Crazy NAS idea

[mikemex]

I've been using a DNS-323 for a while with two 2TB 5400 rpm drives in RAID 1. So far, it works OK, but it's starting to show age, both in capacity and speed.

Speed in particular is quite an issue when handling large amounts of data. Real world speed for my device is around 20 MB/s and, the last time I really needed to move a lot of data (backup a full TB of stuff) I had to leave the computer running for a couple of days. Back in the day this seemed "fine" (I once downloaded a full 650 MB CD-ROM image over a phone line with a 33.6K modem, so you know I'm a patient person) but put in perspective, it now seems like quite poor performance. I can only imagine how bad this problem can be when you really scale things up. Right now I have 2TB of space, but what about 16TB? Capacity is always up but not speed.

So I've been thinking (and this is just hypothetical; say, in case money wasn't an issue): what about building an SSD based NAS?

Thing with SSDs is that price doesn't go up linearly, so a 4TB/8TB drive isn't 2x or 4x more expensive than a 2TB drive, but many times more. A solution I see is to find the sweet spot (it seems to be 2TB) and use tons of drives to build a RAID array.

They sell Micro ATX NAS cases with 8 SATA bays. Micro ATX size is important because ITX motherboards are usually port starved with no real options for expansion. This is far less of a problem with modern motherboards equipped with M.2 slots (there are M.2 to SATA adapters) but a problem still, since the array would eventually be bottlenecked by the network connection. And not many motherboards come with fast ethernet (2.5G or 10G), so having extra slots for expansion seems important to keep things balanced.

There are adapters that allow you to install two M.2 drives and perform RAID internally to an external SATA connection. The idea is to buy 8 adapters and put two drives in each, for 16 in total. By doing RAID themselves, they take a lot of overhead from the host PC. I'd be doing RAID 1 internally and RAID 0 externally. I don't want to do more sophisticated RAID levels (such as 5/6) because they wear SSDs faster and, having to buy 16 of them, I'd probably go for the cheapest possible option.

You may wonder: why SATA? That's because there is the issue of power consumption and dissipation. In this regard, SATA interfaces are much safer as they don't allow devices to consume too much power. And having several SATA drives drives instead of a few NVMe also means having several times more surface area to dissipate heat, so less chance of overheating.

Finally, being a network attached device, you can keep it in a safe physical place. So no spilt cups of coffee anxiety (unlike a drive in an USB case physically attached to the computer).

So let's see (assuming I can't find a better price):

$400 for the case.
$500 for the rest of the components (motherboard, CPU, RAM, extra cards, etc.).
$440 for the adapters (8x55).
$2,560 for the drives (16x160).

$3,900 for 16TB of mirrored storage. It's a lot of money -yes- but it's also al lot of features you can't get from simpler solutions. For example, regular NVMe drives only offer you tons of speed as long as you stay with transfers shorter than the buffer size; speed from the actual flash is much slower. So transferring large amounts of data will not maintain such levels of performance for long. With this device, you should get (at least in theory) sustained transfer rates of about 1GB/s. So the backup that took me two days to complete before would be done in 15 minutes.

What do you think? Am I crazy and overthinking this?

[dr_st]

NAS, redundancy and safe storage is of course the way to go. However, I find it hard to justify spending money on SSDs for backup storage purposes. 20MB/s is atrociously slow this day. Modern HDDs, even 5400RPM ones can easily get over 100MB/s in sequential access, and some faster drives can get almost to 200MB/s.

How often do you find yourself moving whole terabytes data? At a modest average speed of ~50MB/s transferring a whole TB should take ~5-6 hours.

[SMA]

Even with the fastest NAS unit in the world, the speed will not get higher than the speed of your LAN connection.

[mikemex]

Even with the fastest NAS unit in the world, the speed will not get higher than the speed of your LAN connection.

Of course. And the speed of the drive in the other end, too. Actually, I was thinking about the machines in my local network operating in a mainframe model: files never being stored locally, but accessed remotely through the network. With a good Ethernet connection, this shouldn't be much different than working from a local hard drive.

NAS, redundancy and safe storage is of course the way to go. However, I find it hard to justify spending money on SSDs for backup storage purposes. 20MB/s is atrociously slow this day. Modern HDDs, even 5400RPM ones can easily get over 100MB/s in sequential access, and some faster drives can get almost to 200MB/s.

How often do you find yourself moving whole terabytes data? At a modest average speed of ~50MB/s transferring a whole TB should take ~5-6 hours.

Not too often. This is just hypothetical

I'm surprised I have to tell you this, but the main advantage of SSDs is not raw transfer rate, but latency. Spinning drives may offer decent transfer rates when working sequentially, but they have terrible seeking times. Without low latency, a high transfer rate can average to a much lower number in real world conditions.

That's pretty much the whole idea behind TAR files: a tape device seek time is so slow that you better make sure files are pre-glued into a monolithic entity, ensuring only sequential access is necessary for either storage or retrieval.

[dr_st]

I'm surprised I have to tell you this, but the main advantage of SSDs is not raw transfer rate, but latency. Spinning drives may offer decent transfer rates when working sequentially, but they have terrible seeking times. Without low latency, a high transfer rate can average to a much lower number in real world conditions.

Which only comes into play in random access and when transferring many small files. The first is not really relevant for a backup process, and the second can be, if you copy many small files raw instead of compressed into one large archive. That depends on your workflow.