After 55 years of living in and around New York I made the plunge and moved lock, stock and testing lab to Santa Fe, NM. My reasons are too many and in part too personal to go into in detail here but one major reason is having made the transition from consultant to analyst I could no longer justify the cost of living and working in as expensive a place as New York.
When I was a consultant the New York area was the perfect place to be. There was no shortage of small companies that had grown their networks or storage systems beyond their IT staff’s ability to understand them or large companies on wall street looking to implement the latest technology before the big consulting firms had anyone the understood them. I could network with my fellow geeks and take gigs when the phone rang.
Now that I’m an analyst, and running what I in all modesty have to call the best independent test lab in the business, my clients are primarily the storage vendors and the big boys all have offices in New York the people that hire DeepStorage to test and write about their products and technologies aren’t in New York they’re in the storage hot spots of silicon valley, route 128 in Mass and Colorado or scattered across the rest of the world like my friends at Amplidata who were the perfect hosts during my recent visit to them in Ghent, Belguim.
Most of the time it doesn’t matter where my clients are located, we plan our projects online, they send their software and/or equipment to the lab, we test it and write a report that says how well it works.
So Sept 26th the Mayflower moving truck came to the lab in Lyndhurst NJ and took all our equipment to our new space in Santa Fe. We have a little over 1000 square feet of industrial space, with the requisite 3-phase power, that we’re now converting into the best little test lab in the west.
As we were planning this little adventure we realized that we’d have to build out the actual lab space, which since we test equipment that goes in a data center that means we’re building out a small data center. In my years as a consultant I’ve seen some data centers where it was clear the server and network guys were told to build a data center and they just had no idea what the right way to do it was. After all they were server and network guys and never paid attention to the state of the art for server cabinets, UPSes and PDUs let alone CRAC (computer room air conditioning).
So we decided to take the unique opportunity of us building a new data center from scratch and turn it into an education project so we could show you server, storage and network folks out there what your options are when management tells you the company’s moving and you need to build a new data center.
In addition to our own testing we offer remote access to the lab for other writers at NetworkComputing, other delegates to Tech Field Day and some of our friends so we’re going to be looking for the most flexible solutions that will let us run a fully lights out data center.
We’re inviting the vendors that make our favorite data center kit from electrostatic discharge flooring to overhead cable trays to both keep us up to date as new products and categories develop and to supply their gear for use in the lab so we can write about it and show you how you can make your data center better by using the latest gear.
For each product category we’ll show you what the current state of the art is and what your options are if you have a limited budget.
Over the next few months I’ll be blogging here, and at NetworkComputing.com, walking you through each step of the process. We’ll also be shooting some video on YouTube so you can watch as we put everything together.
While you should be able to count on an all-solid-state storage system to deliver better performance than an array of spinning disks, the differences among solid-state systems are at least as significant as those among disk-based systems. While most storage professionals can generally identify a disk array as fitting into the high-end, midrange or value market segment, many have not yet recognized that the market for all-solid-state arrays also breaks down into market segments.
While performance is the main reason we buy any solid-state storage system, some users–including many employed by three-letter agencies of the federal government who would have to kill us if we ever found out exactly what they were doing–are looking for the ultimate in performance and don’t care about anything else.
If speed is what you need and cost is no object, the solution is a rack-mount DRAM SSD from vendors like Texas Memory Systems or Solid Access that can deliver 10 to 20 microseconds of latency for a measly $700 per gigabyte. DRAM SSDs are the drag racers of the data center: Like a dragster they’re really fast but only good at one thing, and they generally lack redundancy, capacity and comfort–that is, management functionality.
If you want enough space for your girlfriend, or more than about 1 Tbyte of data, you should look at flash or hybrid DRAM and flash systems from the likes of Violin Memory or Kaminario. They’ll deliver millions of IOPS and latencies on the order of 100 to 300 microseconds, but are still limited in terms of capacity and management flexibility, topping out at well under 50 Tbytes. These are the Lamborghinis and Bugattis of the data center. They’re so fast that ordinary storage administrators lust after them, but they don’t have room to take the kids and luggage to the beach for the weekend, and suspension so stiff that driving them in the city is a painful experience.
Mere mortals have to settle for all-solid-state arrays in the price-performance category. Vendors like GreenBytes, SolidFire, Pure Storage, Nimbus and Whiptail have feature-rich arrays with redundant controllers and support for multiple storage protocols, snapshots and/or replication. Many of these systems are price-competitive, with arrays full of 15K RPM disks as the vendors use data-reduction techniques like compression and data deduplication to squeeze more data into their expensive SSDs.
