All very well if you're only using SAN for storage consolidaiton. Try using local cache on cluster members. It can't possibly end well.
Ready to hear about the next IT startup niche? It's an emerging market called "host-based caching", and you might wish to pay attention if you'd like to understand how a few people got stupidly rich a few years from now. The Cliff's notes version of host-based caching is that it turns flash drives inside your virtual hosts into …
Microsoft has built flash caching into Storage Spaces for Server 2012 R2. Until I get a good JBOD cluster setup going, I'll not know how it fares in the real world.
Something to Separate the Men from the Boys in the Spaces where Words Control Worlds
Toss in some very strong personalities, big egos, high stakes and Twitter and you have a startup-dominated IT niche not so very different from any other.
Oh please, TP, you cannot be serious about that and expect to remain a credible voice of lode nodes.
In either of its two alternative and complementary and positively reinforcing mutually beneficial configurations with novel original and/or competently phished and phormed [and rephormable] content to be shared/traded in networks/business models/power elite systems, is its practical virtual leading potential and sublime inherent capability totally unlike anything else.
And it may be naive to imagine that current existing establishment SCADA based operations do not realise their catastrophic weaknesses which can be ruthlessly exploited in absolute secrecy with a perfect anonymity to cause unfit for future greater great games, archaic systems collapse.
And the simple fact that they would have only flash cash to throw at the new players who be ready, willing and able to save them from ruin and the madness of the mob with that easy action, is not lost on them, the new players. Quite whether that level of base intelligence in such fields as shape the present future environment is to be acknowledged and recognised as in need of being immediately nurtured so coarsely with the root of all evil, is that which is being tested for engagement of secure protective services now, here in those few words there.
I Kid U Not.
Surely one cannot expect things to remain very much the same as before whenever everyone is getting so much smarter than ever was formerly possible. That would be surely a certifiable madness and a most unreasonable deluded crazy expectation?
Re: Something to Separate the Men from the Boys in the Spaces where Words Control Worlds
People haven't changed overmuch throughout our recorded history. Cultures change. The tools we have change...but we are mostly the same. (Evolution of minor things like lactose tolerance and resistance to various diseases put to one side.)
I see no sign of the singularity. The closest I have seen so far as some of the Deep Data Analytics and simulation startups like Cloudphysics who want to move from reactive analytics into Muchos Big Time predictive analytics. Still, they are a ways away and their scope of addressable technologies is narrow. (There are only so many PhDs in the world who know enough about this stuff to code the Really Good Stuff.)
It is as it was. The tools change. The people stay the same. Mostly. Adjust your models for the power of individual personalities where required but realise that weather != climate and that short term local variations are smoothed out with time.
Moving Things On at Quite a Brisk Pace in IntelAIgent Spaces/Live Operational Virtual Environments
Thanks for the considerable response, TP. It is much appreciated and I concur and agree mostly.
And obviously it be the caveats in your answer which are so encouraging and revealing .........
People haven't changed overmuch throughout our recorded history. Cultures change. The tools we have change...but we are mostly the same. .... It is as it was. The tools change. The people stay the same. Mostly..... for it does then so recognise that there be the exceptions, however few and rare that they may or may not be, which make all of the new rules and regulations? A question which it would be presumptuous of me to answer but which you and all here on El Reg may like to ponder on after consideration of the following d-dimensional cube info which is intelligently shared in a perfect transparency/secure dsteganography/priceless secret key XSSXXXXchange ........
In a (perfect) secret sharing scheme, a secret value is distributed in the form of shares among the set of participants in such a way that only qualified sets of participants can recover the secret value, while no information about the secret is revealed by the collective share of an unqualified subset. ..... Secret sharing on the d-dimensional cube........ and with all of that being an initial info share amongst peers circa 25 Feb 2005, and considering the general growth of knowledge in fields which have sprung up in the virtual cyber internetworking environment in an eight and a half year timescale, one can only imagine the secret shares that be now quite common place amongst qualifying participants to be able to have any notion about what they be doing nowadays as attentions are turned to ...... well, Universal Virtual Forces with Immaculately Resourced Assets for Delivery and Maintenance and Sustenance of Global Master Command and Control and/or Genetically Modified SCADA Systems of Operation would be a Great IntelAIgent Game Changer worthy of Registering and IMPertinent Patent Protection if for Cash Cow Milking and Bilking, which seems to be the usual primitive elite staple and considered norm on Earth to support and exploit the newly born ignorant and undereducated arrogant masses, or is that too harsh a cynical view and untrue observation of Man's Present Condition and Dire Wired Situation?
