderthal can do it.

Unfortunately, neither VMware nor other server layer hypervisors are as predictable.  Storage is simpler to virtualize than servers, I guess, due to the complexity of server workload relative to the simplicity of storage I/O.  (I am trying to be generous here.)

Management, ideally, should occur both at the level of plumbing and kit, and also at an abstracted layer that refers to services.  When I provision storage from my virtual storage pool, again enabled by DataCore, I am able to apply appropriate data protection services simply by checking tick boxes for CDP, snapshot, mirroring, etc. — that is, whatever services I want to extend to the data occupying this allocated storage.  The workload in question automatically receives network load balancing, performance improving caching (using DRAM in the server and FLASH if you so desire) and thin provisioning of capacity (across all spindles in the pool, rather than isolated to one stand of disk drives as in on-array thin provisioning).  These are services and their provisioning is theoretically so easy that users could do it themselves.  Very cloud like.

Beneath the services layer, however, I am careful to ensure that I have plumbing and kit management — the best being X-IO’s on array RESTful management.

Because everything happens inside the corporate firewalls, private clouds are thought to be more secure (hmm — or at least less prone to external interference than by inside ne’redowells), more likely to meet SLAs, and more capable of allocating and deallocating resources to workload from shared resource pools (maybe, if it is a mainframe:  x86 tinkertoys?  Not so agile.)

Public clouds, by contrast, could not possibly deliver security or a reliable SLA.  Why?  Because the ability to deliver secure dependable service is outside the hands of the cloud provider’s hands.  Most cloud providers I have spoken to are very earnest souls — meaning, they think they are on the cusp of a new wave of technology and that they are delivering a very valuable service to what few customers they have thus far been able to garner (88% of public storage cloud capacity are serving PaaS and IaaS vendors, not consumers!).

The fact is that the only way to keep data safe in a public network is not to put it there.  Moreover, despite your earnest desire to deliver an SLA, you are at the mercy of WAN connections and last mile connections, which are very much subject to jitter, latency, failure, maintenance downtime, etc.  If I am not mistaken, and please correct me if I am, all of the special QoS guarantees promised by switch vendors fall apart if data routes through a series of hops that include even one off-brand router in path:  so much for your bandwidth optimization and slip-stream technologies to expedite I/O across WAN links.  Moreover, WANs follow their own version of open shortest path first — an invitation to latency given that the “shortest path” refers to # of router hops not physical distance.  Bottom line:  as demonstrated by many problems with Amazon and other cloudies, the network is the boss, not the service provider.  Anyone who tells you anything else is pulling your leg.

Another fact:  IPvAC (IP over Avian Carrier) remains a faster way to deliver large amounts of data over distance than any WAN.

Then we get to another obvious point:  there are few if any real standards in the world of Clouds.  If anything, the momentum in IaaS and PaaS is running against standardization.  Instead, cloudies are teaming up with one server hypervisor stack peddler or another, setting the stage for cloud conflicts.  Clouds are server hypervisor battles writ large.  So much for the ENRON-like dream of sourcing commodity storage or servers from a “futures market” of providers on a spot pricing basis.

Finally, I have a clear recollection of a fellow back in the 70s who was selling hot site contracts for a facility that didn’t really exist.  His rubes never checked to confirm what they were paying for and all wrote checks for $25K or more that he cashed and took with him to a non-extradition country, where he lived happily ever after.  I encourage anyone seeking the services of a cloud provider to actually visit the facility that is being advertised.  Check to see whether the firm consists of a real “tier 1″ data center (or two, or three), or if the infrastructure is a beat up rack of Dell servers purchased on eBay stood up in a cage at some ISP or managed hosting vendor’s shop.  Find out how they are protecting your data while it is in their charge.  Make them restore data as a test.  In short, make sure that they are doing everything (and more) that you would do to platform, protect and archive the data from your systems.  While I can’t cite any independent surveys, I suspect that very few folks are doing  the due diligence that should be done before you outsource your data assets (and your control) to a third party.

Frankly, I hate all of the hype around clouds.  Claims of huge increases in cloud adoption, written by “analysts” who are being paid by the service providers and leveraging surveys of fewer than 20 respondents — half of whom are vendor plants, are finding their way into publications as “editorial” content.  They aren’t.

Lacking is any reference to history:  in the 80s, during a recession, we saw an interest in Service Bureau computing that waned shortly after the associated recession turned over; in the 90s, there was the phenom we called ASP/SSP (application service providers and storage service providers), all of which failed after the associated downturn in the economy (associated with the dotcom meltdown) was replaced by the housing bubble.

Today’s cloud stuff may very well be just another manifestation of the same historical phenomena.  If I am right, it is already starting to fizzle.

Anyway, that is what I had to say about clouds.  No offense to any sponsors, and I hope you will all prove me to be a complete idiot — out of touch with the real world trends that will shape tomorrow’s IT.  I kind of doubt it.

{ 0 comments }