These systems deliver only hundreds of thousands of IOPS and can have peak latencies as high as a millisecond. Since a 15K RPM disk drive has about 4 milliseconds of latency, and it would take thousands of drives to get 100K IOPS, these systems are still blazingly fast in comparison to even a high-end array full of disks. I consider the price-performance arrays the BMW M5s of the storage world: While they can’t run at more than 200 mph, like the Lamborghinis, they seat four and have enough trunk space for more than just a set of golf clubs.
Shopping for an all-SSD array is like shopping for a fast car: You’ll have to decide how fast is fast enough, remembering that speeding up storage by a factor of five or more may just reveal bottlenecks in your servers or network.
Disclaimer: SolidFire is a client of DeepStorage LLC.
A bit more than a year ago the folks behind the LTO tape format standards, primarily IBM, with some contributions from HP and Quantum, added the Linear Tape File System to LTO’s feature list. While some niche markets, primarily the media and entertainment business, have adopted LTFS, it won’t live up to it’s promise without support from archiving and eDiscovery vendors.
LTFS divides a tape into two partitions, one that holds the file system metadata, and another that holds file data. With a little bit of software on a server, LTFS tapes look to the server like disks, and any application can write files to a tape just like they can write to a disk. LTFS isn’t the first attempt to make tape look like disk, I remember the Backup Exec group at Seagate Software showing me a tape file system in the ’90s.
The difference is that LTFS is standardized, making LTFS tapes a standardized data storage and exchange medium. Now you and I have switched from mailing floppy disks and DVD-R disks to using Dropbox, Sugar Sync and Yousendit, but when you need to move many gigabytes of data from one place to another, it’s hard to beat the effective bandwidth of a box of tapes.
A box of 20 LTO-5 tapes holding 24TB of data will take roughly 12 hours to get from New York to San Francisco via overnight courier. That works out to an effective transfer rate of 2TB/hr or 4.4Gbps. If we allow 12 hours to spool the data to tape, which is about how long it would take to move from a disk stage to tape using a 6-drive tape library, the effective bandwidth is still 2.2Gbps.
Even if you were getting 20:1 data reduction through data deduplication and compression, you’d need a 100Mbps link to match the bandwidth of that small box of tapes replicating that amount of data across a network. Twenty-to-one data reduction may be achievable for backup data, but archives don’t have nearly as much duplicate data as backup repositories, since each archive has just one copy of each data object. Archives of rich media, be they check images, photos from insurance claims adjuster’s digital cameras, or medical images, don’t reduce much at all, making that Fedex box even more attractive.
Without LTFS you’d have to be running the same backup application at both sites to send data via tape as each backup, or archiving, application writes data to tape in its own proprietary format.
In addition to providing a standard interchange format, LTFS promises big advantages to storing data in a standard format over a long period of time. If your archive program stores each object it archives as a native file in an LTFS file system, you’re not dependent on a single vendor for the data mover, indexer, search engine and litigation hold functions. If your archiving vendor discontinues your current product, like EMC did with Disk Extender a few years ago, you can switch to another product and have it index the existing data without having to regurgitate it to disk and ingest it into a new archive. If you have trouble locating data, you could point a Google appliance at the LTFS repository and use Google search to find the relevant data.
We as customers should start pressuring our archiving vendors to support native LTFS as a repository option. Some vendors will respond that they support LTFS, since they support any NAS storage, but most archiving solutions store their data on disk in proprietary container files. While compressed and single-instanced containers may have made sense on disk, the lower-cost-per-GB of tape makes the flexibility of a standard storage format worth the extra storage space it takes up.
After half a century of living and working in the great city of New York and its environs, I decided it was time for a change. In the first week of September we’ll be packing up the DeepStorage labs loading the truck and moving to beautiful Santa Fe New Mexico.
For the 25 years or so I made most of my living consulting with end-user companies; usually helping them clean up the messes that got created as they integrated new technologies faster than they understood them or grew beyond their IT staffs expertise. New York was a great base to do that from as there were thousands of companies, some of them quite confused, between 59th St. and the battery
Since I’ve shifted my professional focus from end-user consulting to hands-on testing and analysis through DeepStorage, location has become a lot less significant. I meet with vendors, who are potential customers for our services, and my peers, at conferences and events from VMworld to Tech Field Day. We agree that on the parameters of a testing project or report they want written and they ship the equipment we are going to test to the lab. Since this means I can run my new business from anywhere I started to question where I actually wanted to be.
