Network Attached Storage components

Having read the recent Choice review of a NAS box and the buying guide I have more questions.

First an observation. It looks to me that a single drive unit that was reviewed is not really that valuable as you get no benefit in either data security or speed that can be obtained from multi-drive units. For many purposes adding another internal HDD to one computer on your home network and sharing would achieve the same result at a fraction of the cost. Sure the data isn’t necessarily on line 24/7 but it isn’t hard to pick the machine that is on most of the time and to deal with a little inconvenience, especially if your purpose is backup.

My projected main use is as as backup for my ever growing photo collection. Secondary uses are to centralise a smaller collection (that may well grow) of movies and music. I plan on a two drive unit with the drives mirrored at a minimum. I am not building a server farm!

First question: where is the slowest component in the system? Is it the NAS box, the HDDs or elsewhere? Put another way, is it worthwhile paying for faster storage or is the limit likely to be elsewhere? If the limit is the drives, which gives more bangs per buck, faster drives or adding drives to stripe them?

Secondly, should you buy specialised NAS HDD drives? Some ‘experts’ say that you should because they are optimised for the type of load and will perform better and be more durable. Others say that this might be technically true but the degree of benefit does not match the price, that for domestic use modern quality HDDs are plenty durable and the extra money just isn’t worth it.

2 Likes

Noted you are looking to a simple and cost effective solution.

For the everyday home PC and network my previous experience at home suggests.

Without going into the next level of tech it is very likely the speed of your home network will be the limiting factor.

A modern mechanical hard drive on a PC standard internal Sata port might be specified with 100-150MByte speeds. Faster than gigabit Ethernet can manage and also faster than any realistic speeds achievable on a home wireless network.

Possibly true if you are in an office that goes flat out with a heavy workload, 12hrs x 6 days.
You could also just buy a quality PC hard drive from a reputable brand, and spend the savings on a reliable second backup drive.

Whether paying extra and buying a drive optimised for server use or NAS is necessary?
It might be more about your budget than any other need? If you get a genuine longer warranty it still might not be worth it? Technology moves on. They all fail! High cost server drives can still fail early or randomly as they age. It is supposedly just less likely. There is no substitute for independent backups they say!

Notes:
Speeds of reading and writing files also depends on the file management system and size of the files being read or written. This is partly an operating system constraint as well as hardware.

If you were reading and writing randomly lots of small files continuously, perhaps you might see some benefits of a fast solid state drive for storage? It’s again about budget, and there are read write limits on solid state drives. General commentary is that for the average home user that should not be a concern. A percentage of solid state drives do fail early and randomly. Something about always having an independent backup echoing!

I’ve had two NAS devices. A single drive Maxtor (pre 2009 vintage) and a newer Unix based box (2012) with dual sata 2 drives in raid. Using 1Gbit Ethernet, the network speed was typically the limit when moving large files between a Linux based PC and faster NAS. Any file transfers to/from Windows or using Windows Network sharing and no NAS were considerably slower, due to Windows being Windows. We now use external USB3 hard drives, and find them fast enough, and more flexible.

An external NAS even with just one hard drive will use less power in standby or when running than a PC. If you rely on wireless networking you can connect the NAS directly using Ethernet into the router which will reduce the loading on your wireless network noticeably. You can use an external USB drive attached to the NAS as a convenient removable backup. Depending on encryption options and authentication a NAS may offer better data security for critical files than vanilla Windows? Not compelling reasons to use a single drive NAS, if you have other options. If you are a laptop and tablet family household a NAS is perhaps the better of several alternatives.

4 Likes

Vanilla PC drives probably aren’t rated for 24x7 operation but as mark_m says all drives will fail eventually. So you have a backup strategy and you carry it out. Right?

SSD may solve the 24x7 operation issue as compared with spinning rust.

Another factor in favour of SSDs is noise. If you use the NAS in a context where silence is a requirement (e.g. HTPC) then an SSD is better.

