Host Your Stuff - Part V
Introduction to Home Server Storage Hardware
So you've got your first piece of your home server purring away, but that single drive is feeling a bit cramped. So what now? Enter stage left, Network Attached Storage or a NAS.
More local storage versus a NAS
Before building a NAS, one may ask if it makes sense to add another system or to just add more storage to the existing home server. This comes down to a couple main factors centering around scalability and flexibility. While you can add more storage to the current server and share that out to multiple clients as a sort of NAS-lite system, this job may not be particularly well suited to a smaller home server long term. Migrating large amounts of data is a tricky proposition and can be quite time consuming, even with high speed networking. This can often lead to the decision that building a NAS shortly off the bat and taking the approach of "buy once, cry once".
So what is a NAS anyway?
A NAS is a specific category of hardware designed to provide a large pool of storage to multiple clients across a network. In contrast to adding drives to each individual system, a NAS lets you provide that storage to your home server(s) and other machines on your network. This can be used for bulk storage, backups, even running VMs or containers from (though there's some caveats) which gives you a lot more flexibility in where your home server projects run.
The big things that make for a good NAS will be the ability to fit multiple hard drives and/or solid state drives, wired (ideally fast) networking and a remote management interface. A good starting point will be a four or more drive bays, a 2.5Gb or 10Gb Ethernet NIC and a quiet case will be good criteria to hit.
Features like redundant controllers and NICs, hot-swap drive bays and redundant power supplies are all necessities for enterprise use but aren't nearly as important for home use and will come with a substantial cost premium. If you have the money burning a hole in your wallet, go for it but there are better uses for it.
In direct opposition to my previous point, if there are multiple people reliant on your NAS being up, redundancy starts becoming a bigger priority. None of us like getting called at 2AM for a Severity 1 bridge call at work, and having that call come from your spouse, kid or family member is no better through outright worse.
Buy vs Build
Buying from a vendor
A pre-built NAS like a unit from Synology or similar vendors can be a decent starting point and are consistent, well-supported and documented for both hardware and software. As long as they're used within the expected workloads with the parts they expect, it is unlikely to run into new and novel issues. This is a double-edged sword as coloring outside the lines of vendor approved configurations will have more edge cases than a Hawthorne Heights concert. If something simple and fire-and-forget is desired, Synology or QNAP will serve you well.
If you want to go bigger, iX Systems also builds NASes running their TrueNAS operating system and can get massive compared to the Synology and QNAP options. This comes with a price-tag to match and is more suitable for deployments where hardware and software issues would be show-stoppers and professional support is required. 45 Drives is in a similar class with their focus primarily on large, high capacity systems and a polished, easy to use interface for small-medium business use. Their HL-15 system may be worth some consideration, but comes at a steep price.
Reputable Vendors:
- Synology
- QNAP
- iX Systems
- 45 Drives
Build it yourself
None of the vendor solutions quite do it for you, there's hardware on hand or you just like to tinker? Building your own NAS may be the option for you!
In a near complete opposite to the packaged solution, a DIY NAS gives you a substantial amount of flexibility in hardware, storage options and software where the only limits to capacity and performance are your budget. However, this does require much more deliberate choices for hardware, software and configuration. The documentation from your various hardware vendors, OS and other services will be your best friend.
A small 4-6 bay NAS is likely better served by a pre-packaged solution but more high performance, high capacity or otherwise intensive storage systems will be more cost effective to DIY.
A mid-tower desktop case and ATX motherboard will be a good starting point. I recommend starting with something that can fit four to six drives. Leave two bays open for expansion later or to migrate to larger drives later and fill the rest with identical drives of your chosen capacity. The CPU requirements are typically fairly modest and a mid-spec desktop CPU will handle them just fine. The CPU recommendations from the first section still apply here. The key difference will be that there is no such thing as "too much RAM' for a NAS. The more the merrier here as the memory will be used for caching which can help speed up read operations. The less a system needs to reach out to disks, the better.
