What's new

TS-364 and Coral accelerator, vs TrueNAS & HP microserver

  • SNBForums Code of Conduct

    SNBForums is a community for everyone, no matter what their level of experience.

    Please be tolerant and patient of others, especially newcomers. We are all here to share and learn!

    The rules are simple: Be patient, be nice, be helpful or be gone!

Duncan DiSauddelli

New Around Here
Hello

Short version:Has anyone got experience of using a Google coral accelerator with a QNAP (esp. the TS-364) please?

Background: my two-bay Netgear ReadyNAS is a dozen years old; working well as a mirrored pair but I fancy upgrading from it to give me RAID5 with SSDs, and also more CPU performance so I can use a new NAS box as a bit of a server too. The new three bay QNAP TS-364 is interesting, as is a home-grown solution using TrueNAS or similar running on a HP Proliant gen8 microserver. this is for a home GbE network with three or so computers connecting to the NAS, LAN only.

in either case I hope to use frigate.video with a Google coral accelerator; in fact, the QNAP is designed for use with one (and I should be able to get TrueNAS to work with it too).

In the case of the TS-364, I presume that if I have (say) eight RTSP streams presented on ethernet, I should be able to use the QNAP to do image classification of the stream content and record short video or jpeg snapshots of recognised images. Does this sound feasible?

The benchmark scores for the new QNAP and for the gen8 Microserver Xeon version are comparable so the available server "power" should be adequate in each case (plus I might run MPD or Plex or similar).
 
Last edited:
Interesting... with QTS, it depends on how 'open' their implementation is - frigate on a regular distro, I think that would be fine, and a lot less work if one needs to go out of bounds of the vendor's implementation.

With the right ML model, coral is/was amazingly fast compared to other solutions of the same vintage - last time I looked at things there was about 3 years ago, so things might be different these days - in the last year or so, seems like everyone is doing some kind of NPU/TPU, etc..

Have you been able to source Coral's - either USB or M2?

(they're hard to find, many resellers I've seen are out of stock and/or backordered)
 
I built my "server" for multiple functions from routing / switching / firewall / wifi / dvr / plex / etc. NAS being one major function. Basically I wanted to roll a bunch of different devices into a single platform and be able to modularly upgrade things as new tech comes out.

All it takes is a PC + Linux and some googling to make it work. I get 400MB/s+ out of it in a Raid 10 format using MDADM in the Linux OS and the ports on the MOBO. I'm using 8TB WD Reds,

I've used QNAP before in the past and it's alright for data but, ask it to do much more and they tend to choke. I have some QNAP parts though for pieces of the server puzzle. I was using their AC2600 card for the AP function before upgrading to AX (external AP) and I'm currently using their 4-port 5GE NIC for WAN/LAN functions.
 
These are interesting - and useful - observations; thanks! I haven't seen any stock of Corals anywhere thus it is likely to become one of those projects which I hold in abeyance for ages (or never complete which is, of course, typical).

I definitely want a robust NAS but also something with additional "power" and I think the Qnap approach might not suit my goals (even though the CPU benchmarks for the TS-364 CPU are only 10% behind the Proliant Xeon's score). The flexibility of going open source and using a PC with linux is definitely attractive .. as long as I don't bugger things up by fiddling too much! Luckily I always have a separate backing store to which I rsync every week, so all would not be lost.
 
The PC option definitely makes for a ton of options as to what purposes you can use it for. If you decide to add / remove functions it's possible to do modularly / as needed. If the Proliant takes a dump and needs to be swapped out it's easy to gather the pieces and put it back together with newer pieces. a Xeon is going to be more equipped for things like transcoding if you're thinking about streaming.

Being open source though makes a huge difference in being able to apply packages w/o fiddling with a "store" and limited options or having to hack something to get it to work. I upgraded my server from an 8700K CPU to 12700K for basically the price of the CPU when rebuilding from the ground up and selling the 8700K setup minus the drives and higher spec parts.

Not being locked into a 3-bay or x-bays for that matter is nice too. The latest case I went with (Meshify 2) holds something like 15 x 3.5" drives. I spaced them out w/ an empty slot between drives to make sure I got decent airflow from the 3 x 140mm fans on the front end. With the open air design it's virtually silent even when it's ramped up transcoding files. There's tons of space to work with inside the case for cabling / mounts, there's a single LED bar on the front to indicate power.... it's basically a stealth box that blends into the background even being front and center.