On the other hand SSD is more expensive, so it depends on your budget. For a photo collection, I would guess that SSD is viable. For a movie collection, maybe not.

If your main use of the NAS is backup then you also need to consider what type of disasters it will protect you from and what types it won’t.

If it were me, I would want my growing photo collection on a centralised always-on box like a NAS, so that I can access the photos from any device, any time, anywhere. A very long time ago I used the “pick the machine that is on most of the time” approach - but it just wasn’t convenient.

In other words, I am suggesting that the NAS is the storage and the backup goes elsewhere.

As far as speed goes, the previous comment covers it. The network will be the speed limit so don’t worry about striping or fast drives. (Fast drives will however benefit you if you do significant local processing e.g. indexing or e.g. backup to removable drive.)

My 2c.

3 Likes

3 backups
2 different formats
1 offsite

4 Likes

In reply to several comments about the need for backup and redundancy you are preaching to the converted. I have seen small businesses that relied heavily on a central system who were too lazy to do backups: they learned the hard way.

Even the average family needs backup. Once our family history and memories were kept in a shoe box, now they are in phones, SD cards and HDDs.

It isn’t that I don’t have anyback but it is slow and inconvenient. I am looking for a modest outlay that will give a big improvement in speed and convenience. I want to automate the process fully or partially. Next stop backup utilities!

3 Likes

A couple of tuned Raspberry Pi3’s each with a 4+TB traveller drive strapped to them in two premises across a wireless link - regularly rsync’d with ‘no deletes’ and a diff report to facilitate periodical review and clean of any needed deletions. Even with one running Kodi, never any performance issues with my level of use, which is not heavy. I always hold a couple of spare travellers, just in case, and these are rotated through the backup pi somewhat sporadically giving pit snaps for which I write the date on the drive - the exposure is in the event of drive failure before resync, of course, the timing of which is reduced by the only slightly out of date replacement drive - happened once so far …

ymmv :slight_smile:

2 Likes

I just have a Mac Mini with multiple external desktop drives. I have a clone of the internal drive, and I also use Time Machine. That takes care of the basics. For movies and TV, I hve a separate drive but I dont bother with backup because the contents are ripped from DVDs I already own, so if needed I can rip them again (though that would take a long long time… maybe I should do regular clones of that drive too.) I had considered a Synology 4 bay but in the end decided that what I have is sufficient.

2 Likes

If a single drive or mirrored then normally the bottleneck is the HDDs. Burst speeds of these can be reasonably high but sustained reads and writes tend to be much lower. RAID without any parity, fault tolerance, or redundancy are not safe options unless you are backing up the data very regularly.

RAID 0 is performance based with the data striped over at least 2 disks but has none of the parity, redundancy or fault tolerance. Striped means the data is written over how ever many disks are in the RAID array eg 2 so that the data is shared between them speeding up writes and reads and increases the capacity by at least doubling the disk space (double if 2, triple if 3, and so on). RAID 0 uses two controllers in a 2 disk system so instead of writing to a single disk, 2 writes are made in the same time and similar for reads…3 disks 3 controllers so 3 reads and writes in the same time as a single read or write in a single disk. There is a little time lost because the controller has to share the data between the disks but it is only a little overhead.

RAID 1 is mirroring and doesn’t tend to have much speed improvement if any as it is writing the same data to the disks in the RAID array, but it is fault tolerant (has redundancy) if one disk fails the other/s continue to function until they fail. Replacing a faulty disk means you do have to re mirror the data but it is much more secure than having no parity, fault tolerance or redundancy. In reality not better than a single disk for speed. Purely mirrored drives only give you the same capacity as a single drive.

There are several RAIDs beyond those 2 with perhaps RAID 5 being the best choice for both speed and parity if not using Nested RAIDs. It does require at least 3 disks and one can fail and you will still not lose any data when it is replaced. With RAID 5 the parity data is shared among all the drives so there is no dedicated parity drive. In this RAID array the data is striped across all 3 or more drives so the capacity is much larger than a single drive but because of the parity data not as much as if purely striped. .