SATA, SAS, NVMe, oh my! Picking your storage medium
One of the key factors that will determine much of how the rest of a storage system works and what it requires will be the drives in use.
SATA Hard Drives
SATA (Serial ATA) hard drives are typically your standard consumer hard drives found at your local big box store. Seagate Ironwolfs, Western Digital Blue/Black/Red/Purple and whatever Toshiba calls their drive line up will be common sights here. Currently, this is something of a minefield due to the two main formats of hard drive recording, CMR and SMR. Without getting into the specifics, SMR hard drives should be discarded entirely. Thanks to vendors being immoral, money-grubbing scum, whether a drive is CMR or SMR can be quite a toss-up. Generally the higher end NAS drives will be CMR, but the vendor should not be trusted. Check spec sheets, independent reviews and cross reference multiple sources. The simpler answer is to tell these vendors (Western Digital especially) to suck rocks and buy used SAS or enterprise SATA drives if you can.
If you must buy new, Western Digital Red Pro drives are often the only ones worth considering. The rest are slower, lower reliability and may or may not be SMR drives. The Blue, Green, Black and Red drives all have their own issues. Primarily the Blue drives are heavily “value engineered” (read: made as cheaply as possible) and are prone to early failures, dead controllers and inconsistent performance. The WD Greens do raise the tripping-hazard of a bar slightly as they’re at least consistently slow. The rest of the caveats remain. WD Black drives are the first tier that aren’t entirely a waste of materials as they typically fail at a less abysmal rate, but are still prone to controller failures and do not have the vibration resistance to use in larger arrays. Additionally some are SMR drives and this is not clearly delineated so should be treated as suspect.
WD Gold drives, their psudo-enterprise line, are consistently decent performers and can be quite reliable and are generally not SMR drives. These are often one the better options for high grade SATA drives from Western Digital. That said, these come with a substantial price premium and do not compete well against their used SAS counterparts.
Seagate has slightly fewer issues with consumer SATA drives and are more consistent with marking drives as CMR or SMR, however this still requires careful reading of the spec sheets. The Ironwolf Pro drives are a good bet for reliability, consistent speed and avoiding many of the issues that plague Western Digital. The Exos line are proper datacenter drives and are rated for use in large storage arrays, have higher reliability than the consumer or pro-sumer lines and are a solid bet for SATA drives in storage arrays.
A particular drive type of note are the MACH.2 dual-actuator drives, these drives effectively function as two independent hard drives in the same chassis, in SATA drives, these are transparently passed through to the host OS as a single drive and are internally striped together into a single faster, higher capacity volume. The SAS variants need to be managed by administrator, but function similarly. These drives offer excellent speed and random read/write performance (for hard drives) at the cost of some capacity. It would be best to avoid these for some time as the price premium is substantial and long-term reliability is still unknown.
Toshiba generally manufactures drives for OEM use and receive far less attention than Western Digital and Seagate. Personally I know very little about these drives and information regarding their performance, reliability and use-cases is sparse from third parties. While I won’t dismiss them out of hand, more research and testing would be required before a recommendation for or against can be made.
SATA SSDs
SATA SSDs fall into much the same trap as SATA HDDs with lower performance, lower reliability and higher cost per terabyte than their used SAS counterparts. Most notably, there are more “value engineered” drives in this space and more pitfalls. The intricacies of SATA SSDs deserves a separate article but the key points will be summarized here.
In the same vein as the HDDs, the lower range consumer drives should be avoided due to poor sustained performance, low endurance and reliability. Drives like Samsung Pros can be used for lighter workloads, however they will quickly become a bottleneck on more intensive workloads and offer a poor value for price per terabyte.
A particular exception to note are datacenter SATA SSDs, these typically have reliability and performance more similar to their bigger SAS cousins and can be good choices for moderately intensive workloads. These drives do tend to come with a substantial price premium compared to new consumer drives or enterprise SAS drives. These should be picked only if the particular system demands 2.5” SATA drives and more reliable and performant storage is required.