Setting up the Proliant shouldn't be that big of an issue. I went Linux but, basically all you have to do is share the drive like you would in windows and leave it alone. The benefit of Linux though is higher uptime unless you like to keep the kernel updated weekly. I just took a simple Ubuntu release to start off with and then modified it for the additional features I wanted to incorporate beyond the NAS. I do 99.9% of the maintenance from an SSH session since it's not being used as a daily driver. If there's a kernel issue on upgrades then it helps to have a screen available to revert the kernel to the last known good.
 
I definitely want a robust NAS but also something with additional "power" and I think the Qnap approach might not suit my goals (even though the CPU benchmarks for the TS-364 CPU are only 10% behind the Proliant Xeon's score). The flexibility of going open source and using a PC with linux is definitely attractive .. as long as I don't bugger things up by fiddling too much!

One of the things about going with a pre-built NAS from QNAP, Synology, Asustor, and others is that everything in the HW/SW stack has been developed and built with a well-defined set of requirements.

They are about as plug and play as they can be - buy it, unbox it, plug in the drives (if needed), follow the wizards, and one is up and running in a couple of hours. Once it's set up, leave it alone. Watch for security updates and apply them appropriately, and have a backup solution (or better yet, two solutions).

QNAP, along with others, has had security issues in the past - they're working to get ahead of it, and some of the causes were dumb mistakes (hard coded passwords for example in one of their apps that had admin level access). They are not the only ones.

Pre-builts also have support, unlike a DIY solution...
 
HP Proliant gen8 microserver

I wouldn't recommend the G8... they're pretty long in tooth, and on the tail end of OEM support.

G10's are current, and there's some decent deals, depending on your budget - I've seen xeon based units with 16GB ram and 4TB of spinning rust for less than a $1000US - doesn't come with a Windows license, but there, one can get Win 2019 Server Essentials for a fairly decent price if one shops around.
 
I do recall addressing this in my post above...
It just needs to be emphasized. There's a lot of security gaps in off the shelf options. It's nice to setup in a couple of minutes and start using these new shiny gadgets but, they have dormant risks.

I've had a couple is NAS devices prior to going diy and while functional they're very underpowered for anything besides LAN storage. A diy SFF PC and adding drives is cheaper than prepackaged units. It's not everyone's cup of tea though.

Going modular though saves cash in the long run. For instance more room for drives vs buying a new chassis. Swapping the NIC for something speedier for $100 vs a new chassis. If you wanted a bump in drive speed and want to do SAS it's an option. There's perks to building your own even with a gen 8.
 
It just needs to be emphasized. There's a lot of security gaps in off the shelf options. It's nice to setup in a couple of minutes and start using these new shiny gadgets but, they have dormant risks.

I've had a couple is NAS devices prior to going diy and while functional they're very underpowered for anything besides LAN storage. A diy SFF PC and adding drives is cheaper than prepackaged units. It's not everyone's cup of tea though.

For most people - that's all they want/need - OP's case with frigate.video and the coral unit - QNAP has their app station, and if it is not there, there's always docker.

I like to find a happy medium when making recommendations - not everyone is super technical and wants to get under the hood, etc with a DYI solution.

I think we're pretty much aligned in our comments here.
 
I contacted Qnap sales regarding the TS-364, including a couple of specific questions regarding Coral use cases. They haven't replied so I decided to go down the secondhand Proliant route . It might turn out to be a bit beyond me, but I hope can get it sorted out a step at a time. I'll practice on a virtualbox VM of TrueNAS, experiment with adding/configuring Docker images, then do it for real on the Proliant, and then finally transfer my data to the TrueNAS/Proliant once it is working.

Thanks for the advice and comments and I am sure I can come back for help!
 
While it's pretty there can be drawbacks of using ZFS on the drives.


I went with the KISS method and just used Ubuntu + MDADM / Ext4 with Raid 10. I've been running it for a few years now and it's rock solid. I have a 5th drive in the array as a backup if one fails it will automatically take over. If you want to expand it's an easy operation to add a new disk to the mix with a couple of commands.

If you want a pretty GUI there are options to add packages to do that. If you want to just monitor things there are apps for that as well. Since the device will sit somewhere out of sight most of the time it should just work. There's some automated stuff that you can configure when setting it up or after that monitor the health / rebuild if a problem is detected. You can tell it to e-mail you updates / problems.