Much as I love New York, and took full advantage its charms in my youth, as my beard has turned gray I spend less time at the theater, concert hall, or club. Since I’m also dedicated myself to getting back in shape, okay some shape other than round, I’m also spending less time at the four-star restaurants. While I loved the fast pace when I was young enough to keep up with it, today it wears on me more than it energizes so I’ve decided that it’s time to turn the tension level down from 11. Add in that New York is, with the possible exception of San Francisco, the most expensive place in the country live and the time has come to move.
My ex-wife and best friend, how many people can say that, grew up in Santa Fe, then moved back there after getting her PhD from the University of Robin Hood in Nottingham UK. Over the many visits I’ve made there I felt surprisingly comfortable after years of traveling a lot and usually wishing I was in New York; so Santa Fe is the place for me.
I just signed the lease for the new space for DeepStorage’s new lab and office space. We’re now entering the exciting construction phase. Since construction on the New Jersey lab which was originally scheduled to take 90 days ended up taking almost 9 months I’m planning this part in excruciating detail. Hopefully building permits in New Mexico will be somewhat more straightforward than the process here in a city where our last three mayors have all done time at Club Fed.
In addition to getting the normal permits for walls and the like I’m also of figuring out exactly how to build the data center. I’ve learned a lot working in way too many bad data centers. Server rooms with window air conditioners, bad lighting, asiles too narrow to open the doors on the racks on both sides of the asile and of course the ones where an overloaded circut breaker causes a massive cascade and total loss of power.
Much of the problem is that SMEs give the poor IT guy a room and tell him to turn it into a data center . The poor IT guy who’s spent his whole career learing about switches, routers, servers and storage has no idea how to choose air conditioning, power distribution and the other physical infrastructure portions of the data center.
I’m going to be putting together a series of blog posts under the heading My Next Data Center where I’ll discuss the choices we made, discuss the good options available that we chose not to use and generally provide a roadmap to building a small, in our case 6 rack and 20KVA, data center to provide the level of reliability and degree of flexibility most organizations need without breaking the budget
I’m excited at the prospect of making the big move we’ll basically be closing the lab just before VMworld packing everything up and reopening in the middle of September
Two of the leading analysts in the storage business, Jim Handy The SSDGuy from Objective Analysis (who knows more about Flash from the component level than anyone I know) and Tom Coughlin (Who’s plugged in to the hard disk and entertainment businesses) are running a survey of IT professionals to get a rough idea of their IOPS requirements (real or imaginary – if they don’t know for sure a guess is good!) It’s broken into broad, “Orders of Magnitude” ranges:
Do you need a hundred?
Ten thousand? etc.
Take the survey at http://www.SurveyMonkey.com/s/KGKVR6X and help a couple of the good guys out.
My buddies Rick and Nigel had me back on the Inforsmack podcast this week reviewing the 42 announcements at last week’s EMCworld. As usual we had a good time and discussed the goings on in Vegas, after all not everything can stay in Vegas.
Subjects include EMC marketing czar Jeremy Burton’s love of spectacle and cheese, EMC co-opting their competitor’s best features and XtremIO.
Check out the podcast at http://infosmackpodcasts.com/emc-world-2012-review/
Now that I’m a vExpert for 2012 and elbow deep in researching our Server Side Solid State Storage report I’ve proposed a session at VMworld on the subject. While there are about a dozen of the 1300 or so proposed sessions on solid state storage this is the only one not being proposed by a vendor of SSDs or flash based storage system. Flash is revolutionizing the storage business and I’m excited about the possibilities adding flash to the server brings to those not ready to replace their existing storage infrastructure in a forklift upgrade.
This session will explore how users can integrate server side caching into their virtual server and VDI environments. We’ll examine the different architectures and caching methods used by the dozen or so server side caching products currently available. We will then look at the pros and cons of these approaches to solve common storage performance problems in virtual server and VDI environments with particular attention to how the caches integrate into vMotion and other management features.
Unlike other conferences I’ve spoken at over the years like Interop where sessions are chosen by track chairs or an advisory board VMware lets attendees vote for the sessions they’d like to attend. If you want a vendor independent session on making flash work in your data center to speed up your vSphere and View environment please vote.
As they say in Chicago, and here in Hudson county New Jersey, vote early and vote often.
The session is number 2247 Using Server Side Flash Caching
Cast your vote at https://vmworld2012.activeevents.com/scheduler/publicVoting.do