Nexsan Technologies — Green Storage

The response to our announcement last week of the first members of our new E-Series storage family has been truly amazing.  In Part 1 of this series, I told you a bit about our design philosophy, so today I’d like to introduce you to the technology and some of our newest innovations.

The first thing that people notice is our new Active Drawer™ technology.  It’s something our engineers started working on over five years ago, as part of our quest to deliver higher and higher density without the sorts of painful compromises that had afflicted our competitors’ earlier efforts.  When we announced a chassis in 2002 that packed a whopping 42 disks in 4U, it was a game-changer.  As time marched on, however, it became clear that 60 drives in 4U was the next milestone, but we just didn’t like the idea of having to slide a honkin’ big 300 pound chassis three feet out of a rack to change a drive (a process likely to involve at least two people, a stepladder, gigantic metal barn doors, and quite possibly an ambulance).  There had to be a better way.

There were a few vendors that tried to improve density by creating disk packs that could plug into a shelf, but all the drives in a pack had to be taken off-line in order to change just one of them, necessitating a huge RAID rebuild once the pack was re-inserted.  Rumor has it that at least one of them attempted to speed up the rebuild by scoreboarding changed stripes, but I guess for whatever reason they gave up on that approach.   Another problem with this approach is that in order for the canisters to be safely handled by one person, they could only carry a relative handful of drives, which means that a lot of them would be needed to build a 60 drive unit, increasing costs and reducing reliability.

After years of development and experimentation, we settled on a drawer system, with three drawers each with 20 drives, each drawer permanently connected by a redundant pair of custom power and data cables that slide in and out as the drawer moves.   We call it Active Drawer because our system is able to sustain I/O to any of the drives in a drawer, even when the drawer has been pulled out to add or change drives.   This means that the other drives are not impacted during a maintenance process.

It is a common  misconception that the goal of high density is to reduce rack space requirements.  That’s nice, of course, but it’s not the reason we create these amazing designs.  Rather, the real win from high density is in the reduction of components and complexity required to implement any given amount of storage.  Less parts means greater reliability and lower power; less complexity means fewer installation headaches and quite possibly a big win in long term reliability as well.  And it’s usually cheaper.

In the spirit of “no-compromise”, we wanted to ensure that the design could accommodate even the most power hungry 15K SAS drives, while at the same time being an economical solution for SATA drives and a worthy home for ultra-high IOPS SSD drives.  This was partly an electro-mechanical challenge, necessitating state of the art power, cooling, and vibration-control technologies, but also we needed a major jump in RAID controller technology.

Our latest RAID controller design is the beating heart of the new E-Series products.  It’s much faster of course, but amazingly it uses less power and is half the size of our previous generation.  Much of the improvement is due to our transition to PCI-Express Gen2, but also we completely re-designed our custom hardware RAID engine, and moved to a new generation of support silicon.  We also went to a dual core CPU, but unlike many of our competitors who need gigantic, power-hungry, processors to do their RAID calculations as software subroutines (RAID 6 or other multi-parity schemes in particular are huge power pigs when you do it in code), we took the time to create a custom-designed hardware RAID card with a custom-designed RAID chip that does all the heavy lifting, so our processor is relatively lightly loaded.

Another cool thing about our new controller design is that we put the host I/O on pluggable cards.  This week we announced our dual 8 Gbit Fibre Channel I/O card, but others are coming right behind it including dual 10 GbE iSCSI, and a dual SAS card (the latter for DAS applications which are surprisingly common among our customer base).  We plan to let moderately tech-savvy customers change the cards themselves if their interface needs change, or we can swap it various other ways.

There’s a lot to love about our new E-Series family.  We can pack 1200 TB into a single rack, using field-proven enterprise-grade 2TB SATA drives (1800 TB when the 3 TB enterprise-grade drives come out in a quarter or two), or if SSD is more your thing we can deliver about 1,000,000 truly random IOPS in the same space.  For archival applications, or applications that have periods of downtime like D2D, we have added an even deeper level of power savings called AutoMAID® Level 4 that can reduce the power consumption of this 1.2 petabyte rack of storage to about 2 KW compared to alternative vendor solutions which consume as much as 20 KW for the same capacity.