SAS Hard Drives
Moving over to SAS (Serial attached SCSI) drives, the same vendors come up with Seagate's Exos lineup, Western Digital Gold, HGST and again, some Toshiba drives. Note a trend here. These drives are more commonly found in big, enterprise storage systems and have some features that make them better suited for that role. These drives look similar to SATA drives, but take a separate interface to support the SAS-specific features. These are more commonly found on the used market and are often substantially cheaper than new, still similarly reliable to new consumer drives and available in bulk (often at a discount). These drives often form the bulk of the storage used by self-hosters and homelabbers.
SATA/SAS Interoperability
Note: SATA drives can generally be plugged in to SAS ports, but the opposite is not true. You can mix SATA and SAS drives on the same backplane or controller, but this is typically inadvisable due to factors discussed later.
NVMe SSDs
NVMe or Non-Volatile Memory express drives are the current darling child of gamers, enterprise storage administrators and anyone who wants to fast. These are going to be SSDs and will be your fastest, but highest price per TB of storage option. These drives come in four main form factors, M.2, U.2, EDSFF and 3.5 HDDs. M.2 and U.2 drives are the most common with EDSFF and NVMe HDDs being fairly exotic and rarely seen outside cloud providers and some supercomputers.
M.2 NVMe Drives
M.2 drives are great options for low power, high burst performance workloads like consumer laptops and desktops or as boot drives for servers. However, these drives generally do not have the sustained performance or longevity for more demanding workloads. Additionally, these drives tend to be disproportionately expensive at higher capacity versus the enterprise options.
Long story short: These make good boot drives, can be decent options for VM/container storage but should be avoided for bulk storage. The primary issue with these drives is that they cannot sustain the high performance numbers that are advertised and will quickly crash back down to SATA SSD or SATA HDD levels of performance under the sort of sustained workloads that a moderately active home server will have.
U.2 NVMe Drives
U.2 NVMe drives are the step up from the consumer M.2 drives, sharing the same form factor as 2.5” SAS SSDs but talk to the host machine using NVMe as the name suggests. A primary benefit of these drives is the much higher transfer speeds compared to SAS SSDs, as well as higher endurance and more consistent performance than M.2 drives.
A critical factor against these drives are that they do consume enough power to warrant active cooling through fans. While all drives should be cooled, the lower volume of airflow through a case is generally sufficient for SATA or SAS HDDs. This is not true for U.2 drives and will often require the drive cages they’re in to be cooled more directly through a fan attached to the cage, a fan mounted near or on the drive (though a small, lower airflow 40mm fan is generally sufficient) or substantial airflow through the entire chassis. Despite the additional considerations and cooling requirements, these drives offer the best density and performance available without moving to more exotic solutions like EDSFF E1.L drives or vendor-specific, custom flash.
These drives make excellent choices for high performance storage and are a staple of current-generation enterprise hardware. For a smaller home server or homelab these drives are often overkill but due to their prevalence can often be found cheaper than SAS SSDs. These drives will be fairly expensive options regardless but can provide excellent storage density with options as high as 122TB per drive available. Currently 7.68TB drives are a good option for density and price per TB.
Currently, the top vendors for these drives are Kioxia, Samsung, Intel (now Solidigm) and Micron. All of these vendors build high quality through exceptional drives for datacenter use and will handle nearly any workload home use can throw at them. These do warrant more careful examination of the spec sheets (which unlike consumer drives, can be trusted to a fair degree) to check that the drives meet requirements for transfer speeds, endurance and interface.
SAS SSDs
These drives follow much of the same benefits and caveats of U.2 NVMe drives, though to a lesser degree for both the benefits and downsides. Unlike U.2 drives, SAS drives communicate over the same SAS bus as SAS HDDs allowing you to reuse the same HBAs and cabling for both drive types. These will be more reliable than consumer SATA drives and often as fast or faster. These drives do consume more power than their SATA counterparts, but less than U.2 drives making active cooling required but to a lessened degree.
The best vendors for SAS SSDs are much the same as U.2 drives with the inclusion of HGST and removal of Solidigm who did not yet exist in the heyday of SAS SSDs.