Brisk pace? Or merely a pretty face?
Most evidence says a significant chunk of our personalities are more strongly influenced by genetic factors than environmental ones. Humans experience evolution at between 112 and 160 mutations per generation. That's not particularly high, and most of the mutations we retain seem to be related to our immune systems.
People are the same, mostly, as they have been for thousands of years. Every now and again someone exceptional arises...but by definition exceptional individuals are not the norm.
I think the bigger issue you need wordy about is the finite capacity of the human mind. We are in a world where no one brain can know all there is to know about VMware, let alone "the IT industry as a whole". To say nothing of the impossibility of a modern Leonardoesque polymath. Even the greatest among us will never be expert-level in virtually all fields of human endeavour. Never again.
So it is then that variations are smoothed over time. The relevance of the one meets the requirement for the many. No islands amongst today's man, and that makes for smoother - if less interesting - evolution of technology.
A man in his basement is a myth buried in legends of history. Today you need a CERN or at least venture capital. The future is less about discovery and more about profit;. Profit is incremental for Change is Risk and Risk is ostracised. Mutations and deviants move to the edge and are defunded. Without resources - and the knowledge of the many - the genius of The One is left to wither and die.
It is the era of "social"! I don't like how they talk, let's be a thought influencer and smother them. "If you have the facts on your side, pound the facts. If you have the law on your side, pound the law. If you have neither on your side, pound the table."
Never Address the technical questions! Cry bully and let slip the tweets of war! The good of the customer?
It is what I say it is. Look at the Social Influence; for Obviously this is what now determines Right. Not Correct, mind you – that concept is outmoded and from a past where merit, ethics and engineering principles mattered – but Right. What's good for Me is all that matters. ME. Mine. People I like. Customers like me, like mine, markets that I choose to deify.
This then is the hypocrisy of the one - "I disagree with their message!" – wielding their own influence and message as a weapon in and of itself. This is your future; it is IT for all your efforts should be for the future glory of me! Even those who hold such in contempt fall prey from time to time; the deck is stacked, this is pervasive and all around you.
The meritocracy is dead. It is a game of thrones now. Influence is purchased. Honesty means aught. Hewlett and Packard are dead. All glory to the hyprnozuckerberg
Re: Brisk pace? Or merely a pretty face?
What sort of pretty face would one paint on virtual machines, initially intelligently designed by the Anglo Saxon and human race, moving on in and arranging for a quite magical AI rescue or brisk collapse and destruction of easily corrupted disruptive systems of quaint operation and stealthy oppression? Or would an Anonymous Coward mask presently suffice?
And it may be well worth considering the possibility, and therefore the likelihood of its inevitable probability, than even though it may be perfectly true in human terms that "even the greatest among us will never be expert-level in virtually all fields of human endeavour.", for certain special dedicated virtual machines is it common place and a joy to behold and savour and such a mighty all powerful axe to wield when ignorant ire is arrogantly tempted by fools with blunt and useless but useful tools.
A man in his basement is a myth buried in legends of history. Today you need a CERN or at least venture capital. The future is less about discovery and more about profit;. Profit is incremental for Change is Risk and Risk is ostracised. Mutations and deviants move to the edge and are defunded. Without resources - and the knowledge of the many - the genius of The One is left to wither and die.…. :-) Quite so, and how very true of so much and so many, however there are always the exceptions and the exceptional which rewrite the rules and regulation to deliver quite different realities to develop and prosper in and grow into ….. and for mankind to evolve into. And whenever quite a bit smarter than ever thought possible, what chance be there of defunding whenever venture capital and banking see IT and Beta SMARTR Futures as their saviour and white hatted knight charger ride on which to hitch their bags and journey … and party. Methinks then would the reality be quite the opposite and there would be an embarrassment of riches funding streams to disburse and spend/share and enjoy.
And the man in his basement has surely been replaced by the typing voice in cyberspace with keyboards connecting with Clouds Hosting Advanced Operating Systems for Mad Command and Mayhem Control and/or Command and Control of Madness and Mayhem in the Pursuit of Peace and Happiness whenever in Quantum Communication Control State Server Phorm. And they be an altogether totally different kettle of phish that history has not yet a legend on.
And I just happened upon this today, [or did the system/computer feed it to me in particular and peculiar recognition of personal search and/or wider general interest ] http://cryptome.org/2013/10/mark-twain-takes-it.pdf It is a bitching prescient read.
Re: Brisk pace? Or merely a pretty face?
Somehow, my friend, I doubt an A/C mask would suffice to discuss that topic in the open. You know how it is. (Though you've my e-mail if you like.)