The first three members of the E-Series family are the NEXSAN E18, which holds 18 SAS or SSD drives in only 2U (these are 3.5” form factor drives, mind you, so this is very dense indeed), the NEXSAN E60, which holds 60 drives in only 4U, and the NEXSAN E60X which is the expansion chassis for the E60 (giving you a total of 240 TB in 8U including redundant RAID controllers and power supplies).  In the next blog post in this series I’ll tell you more about how our E-Series drives efficiency to a new level.

Sometimes I feel like Steve Jobs – just when some of my competitors have caught up with some of our specs and copied our features, we release something new that leapfrogs ahead and sets the new standard.  Take a stroll over to our newly-updated website and have a look – we have videos, white papers, and all sorts of new stuff.  We’ve invested heavily into the web infrastructure to make our site much more dynamic and useful, so let us know how we can help you.

Today Nexsan introduced its most advanced storage family yet – the revolutionary NEXSAN E-Series.  We will get into the feeds and speeds in Part 2 of this blog post, but first I think it would be useful to explain the strategic thinking behind the new family.

We have watched with a mixture of amusement and incredulity as the “big iron” vendors try to reposition their high-end offerings as being optimized for the mid-market, while at the same time the commodity vendors in the market are saying that because they have more “tick boxes” than before, that they should be seen as serious competition in the midsize storage arena.  Either way, the fundamental error is the same.   The mid-market has its own set of challenges, and the only way to address them is from day one of the engineering project, focus like a laser on what the mid-market really needs.

Nexsan has been focused on the mid-market for over a decade, and we have observed how much pain these folks have encountered when trying to adapt their business to the costs and complexities of a high-touch enterprise storage solution, or dealing with the blind panic resulting from reliance on the “black hole” tech support one encounters with products from commodity vendors.

A mid-market customer cannot often afford to add storage specialists to their IT department and also send them to weeks-long vendor training, nor can they afford the armies of Professional Services consultants that are often recommended by the Enterprise vendors as an alternative.  Especially when you consider that these vendors usually want a huge annual spend on maintenance and of course a forklift upgrade after Year 2.

A well-crafted storage solution for the mid-market must be designed around an IT department that has few if any dedicated storage managers, and thus must have an installation and management process that’s virtually free of arcane industry jargon.  With shrinking budgets these days, it’s entirely possible that the IT Manager will be racking the gear himself (or herself) as well as configuring it and perhaps even maintaining it (to avoid the annual tithe to the vendor).  Creating a storage product that fits these needs cannot be a simple afterthought.

In the next installment of this blog post I’ll tell you about the new products we are announcing today, and why we think they will set the new standard for mid-market storage.

As you may have heard, we just announced our DATABeast storage solution, which incorporates the energy saving technologies for which we’ve become a little bit famous.  Today I’d like to offer some suggestions for optimizing AutoMAID in these and similar environments.

For those unfamiliar with AutoMAID, a brief tutorial:  imagine if you took a RAID shelf and added an optional mode which spins down arrays that are not being accessed, and that the spin down process is divided into three steps which save progressively more power.  You don’t access it for a couple of minutes and it goes into the first saving level which unloads the heads, saving power due to reduced aerodynamic resistance in the drives, then if it stays idle for a while longer, it slows to half the normal RPM, then finally if it’s idle for a long time it’s spun completely down and put into a low power sleep mode.

Many new storage offerings have the capability to virtualize huge amounts of storage and present it as a flat file system or a flat sea of blocks.  The DATABeast, for example, can build single physical LUNs as big as 64 terabytes, striping the data across many individual RAID sets in multiple disk shelves.  Very tempting to put all your storage in one huge volume, isn’t it?  But just because you can, doesn’t always mean you should.  Especially when trying to save power!

