We continue with the story of the new NAS system at Chaos Manor.
From what I’ve read on Netgear’s support forum, the initialization of the RAID 6 setup can take a LONG time but since the unit isn’t in active use yet this will only be a mild annoyance.
Eric noticed a Netgear RN10400 RAID system was advertised by Fry’s, so he decided that would be appropriate for the project. (See http://goo.gl/lYAl2p for product details.)
There wasn’t so much a survey [of system possibilities] as a Fry’s ad offering the unit for approximately $100 off its normal price. At the time I grabbed it there were potentially four different places it might have gone. If I had used it as a media server at home I would have gone for a JBOD configuration to maximize capacity as nothing on would have been irreplaceable.
The initial setup [of the RN10400] was very straightforward. A full set of four 4 TB drives from Seagate (Model ST4000VN000) were purchased and installed in the carriers. This is familiar to anyone who has done much PC construction with cases that provide removable carriers. Once the drive is installed in the carrier, it slides into its slot and connects to the SATA interface.
Upon powering up the NAS defaults to a RAID 5 configuration and immediately sets about verifying the drives and creating the RAID. This would eventually result in a capacity of slightly over 10 TB of storage capacity. Netgear offers an app called RAIDar that searches the LAN for any of the company’s NAS products and reports their address and status. From there you can log into the NAS itself and see what progress it has made and specify a wide variety of options, such as user accounts, access limits, and backups to other NAS or offsite via cloud services. This is also where you’d break up the default volume to recreate in a different RAID configuration.
That part that got tricky was making sense of the instructions. I cannot say if Netgear has updated the firmware and the PDF manual hadn’t caught up yet or it had always been wrong. It also may be intended for a different model and put in this version by mistake. The method detailed for destroying the existing volume simply didn’t apply to the interface presented.
The System Setup and Configuration
After much consideration of the choices, I decided to accept safety over capacity and move the Netgear’s four 4 TB drives to a RAID 6 configuration. This effectively means that half of the volume’s capacity is consumed by the non-data use of sector to insure the volume can be reconstructed without loss if any one drive should fail.
Eventually I noted the gear icon that had appeared elsewhere and found it was an active link to bring up an otherwise undocumented menu. From there the task more or less matched that described by the manual. The four drives were applied to a new RAID 6 volume which would offer 7.6 TB after 60 hours of setup. They weren’t kidding. It took that long. A search of the forums on the Netgear site produced a post that explained the RAID 6 setup had far more overhead than RAID 5 or lower, and that this was one of the reason their higher capacity NAS products used Intel processors with specialized function for the task.
Another optional setting was ‘Bit rot protection.’ This tries to head off failures on the drive media before data is lost but they warn of a potentially dire performance hit. I turned it on and did note that when uploading the backups from the recently retired machines the throughput was frequently awful, often descending low enough to be measured in KB. Before the next big upload I’m going to switch this feature off to see if it is the culprit. Once the bulk of initial backups are made it may have little bearing on day to day use but we’ll just have to watch and see. I recall some line about those who would sacrifice performance for data security may end up with neither.
Because this is a fairly inexpensive model using an ARM-based SoC [ARM processor, System-on-Chip – think Raspberry PI, which also uses an ARM-based SoC], the process of creating the new volume was quite lengthy, coming in at 60 hours. High-end models most often use Intel processors with substantially better performance in the specific areas complex RAIDs place their demand.
SOC is System On Chip. ARM processors are rarely found all by themselves. The myriad licensees package them with other functions, licensed from other companies or created by themselves, to produce the device specific to the intended purpose. This has become increasingly common across the industry, especially in the consumer sector where the other functions are a given.
Thus most Intel processors these days have a GPU included, and numerous other functions that were formerly found in separate chips. Due to the way ARM’s business model works, an ARM-based device is likely to be far more customized than something out of Intel as it simply isn’t worth Intel’s time to pursue markets below a certain size. Intel does have a line within the ATOM product range that is aimed at NAS builders but at a significantly great cost than ARM-based competitors. You’ll find these on higher end NAS models with a greater range of abilities than were needed here. For example, some high powered NAS products can act as a self-contained media library with the hardware to drive a TV all by themselves rather than relying on a client device.
As Eric worked on the NAS/RAID setup, he set up the volumes as RAID 6 instead of the default RAID 5.
The Netgear ReadyNAS 104 was populated with 4 TB Seagate NAS drives. This, when it finally finishes, will offer 7.6 TB of space to work with
I intend to write at greater length about it but was waiting until the change was done and more material might be added. There is a lot of capability in the device, making for a dizzying array of choices for how best to make use of it.
I’ll also make note that the 60 hours needed to perform the RAID 6 volume creation would be greatly reduced on higher end product using the Intel products that have been targeted at the NAS business. This model, and most like it in price and features, use ARM chips that cannot match the Intel products on the more computationally intense RAID configurations, according to Netgear reps in their forums.
Minor Annoyances Solved
There was some initial difficulty with the NAS alerts emails not getting to Dr. Pournelle’s account; they were going to Eric’s email account. Eric noted that, along with some additional info on the configuration.
Currently the email alerts are coming to me [Eric]. The mystery is why Jerry’s account wouldn’t work using the same setting displayed Outlook remains unsolved.
The NAS has a pair of gigabit Ethernet ports and is administered through an intranet web UI. It supports the installation of apps to add custom features but a brief survey didn’t make any stand out for Jerry’s needs.
Netgear warns that RAID 6 on this product range will significantly reduce write performance but with so few users on the Chaos Manor network this isn’t likely to be a problem. After the initial backup from an actively used PC the following updates should be quick, unless a complete image is being made every time.
The Swan system has an eSATA drive docking slot that lends itself better to that level of backup. A USB 3 external drive should suffice for AlienArtifact and the same drive could also serve the Surface Pro 3 since it has so little storage capacity compared to the desktop systems.
As the project continued, with status reports to the Advisors for their reading pleasure, Advisor Peter wondered:
I’m still curious to know how much of that gigabit capacity can be delivered over the Chaos Manor network, but it probably won’t come very close to the theoretical performance of the array itself, which should be well above 1 GBPS for reads, at least.
That’s really saying something, since I think all the computation is just XOR operations and table lookups in normal operations. But cheap ARM chips have few of the data-movement and vector-processing features of Intel’s cheapest processors, and that’s probably the real issue rather than “computation” per se.
Advisor Brian Bilbrey chimed in about initial and ongoing performance of the system.
A reduced write performance is primarily an issue on initial seeding of the device. Jerry undoubtedly has a lot of data he’d like to centralize, and getting all of that on the first pass will be the ordeal. Keeping it updated thereafter is less likely to be affected by the performance issues.
Eric continued the discussion:
I won’t be surprised if the network speed is much more of a limiting factor and the difference in write performance impossible to discern without much bigger loads than we’re ever likely to create.
Perhaps [it could be measured] if we tried to save an image from every workstation simultaneously. None of those machines has more than 1.25 TB of local storage (250-ish GB SSD and 1 TB hard drive) and most of that empty. (There is 192 GB on the Surface if you include the microSD card.)
If we did full backups every night there would be plenty of time to avoid overlap.
The saga continues in the final installment, as data is moved from systems, and older systems are considered for retirement. Let us know what you think in the comments below, and share this article with others.