Picking Your Storage
SATA drives will be your most readily accessible option as they are plug-and-play with consumer and prosumer hardware. However, these drives are often going to be pretty poor choice for price for capacity. Used SAS drives can often be found in bulk at the same or higher capacities for cheaper. These will require some extra hardware and cabling discussed later in this article but often the cost of the added hardware is dwarfed by the dollar/TB difference.
SAS drives will typically be your best bet for higher densities, bulk purchases used and are often substantially cheaper per terabyte than new SATA drives. Generally, targeting $10USD/TB or lower will be the minimum for a "decent" deal, with $6-8/TB being the range for a "good" deal.
Capacity also plays a massive role in drive pricing. Both the top and bottom end of drive capacities are going to be your worst options for $/TB efficiency. The older, smaller drives only get so cheap before they're not worth selling and the newer, high capacity drives have yet to be phased out and still command like-new prices.
At time of writing (January 2025), 10-16TB SAS drives are currently the sweet-spot for storage pricing with 18-22TB drives offering a good balance of cost, density and total power consumption due to a smaller number of drives being required.
2.5" HDDs, SATA or SAS
2.5" HDDs are generally a relic of a bygone era before SSDs became popular. The consumer SATA drives are typically going to be fairly low quality, lower reliability and more expensive per TB of capacity versus the 3.5" variants and don't have a good reason to be purchased versus 3.5" drives or SATA and SAS SSDs.
Okay but what do I pick?
Simple answer: Whatever your hardware supports the easiest. If you're getting an off-the-shelf NAS, use the drives the vendor supports. If you're building your own and you just want to use the SATA ports you have, then get SATA drives. Grab some 12-14TB drives, at least four of them and you're off to the races.
SAS Drives, Host Bus Adapters and Disk Shelves
SAS drives largely function the same as SATA drives to the average user but have a few key differences that can make or break a storage setup. As mentioned above, SAS drives are not compatible with SATA connectors without an interposer, allow for multiple paths to the disk's controller and easier scale up for large storage arrays.
SAS Muti-path
While this is likely to be out of the scope of many people's first (or even fourth) NAS, SAS offers the ability to multipath or provide multiple physical links to a drive's controller. This can be used in more professional/enterprise setups to allow for redundant controllers, cabling and disk shelves.
Critically for a home server, multipathed drives will present themselves as one physical disk when properly configured, but adding one or more SATA drives to the array breaks multipathing and will instead present two physical drives for each path that can be taken to the disk. In smaller, single-pathed, single controller setups, this typically does not present an issue but will create problems in larger, more redundant setups.
SAS Host Bus Adapters (HBAs)
Many motherboards will have a built-in SATA controller (directly attached to the chipset on the motherboard) and allow for a small number of SATA drives to be attached. Now what about SAS drives? Since they require a different physical interface and different controller, this is where hardware RAID controllers and HBAs come in to play.
Hardware RAID will be skipped for now as it is largely a legacy technology and is inadvisable for most home server uses.
HBAs are "dumb" controllers that act much like the SATA controller on a motherboard. They provide ports to attach SAS cables to the card and drives, just using different cables and ports than a traditional SATA controller.
These HBAs are typically desirable over RAID controllers as it allows for more flexibility in the storage configuration, portability between systems and more choices for your filesystem on those drives.
Which SAS HBAs are any good?
Short answer: LSI 9300 series or newer. The older LSI 9200 series cards are rapidly being dropped from support for most operating systems. The 9300 series cards support up to SAS 12Gb/s speeds and have current driver support. The newer LSI 9400 and 9500 series cards also have varients that support NVMe drives, though they command a significant premium for this capability. Generally this will be of little concern to most building a NAS, though may be the topic of a future article.
Dissecting LSI model numbers:
- LSI - The Broadcom subsidiary who makes the controller
- SAS/HBA - This specifies the specific protocol (or lack of one) used by the HBA.