Nested RADs come in a variety of choices but mostly seen are RAID 01 (0+1 or mirrors of stripes) (both mirroring and stripes) with 3 disks needed but generally 4 at least actual (2 mirrors and 2 stripes) or RAID 10 (1+0 or stripes of mirrors) with at least 4 drives required. In my opinion RAID 10 is the better choice.

For a more detailed read of RAID 01 & 10 see https://www.thegeekstuff.com/2011/10/raid10-vs-raid01/

For networks at least a network controller of 1 Gbps is needed to ensure good data flow. You can buy NAS boxes with 2 X Gbe ports or better such as 10 Gbe ports but these also require similar support on your router or server. But 1 Gbe should more than adequately serve a normal household and is not normally a bottleneck.

NAS drives are specially made to support the continuous 24/7 power on. Many have vibration control and other modifications that suit NAS usage that standard household drives are not designed for. So if using a NAS with rotational drives choose NAS drives. SSD memory cells have a limited amount of writes they can do before they will fail, so while faster read and write access is part of the allure be aware that the NAS box hardware and software should support the use of SSDs particularly the TRIM command. Most modern NAS boxes do support SSDs but be sure to check.

4 Likes

Thanks for the rundown on backup and storage generally. On this there seems to be good consensus. However on the slowest component in the chain there is not. You said:

While others have said the slowest link in the chain is the network. I am not trying to provoke a blue here but to collect evidence. So people what is your evidence for your position? Do you have numbers? How representative and reliable are they?

I see “book” maximum speed figures quoted for various components but just how (or even if) these are comparable is not so simple. Theoretical limits are not much use to me, I cannot stop being an empiricist.

I would be particularly interested if there were any real-world tests where one parameter was varied while all other aspects were kept the same while various backups, such as big video files or many small files, were measured. For example the same network and NAS box with various HDDs. Anybody?

2 Likes

You mean supporting link aggregation (802.3ad or similar)? Yes that would be nice but unfortunately my file server only has a single GbE port. :frowning:

With only a single GbE port I can’t possibly get full speed out of an SSD drive. The real benefit of an SSD drive is in doing the local backup to external drive, where I can get > 500MB/s. This is a genuine issue because one of the best ways of ensuring that backups actually get done is ensuring that they are fast and easy. :slight_smile:

3 Likes

The actual measured sustained speeds on my disks are so far above even the theoretical speed of GbE that I don’t feel the need to prove it. :slight_smile:

2 Likes

Network at 1 Gbps ie 125 Megabytes per second is not sustainable by rotational drives which peak at about 170 MB/s but at sustained are much more likely to be 50 MB/s or commonly much less. So a striped RAID of 3 rotational drives would possibly make the speed limit sustained of a 1 Gbe network. SSDs are more likely to saturate a 1 Gbe network (why gamers love them). SSD 500 MB/s in real world tests are for sequential reads and writes not the usual with most read and writes being a mix of block sizes, I must distinguish here that these are SSDs that are using SATA ports and much faster speeds are obtainable when using NVMe SSD drives. Why do I remark about rotational drives, it is because when it comes to a home system to buy SSDs of any great capacity to use in a NAS you are talking large amounts of money. To buy a 4TB NAS HDD you would pay about $200 each (or less) for a decent brand eg WD Red or Seagate IronWolf but a 4TB SSD would set you back at least $700 and perhaps $1,500 each.

It is what comes after your server or router that makes a difference eg if using a 100 Mbps network then it is likely this will be less than the data being sent, but wireless speeds if using bands n, a, or ac will not be a usually limiting factor.

1 Like

This isn’t a competition, I am not issuing challenges, I am just after data. So what kind of testing have you done and what were the results?

Where do these figures come from?