It's a good place to start as well so you have an idea of what's actually going on vs relying on a pre-built piece of software that might bite you in the arse later when something goes buggy. You can still RDP into the box if you want a GUI experience to manage it or you can SSH into it and just run commands. The only time I need to go hands on is when I update the kernel and it doesn't agree with the HW I'm using. This kernel issue for me is using a 5GE card from QNAP that sometimes the dev's don't put the code into the newer kernel image and need to rollback to the prior version. Chances are though the PL box will be using popular HW that won't have issues with updates.

Going with the latest manual kernel for me is something that wrings out the most potential from the HW since I rebuilt it with ADL from the ground up last year. The kernel updates though take care of a lot of the bugs / security issues quicker than waiting for the updates pushed automatically. Most of the system releases stick with a particular major release with incremental updates as they come for that version.

You'll be happier with the DIY performance over being boxed into a specific vendor like QNAP / Synology / etc. If you decide to dive deeper into making tweaks you'll have the ability to use the box for other needs easily by adding the configurations needed.

I use mine for a few different things - router / switch / firewall / wifi / DVR / NAS / etc. That was my goal when looking at all of the devices I had sitting around taking up space / power outlets. Rolling the routing function into the device though I was able to take back control of data vs using a Netgear or some other black box off the shelf. I don't worry about a buggy router OS release causing potential issues at this point. I don't have to toss a box and pay more than I need to in order to upgrade to the next technology i.e. AP for 6E is $360 vs Router @ $500+. I can modularly swap the NIC for the speeds I need w/o being boxed into either 1GE or 10GE or 1 x 2.5GE port. I went 4 x 5GE to cover WAN / LAN needs currently. If I boost the speed of the NAS I can always add a 10GE card to the mix for $100.
 
While it's pretty there can be drawbacks of using ZFS on the drives.

The 'hidden' cost of using ZFS for your home NAS
I went with the KISS method and just used Ubuntu + MDADM / Ext4 with Raid 10. I've been running it for a few years now and it's rock solid. I have a 5th drive in the array as a backup if one fails it will automatically take over. If you want to expand it's an easy operation to add a new disk to the mix with a couple of commands.

The linked article is from 2016...

ZFS can bring things to the party, just like btrfs, or LVM, MDADM, etc...

They're all good, truth be told. The main thing with implementing a RAID solution is to do some planning up front.

Adding drives to an existing array usually is going to have some impact - I'd rather not add to an array, but rebuild it from scratch and restore the data from a very recent backup (yes, RAID's need to be backed up just like other storage solutions)

MDADM/RAID10/EXT4 - about as simple as it can get, and has very good read/write performance overall - that would be my choice as well. This approach is very mature, well understood, there's enough data redundancy to recover data from a broken array - and one thing to consider - it's been my experience several times over - when a member of a raidset (disk) dies, usually one or two are planning to attend the funeral - in other words, time to plan on replacing all the drives if they are from the same time/date set.
 
I'll practice on a virtualbox VM of TrueNAS, experiment with adding/configuring Docker images, then do it for real on the Proliant, and then finally transfer my data to the TrueNAS/Proliant once it is working.

Thanks for the advice and comments and I am sure I can come back for help!

Quick tip - use the array only for storage, keep the OS and applications on another drive (SSD or spinning rust) - that will make recovery easier when things eventually die (all disks will die at some point), and also makes OS and application updates easier - one can unmount the array, do the upgrade or OS reinstall, and then remount the array once things are golden.

Also, keep the swap file off the RAID, which is another reason to have the OS/Apps running on their own drive.

Good luck, sounds like a fun project
 
when a member of a raidset (disk) dies, usually one or two are planning to attend the funeral - in other words, time to plan on replacing all the drives if they are from the same time/date set.
I knew this going into it and ordered from different batches.

Quick tip - use the array only for storage, keep the OS and applications on another drive (SSD or spinning rust) - that will make recovery easier when things eventually die (all disks will die at some point), and also makes OS and application updates easier - one can unmount the array, do the upgrade or OS reinstall, and then remount the array once things are golden.

Also, keep the swap file off the RAID, which is another reason to have the OS/Apps running on their own drive.
1000% agree with this. It keeps life simple when things aren't. Even w/o the native OS installed you can mount / scan the array from the livecd if you're troubleshooting something else.
 

Latest threads

Sign Up For SNBForums Daily Digest

Get an update of what's new every day delivered to your mailbox. Sign up here!
Top