- This four digit code represents the general model (93xx, 95xx) and any trailing digits (9531 for example) are more specific model versions but are generally interchangeable. As with most hardware, check the vendor's spec sheet to be sure.
- 8e/16i - This denotes the number of SAS lanes the card has and whether they are internal (For drives inside the case) or external (For disk shelves, tape drives and other devices).
- Tri-Mode - When present, this denotes an HBA that can communicate over SATA, SAS or NVMe protocols making it effectively "omnivorous" regarding the drives it can use.
Disk Shelves
Disk Shelves or DASes (Direct Attach Storage) are external chassis designed to hold, power and connect a large number of drives to a server or storage system. These shelves are "dumb" as they are not able to manage the disks themselves and require a connection to an actual computer.
The allow for a single server to add more disks than it could physically hold and expand storage "vertically" as opposed to adding more and more discrete NASes. This is a strategy known as scale-up storage and can be used to create some truly massive storage arrays, though this can have throughput limits on the larger end of the scale.
These disk shelves are almost always pieces of datacenter gear with the noise and power draw to go with the category. This is likely best avoided for beginners and for small labs until truly massive amounts of storage are required. However, these systems are excellent options for going big where hundreds of terabytes through petabytes of storage on a single system are required.
How much storage do I want?
To quote myself and other tired, hangry regulars of the Homelab Discord: "How long is a piece of string?".
Data, like a gas will expand to fit the space available. For those of us that remember when a 120GB hard drive (adjust units for degree of joint pain upon reading) was impossibly large and we'd never think we would accumulate that much data, we're likely now bemoaning the measly 1TB of OneDrive space Microsoft thinks we wanted.
The main point being that oversizing your storage slightly can be beneficial as the amount of thins you store will likely exceed your expectations but building massive pools on older technology is a loosing battle against Moore's Law.
As mentioned above, even though smaller drives like 10TB SAS HDDs can be found for quite cheap on the used market, building a several hundred TB array of them will require substantially more supporting hardware like Host Bus Adapters, cables and disk shelves than a smaller number of higher capacity drives which can eliminate the cost savings.
Some general estimates of thumb:
- 3GB per hour of 1080p video
- 12GB per hour of 4k video
- 140MB per hour of CD-quality music
- 300MB per hour of audiophile-qualty music
- 10-30MB per book (PDF format)
Examples at scale: - Anna's Archive copy: ~1PB
- Season 1-8 of Game of Thrones: 210GB at 1080p, 843GB at 4k
My rules of thumb for storage volume
For my homelab, the rough numbers I go with for planning out storage works something like this:
- Eyeball how much data I need, for fixed-size projects like archival work this can be found outright or estimated easily enough, however for more nebulous workloads, I typically work off of educated guesswork and identifying what size of storage meets my performance requirements, budget and the hardware it needs to work with.
- Add 25-50% of that capacity as buffer for redundancy like RAID and expected underestimation of needs.
- Figure out when to get rid of data. Adm. Grace Hopper has an excellent talk discussing the value (or lack there of) for storing data that looses usefulness over time. Once you figure out how much data you need, figure out how much is actually important. Once data has outlived it's usefulness removing it can help reduce the "bloat" of storage.
This last step is deceptively critical. Once the archival data is pushed to tape, it has limited to no use on hot/warm storage and can be removed. VM, container and other backups can get increasingly sparse as time goes on as well. You may want to take hourly snapshots for systems you're actively working on or for quick roll-backs, but retaining those backups can drop down to days, weeks and even months between copies for older versions of the VM or container.
Likewise if data is stored in some other medium (like tape, DVDs, readily available online), the importance of backing it up is greatly reduced. Important personal documents, family photos and other similarly valuable bits of data may be very important to retain, but are often much lower volume of data than your average Plex library. Prioritizing what data is actually necessary and when the cost of maintaining copies is outweighed by the value of the data helps balance the capacity of the storage required.
What next?
So you've got a shiny, new NAS built. What now? Next we'll discuss the operating systems used for NASes, how to get it talking to your devices and some basic configuration from there.