1 Like

Places like these:

https://www.harddrivebenchmark.net/

https://www.userbenchmark.com/UserRun/16457811

https://hdd.userbenchmark.com/
(use the dbl arrow at the end of the table to add the 4K & sustained reads etc)

There are lots of magazines and similar that benchmark new drives when they appear in the market. HDDs have caches to store data both for reads and writes, these do make a difference when it is sequential data or initial bursts of data but if reads and writes are random (which a lot are) caches do not help too much. SSDs also have similar caches and combined with the controllers & memory chips used they do much better but again are much more expensive per megabyte than rotational drives, this pricing continues to fall and smaller SSD drives are much more cheaply obtained nowadays.

1 Like

Yes. Since I measured this doing backup, it is sequential read on the source disk and sequential write on the destination disk - and a large transfer size.

I expect that the transfer size makes the biggest difference, when comparing with a more usual workload. I think with SSDs whether it’s sequential or random makes less difference, by contrast with spinning rust, that has rotational and seek latency.

If you are throwing around media files (a common use case for home NAS) then the transfer sizes are going to be well above the worst case for disk performance.

In fact, since I am backing up to external drive, there is USB in the mix on the destination side.

Indeedy. Hence the suggestion above that for a movie collection SSD might not be financially viable for the average customer.

2 Likes

In addition that is a theoretical figure. Here is an article that discusses just some of the overheads that apply to the real world on the network side. https://www.cablefree.net/wireless-technology/maximum-throughput-gigabit-ethernet/

2 Likes

Yes, however there are so many different configurations and variables. To avoid this complication, components tend to be assessed and tested in isolation. This is to ensure that the variations in results are only due to the device being tested, and not the test environment.

Things to consider:
Budget limitations?

  • Current network design and specification.
  • Storage device reliability, speed.
  • Processing speed and critically operating system of the PC or NAS etc the storage is installed on.
  • Type of data and file formats and sizes most often accessed or saved.
  • The maximum real data transfer speed each client is capable of.

Every home is different in some way!

There is a good summary of the storage options and some considerations using RAID per @grahroll prior. That’s if there is a need for more secure storage. It does push the limits of a faster cabled network where there is the need. I’ve seen this as the norm in offices I’ve worked in with advanced 3D CAD animations.

To keep it simple for a home network, about as good as it might get, some data.

While this relates to a NAS, a NAS is still a computer. Typically one with a lowly dual core CPU and 1-2Ghz speed rating to keep the power low. The big difference when compared to a PC is the NAS operating system is optimised to do one main task. Saving and sharing data! The test results are all over a 1Gbps or faster ethernet.

The file transfer results include all the impacts of NAS overhead running a cross 1GB ethernet connection. The test assumes there is no other network traffic at the same. Objectively there is a little spare to go faster on the test network which is as good as it might get at home?

On the following you will also find a variety of speed results for NAS including the Synology DS119 single drive model, with a throughput of 72.2 MBps.

https://www.smallnetbuilder.com/tools/charts/nas/view

A suggestion:
With a fast network at home it is possible to test how fast it is using Windows task manager or similar but moving a large file and by moving a directory with lots of small files. It will give an indication of the gap in performance and potential benefit when using a home NAS and mechanical drives.

With a slow home network, typical Wifi or 100Mbps Ethernet, the same experiment will demonstrate just how much the network might be slowing you down.

P.S.
Caveat.
If the device used for file storage is a slow computer and the device at the other end is a cheap tablet, the actual file transfer speed will slow to which ever is the slowest of the two at processing the data. This may be many times less than the highest theoretical speed of the communications link. One fast device + one slow device = one slower data transfer speed.

Maximum Wifi speeds are a bit like the maximum speed on the dash of older cars. The car could only ever reach the speed on the far right of the dash if the car was dropped out if the back of a plane at altitude, approaching terminal velocity! :flushed:

3 Likes

You could buy four of these and get close to 4TB for $556. Five, and you’re well over 4TB with $5 left over of your 700 :wink: .

Those prices, they are a changin’.

2 Likes

Plus

  • the level of expertise of the user
  • the amount of time and effort the user wants to put in to the initial setup and/or ongoing maintenance
2 Likes