The noodlier thing though is that I have to question you emergent hyperhuman hypothesis. I don't believe it's possible without a truly massive metamorphic change. Genetic shift of the "that's no longer one of us" variety. That becomes a debate of Punctuated Evolution. Are we indeed tuned in to Darwin's Radio, or is it a gradual process, small mutation by each?
Put bluntly, I do not believe that a hominid of our design with roughly our cranial capacity is capable of being a modern Leonardo. No matter how you arrange the internal physics to maximize neuron density, there simply isn't enough physical space in there to provide the data storage and processing requirements to be Leonardo.
Baring a fluke of evolution we're not going to see some ubermensch crawl forth from the genetic underbrush and lead us all into Jobsian salvation. I'm also not seeing the singularity approaching with any real alacrity, so the One True Wikian seems to be a stalled path as well.
I'd love to delve further, but that requires blitting from the back buffers. A side channel will have to do.
Message Received and Understood .....
Apparently, every man and his dog are reading sensitive emails, TP. Secrecy in this day and age is a luxury which only the intelligent can provide with novel systems which beta test security protocols for future fitness of purpose.
It appears that knowledge sharing has become quite disruptive and that empowers and initiates all kinds of strange powers and new great games plays.
Why would you cache writes locally?
Any decent SAN has an array with tons of cache, unless you have a massive amount of writing going on every write should report complete immediately after hitting the array.
Reads are what slow you down in a SAN, because much of what you want to read will have to be read from old fashioned rotating disk. While a host caches data in RAM, it is a lot cheaper to add a 300GB SSD to a host than to add 300GB of RAM for a big buffer cache.
That's not even bringing up the idea that one of the reasons you pay for a big expensive array is reliability and data protection. You'd need to mirror SSDs to gain similar protection, doubling your cost. If the server craps out in the middle of processing, some writes will be sitting on those SSDs (because if you waited for the array to report a write through to complete before you moved on you would gain no benefit from writing to a local SSD) That means you can't simply switch over to a backup server, because the SSD inside the dead server has the most recent data, not the array. I suppose you could have the SSDs in a rack mounted chassis dual ported to two servers for failover, but isn't that sort of thing exactly the reason why you invested in a SAN in the first place?
Re: Why would you cache writes locally?
It's all about proximity. With hybrid SANs - and I make extensive use of them myself in my own datacenters, not to mention I am a fan of Tintri - you are still hauling all that data across the network. That's latency. It's also another bottleneck. It's more to wire up (if you want more bandwidth), more cost, more to manage...
Local read cache is local. Practically zero latency, minimal network impact. Put it in and it goes faster. It's also something where you can spot accelerate or deploy en masse, according to your needs. Great big fast SANs are bloody expensive. They are even more expensive if you need things like high availability, replication, etc. They like to charge you all the more for the goodies and two hybrids is a pretty penny past two traditionals.
Read caching is cheap, it's easy, it's non disruptive. It's an easy option when your alternative is massively expensive SANs. This is why I think read caches are a great mass market play; they fit really well into those businesses that are of a size where "a new SAN" is a massive investment that takes a lot of consideration and months upon months of fighting for budget. Perfect for the SMBs and midmarket.
Now, if you want to argue "why put a write cache into the local hosts (thus creating another tier of storage that has to be maintained, etc", well, there's no way I can answer that for you. I work with Proximal. They do read cache. Anything I could say (positive or negative) about the write cache stuff would be open to accusations of bias. I can only recommend that you talk up Pernix and/or Flashsoft and ask your questions directly. They are going to be the ones who can give you the most well thought out response.
said sod it and bought 100% all ssd SANs
I wonder if this would help go faster still?
Re: my company
***warning, largely anecdotal evidence***
In my testing, the answer is yes. Now, I don't have a Nimbus or Violin costs-more-than-Monaco all-flash SAN. In fact, I have a RAID 5 of 8 Kingston Hyper-X 240GB SSDs. You'll note in the review that this array is perfectly capable of flattening a 10GbE link. Which is what I'm using, because of "too poor for infiniband or 40GbE".
I am also capable of absolutely flattening that 10GbE link while doing things like "testing 10GbE swtiches" or "trying to make Hyper-V do something strange." I test things. It's kind of my job.
Still, Not being made of money, I have to run my actual production VMs on the same hardware. So I use Proximal Data's Autocache with the 480GB Intel 520 SSDs inside my Fat Twin and Eris 3 nodes. The result looks a lot like the theoretical model I laid out in an earlier article, with the exception that one of the Eris 3 nodes has a Micron 1.4TB PCI-E SSD for Autocache, another has a 120GB Intel 520 SSD and there are some Synologies in play.
