Smart concentrators link conventional storage devices over IP networks
For many midsize companies, the major obstacle to adopting IP storage is not necessarily the fear of new technologies but the need to preserve a recent, conspicuous investment in conventional storage arrays.
That the benefits of networked storage might ultimately compensate for the loss of new disk arrays is an assertion difficult to substantiate. And it may not resonate well in the company boardroom.
Luckily, there is an elegant way out of this predicament: Virtualization routers from companies such as Stonefly Networks extend the life of traditional disk arrays by allowing them to provision storage for an IP SAN.
These virtualization routers serve as network enablers for conventional disk arrays, making it possible to connect those SCSI-bound devices to your GbE (Gigabit Ethernet) network. Add GbE NICs and iSCSI initiators to your server, and voilà! You have a SAN without resorting to unpopular write-offs.
In October, Stonefly released the i3000 Storage Concentrator model, which performs better than its predecessors, automatically balances network traffic over two GbE ports, and includes the exceptionally flexible Replicator mirroring software that can create both synchronous and asynchronous data replicas.
At a price of about $20,000 for two clustered i3000s, Stonefly offers one of the most affordable entry paths into networked storage, but you will have to add your own disk arrays, networking cards, and switches.
Old Dog, New Tricks
The i3000 is a 1U server based on Dell PowerEdge 1750 hardware running Stonefly management software on a Linux OS residing on a local drive. The units I reviewed mount dual-channel SCSI controllers from Adaptec, but you can use Fibre Channel adapters to connect compatible storage arrays.
The i3000’s HTTP-based management application is simple to use but powerful, with everything needed to manage the i3000 itself and storage devices, provision host servers with storage, set up a second unit for fail over, and monitor and remedy error conditions.
Unlike a monolithic IP SAN in which a single management application often controls everything, the Stonefly GUI controls the configuration of the i3000 but not that of the storage array. For example, you cannot change the RAID level or set a new LUN from the GUI — for that, you must use the software and tools native to that array.
Depending on the number of arrays and the frequency of changes, this separate management may be more or less of a nuisance. It’s the price you’ll have to pay for promoting your existing arrays to networked devices.
After powering on, the i3000 immediately discovered my array, which was configured to provide a single, 250GB LUN. The most common deployment is to let the i3000 grab storage from attached devices and segment that space in logical volumes, giving you all the benefits of seamless virtualization and storage provisioning.
Alternatively, you can instruct the i3000 to leave a discovered storage device in pass-through mode, which allows a server to access it as a directly attached LUN. This option can be handy during the transition from server-attached to networked storage, because it preserves existing LUNs (managed as a pass-through) while the i3000 concurrently manages unallocated space on the networked array.
Using one i3000, I began creating volumes from my managed LUN and assigning them to my hosts. Managing volumes is easy and intuitive but has one glitch: Expanding a volume takes it offline. It’s only offline for a short time, but it’s enough to break an application.
Otherwise, I could easily gain access to my volumes using the Microsoft iSCSI Initiator loaded on my servers and everything ran well; thanks to the Stonefly i3000, that SCSI array — once confined to attached hosts only — could now provide storage to any server on my IP network.
In this configuration, however, the i3000 is a single point of failure. To avoid this, Stonefly offers a configuration that pairs a second i3000 in a cluster, ready to take over if the first fails. The fail-over cluster is easy to set up and works smoothly, but the second controller is purely a stand-by and does not participate in nor improve performance of data transfers.
I simulated several different failures and the stand-by unit always resumed operations without disrupting ongoing data transfers. The i3000 GUI has also a “manual fail-over” option, a useful way to test the recovery procedure. I put the manual fail-over to work immediately when one of the controllers was not behaving as expected during recovery and had to be replaced. It worked nicely.
Purchasing a single i3000 or a fail-over pair also includes one server license for Replicator, and once you try this remarkable mirroring software, I bet you will be addicted as I am.
Replicator goes well beyond the usual list of required features for a mirroring application, including storage-independent, cross-server, coordinated replicas of related volumes; custom buffer size for asynchronous writes; and persistent repository for execution parameters. It’s extremely adaptable and well thought out.
In short, Replicator intercepts any application’s writes to a source and replicates them to a target disk, using safety features such as a journal and a dedicated memory buffer to make sure that the mirroring executes quickly and will survive connectivity failures. The copy of Replicator I reviewed worked on Microsoft Windows only, but Stonefly plans to release a Linux version soon.
Replicator is a tool that can add an extra layer of resilience to your volumes, creating reliable mirrors of your data across the room or miles away. It’s one of the best mirroring products that I have seen — a worthy addition that makes the i3000 Storage Concentrator a solid and reliable option for your storage network.
Overall Score (100%)
|i3000 Storage Concentrator||8.0||8.0||9.0||9.0||9.0||9.0|
This weekend's Windows 10 upgrade has users angry, and it's unclear if the ploy will continue
Speaking at the O'Reilly Fluent conference, Eich also endorsed the Service Workers mobile app...
You don't need a tinfoil hat, either. Opportunists have exploited consumer fears to create an industry...
Misconceptions and 'best practices' may have your team spinning wheels rather than continuously...
Today, Databricks subscribers can get a technical preview of Spark 2.0. Improved performance,...
The latest version of the Python runtime sponsored by Dropbox uses garbage collection to enhance...