The problem is, that any type of striping arrangement means that just a small handful of host operations will very likely result in at least one operation to each RAID set, which means you are not saving much energy.  Many operations, such as read commands which don’t happen to hit our cache, will require us to spin up at least one drive of the RAID set and possibly all of them.  The same effect can happen when using LVM software on the host if you allow it to stripe data across all the RAIDs, or if you combine all the RAIDs into a ZFS pool.  This is fine if you have a deep archive application which is 100% idle for long periods of time, in which case the arrays will unload the heads, then go to half speed, and finally stop until you start hitting the storage again.

By arranging your data such that infrequently used data is pooled together, the pool may enjoy long periods of zero activity, in which case the AutoMAID savings are maximized.  Similarly, resist the temptation to store active data on arrays which are mostly there for future expansion – let them stay asleep until you need the TB.  DATABeast can let you add the arrays to an existing storage pool on the fly, whenever you need it, so there’s not much reason to add arrays until the space is required.   Our thin provisioning technology allows seamless expansion of existing volumes, with convenient operator alerts when approaching the watermark setpoints.

DATABeast also has tools which allow you to migrate your data from one storage pool to another, which will assist in maximizing your energy savings.  The reporting tool will provide information on utilization, which will help streamline your management decisions.  If you want to check your handiwork, each array gathers AutoMAID “efficiency” statistics, which will tell you how close you are to achieving maximum power savings.

In future blog posts we will dive a little deeper into power savings and offer additional ideas on how to streamline the process.  It’s easy being green!

When Nexsan introduced the concept of Enterprise ATA in early 2001, we were thinking primarily about how much money could be saved in the data center by using low-cost desktop hard drives to store Tier 2 data and D2D images.  Soon, however, we started hearing about how ATA storage was saving power.  Raw electricity wasn’t particularly expensive, but once you factor in the costs of UPS protection and redundant air conditioning, the picture starts to change, especially if you are operating near the limits of the existing power and cooling infrastructure.  I was recently told by a data center manager in midtown Manhattan that if he wanted more power, there was a three-year waiting list and he had to pay to rip up the streets.  Not good.

About five years ago, we started to specifically engineer storage for power efficiency, which led to our revolutionary 42-drive ATABeast (and its descendant, the SATABeast).  We basically shared two power supplies and two RAID controllers across what is essentially three shelves-worth of disk drives.  Saves money and increases density of course, but also saves a lot of power.  We continued to improve efficiency in later releases because disk drives were improving in density and efficiency, but also by working on various power optimizations, which led us to develop AutoMAID.

If you aren’t familiar with AutoMAID, it’s the idea that when a drive is not in use, it is gradually spun down (the heads are unloaded first which reduces air resistance in the HDA, then after a further period of inactivity the drives are reduced to half their nominal RPM, and then if it remains idle we put it to sleep).  AutoMAID takes advantage of the fact that typical Fibre Channel command timeouts are in the realm of 60 to 120 seconds, which is plenty of time to pop the drives back into full speed mode if an I/O request comes down the wire. 

The key point here is that the array still acts like any other disk drive or RAID from the perspective of the host O/S, while opportunistically looking for ways to save power if a drive or a bunch of drives are not needed at a given point in time. This is often the case in D2D or deep archive applications or in applications which are not used 24×7. At the same time, the way we’ve executed AutoMAID means not compromise on the ability to return to maximum performance should I/O recommence.

Green storage is important to Nexsan and is included in all of our products at no extra cost.  We’ve been green since 2001 (long before global warming and green messages began permeating our daily lives) and we know how to provide green storage without compromising performance.

The industry is catching on to the Nexsan AutoMAID idea labeling this class of technology MAID 2.0 or next generation MAID.  MAID 2.0 embraces the idea of intelligently idling disks based on application and need while negating the performance trade off.