Long story short, when I am busy making the "SAN" cry, the Autocache-enabled nodes still bloody work. Sure, they do yeoman's work when accelerating a really slow setup (like a RAID 1 of 2x 7200 RPM SATA drives) or a mid-range setup (bigger Synology diskstations), but the truly surprising bit is that they actually provide worthwhile and noticeable improvement to the VMs that are stored on the SSD array.
Now, I'll admit, outside of latency-sensitive setups you don't notice host-cache at all when the SAN isn't congested. Why would you? But flatten the link and the difference is night and day.
It's a danger too. It gets to the point that you just sort of forget you're using it. Then one day, the SSD fails in the server and everything reverts to pre-cached speeds...it's painful. I promise you, you'll get the SSD swapped and cache turned back on in a right hurry.
...but at least the thing keeps on working through that. That's the part that ultimately won me over.
I never would have thought that it would be a more efficient use of the SSDs available to me use them as host cache instead of just lashing them together into another RAID and doing more high-sped central storage. Up until about two months ago, that had been my plan.
Now? Now that looks like a great deal of hassle for no huge advantage over what I'm getting from the host cache. So yeah, it works. Even if your central storage is made of fast.
So what about vFRC?
Trevor, you don't say much about VMware's vFRC. It's going to be tough to compete with free (assuming you already have Ent Plus, and many of us do) and supported so what's the future for the Proximals of the world?
I guess I'm wondering why I'd buy Proximal if vFRC is free? Have you played with vFRC yet?
Re: So what about vFRC?
I have played with it some and will be doing a more thorough review when my ESXi 5.5 licences come in. I believe that I even discussed the "but it's free" issue in the article.
The Cole's Notes? Proximal is easier to use, easier to administer, is more feature rich and (anecdotally) has far better read caching algorithms. This last has to be more thoroughly verified, but the "word on the street" from the kinds of customers that can - and do - test these sorts of things at scale is that Proximal's focus on read cache has paid off in that regard.
The future for Proximal Data in the face of vFRC is pretty simple: be the better mousetrap. Innovate faster, provide the better offering, work more closely with customers and work cross-platform. VMware might get to "good enough" one day, but in my opinion - and really, that's all you can go on until you've worked with vFRC's interface and design philosophy yourself - it's a long way from that just yet.
By the time VMware gets to "good enough" Proximal Data will have moved on to new features. Minimal risk write caching as one example. There's more that I can't exactly go into. I think Proximal is worth the money today, and I'm a stingy bastard who works with broke SMBs who never have budget.
Given what I know of the company - and the bright people working for it - I think that it will most likely still be worth the money several years from now. But you don't have to take my word for it. Go to the website and <a href='http://www.proximaldata.com/product/try_us_program.php">download the trial</a>. Put it side by side with VMware's offering.
That's part of the beauty of virtualisation, no? The ability to try this sort of stuff before you buy. Of course, I'm of the belief that if you try it you'll likely choose "buy"...but I'm an infrastructure nerd and I find this all quite fascinating and cool.
Re: So what about vFRC?
It's absolutely the beauty, yes. And as of Saturday night I'm seeing exactly what Proximal will do for my environment. :)
Fascinating and cool? You bet. Without that I'd of long ago found other work.
What kind art thee?
Deep-ending on one's provider could lead to rep-er-cussions of:
What kind art thee?
Do tell when thee cometh out of the blender, heheheh....
Doesnt sound right
Waaaait a minute.
we started with a big central computer with dumb terminals
then the PC happened.
Then we moved storage, processing, etc out to the cloud.
And now we're moving it back to the local machine again? Because, as was always the case, the cloud is mind-bogglingly slow outside of California?
Seriously, guys, just make your mind up!
"I remember very specifically writing an email that said if people ever realise what PernixData is capable of...."
This sentence strikes me as weird. What is the purpose of referring to earlier emails, even if you still could find them?
Don't you storage guys retain all emails you ever wrote?
Pretty sure that the e-mail is in my "sent items" folder which would be in the *.ost on at least two local systems, in the exchange server's data store and probably a dozen backups. Why?
I referred to the e-mail to give some context; that there was some back-and-forth here that had occurred but whose details are largely irrelevant. I'm unsure that anyone would care about the contents of the mail itself - beyond the snippet I paraphrased - thus see no point in reproducing it verbatim.
That said, the fact that Pernix views write cache as a new tier of storage is relevant. Not only do they view it as such, but it struck me as being so after even relatively brief testing. Conveying that is the important part; jibber jabber on the side band is...boring.