Post B5FlHiChWJl6dliytE by quad@akko.quad.moe
 (DIR) More posts by quad@akko.quad.moe
 (DIR) Post #B5FkYpNnyw1zsIW3Jw by quad@akko.quad.moe
       2 likes, 0 repeats
       
       The terramaster arrived. Now I just need to hold out until drive prices fall again
       
 (DIR) Post #B5FkxLVqAzcGTNxvsG by quad@akko.quad.moe
       1 likes, 0 repeats
       
       Luckily the HDD price seems to be falling faster than I expected. Last year an 8TB Ironwolf was around 2250-2500kr, but by Februrary this year they were going for 3500kr. But now they're already back down to ~3000kr.At this rate, hopefully I can get some hard drives for regular-ish prices again around summer.
       
 (DIR) Post #B5FlAxOEf11TiCtmCG by quad@akko.quad.moe
       0 likes, 0 repeats
       
       I still haven't figured out how I'll set it up.I could do nine drives in a RAID-Z2, but that would require 9 drives with the same capacity, and I just don't think that's feasible.So I suspect I'll end up doing RAID-Z1 in groups of three.
       
 (DIR) Post #B5FlHiChWJl6dliytE by quad@akko.quad.moe
       0 likes, 0 repeats
       
       Three RAID-Z1s is probably better anyways, I think putting all drives in a RAID-Z2 isn't the best idea as far as performance goes.I'm just not sure I like the idea of giving up 1/3 of my capacity to redundancy.
       
 (DIR) Post #B5FltnfNeuZ4aQIisi by quad@akko.quad.moe
       0 likes, 0 repeats
       
       Unfortunately the price drops seem to mostly apply to lower-end capacities, like 8TB and 12TB. Larger capacities like 16TB are still holding their prices high.So I guess there's no point thinking about it, I'll just have to wait at least a month or two before I start considering buying drives anyways, and unfortunately I'm not psychic enough to know what prices will be. If I was I would've bought drives in November instead.
       
 (DIR) Post #B5Fmbbvwdtc8wIwIm8 by ffaw@decayable.ink
       0 likes, 0 repeats
       
       buy some drives from a recycle center. Most of the time datacenters and big companies get rid of a shit ton of perfectly working drives and equipment and the recycle centers if they let you buy from them let you take shit home for nothing.
       
 (DIR) Post #B5FmbcQ4priWRl2Mk4 by quad@akko.quad.moe
       2 likes, 0 repeats
       
       @ffaw mostly not an option in norway because a B2C used market essentially doesn't exist here.i used to do this as a kid, but regulations now require them to wipe any device before giving it away. unfortunately none of them give a shit so it means basically every recycling center forbids people from bringing out electronics. even if you ask to take only a computer while removing the storage in front of them they'll still refuse. i can get random wood and metal from them, but no electronics and DEFINITELY no storage.direct B2C used markets essentially do not exist because in norway a business can't just flip shit on a used market, they have to do stuff properly with a VAT and honor a warranty. norway is extremely strict (even more so than the EU) about it being entirely impossible for a company to sell something without a proper 3+ year warranty.this means the only remaining option is refurbishment shops, where the prices are pretty ass. for example i can get a base model T480 with an i5 and 8GB of RAM for 400 EUR. And I haven't seen them sell hard drives whatsoever.
       
 (DIR) Post #B5FnFU13V0WNy0LHwO by ffaw@decayable.ink
       0 likes, 0 repeats
       
       Damn that really fucking sucks. Maybe you could find some retards who are lax on the rules but from what you are saying it sounds rather difficult.  I guess you could also go for ssds as I have seen people who drill holes in the drives thinking that they erased the data that way but didn't realize the boards took up 20% of the chasis so it didn't actually ruin the drive but at the end of the day ssds are bad to use due to the read and write limitations.
       
 (DIR) Post #B5FnFUEAiF0scgTlNw by quad@akko.quad.moe
       0 likes, 0 repeats
       
       @ffaw having a tech hobby in norway unfortunately just kinda sucks at times.the lack of a B2C used market is honestly one of the worst. sometimes you'll find it sidelined (meaning an employee just took shit home for free and then flips it for a profit, technically making it C2C), but overall it means the used market for tech sucks here. i'm always jealous when i hear of people in the EU/US who get a whole pallet of network switches for next to nothing.Being outside the EU also means I can't access the used markets on sites like ebay, I'll get hit with import fees, so it's not worth it 95% of the time.
       
 (DIR) Post #B5Fnkb8t3Nxp2tqHSa by ffaw@decayable.ink
       0 likes, 0 repeats
       
       Even outside the eu ebay sucks. I have a us friend that told me that what some americans will do is buy up everything of a specific electronic after a youtuber makes a video about the product then they'll mark up the price by 200+%
       
 (DIR) Post #B5FnkbJWPqTFZsom2K by quad@akko.quad.moe
       0 likes, 0 repeats
       
       @ffaw yeah. i work in IT so I can bring stuff home sometimes, but unfortunately it's rarely stuff that I actually want.I do have two Intel NUC Rugged Kits which we used for running infoscreens though, which I've now converted into LibreELEC boxes. I love those.I check the used market on the most popular national site (finn.no) pretty regularly, but it's extremely rare that I find much worthwhile.
       
 (DIR) Post #B5FoO2RtSiwZ1sXU2K by antsu@p.antsu.net
       0 likes, 0 repeats
       
       @quad Is that just a case or does it also include the thinking sand?
       
 (DIR) Post #B5FoO2dalEIjcA0pGq by quad@akko.quad.moe
       0 likes, 0 repeats
       
       @antsu Neither, it's a DAS.It contains a USB hub, some pile of USB-to-SATA chips and a PSU: https://www.terra-master.com/products/d9-320Unfortunately only 10 Gbit USB. Which is fine for my HDD use, but might be a bit limiting if I ever use the daisy chain port.
       
 (DIR) Post #B5FoaYLKhqmVQFRCXw by quad@akko.quad.moe
       0 likes, 0 repeats
       
       @antsu I was originally planning to build a new NAS at the start of this year, but considering both RAM and HDD prices went to hell, I decided to instead buy one of these to plug into my existing server. Then I can buy HDDs when the price falls, and sometime later I can replace the server with some small compute box since the storage is external now.Terramaster also sells a NAS version of the same thing if one would prefer to have magic sand inside of it: https://www.terra-master.com/products/t9-450But the one I bought is just for drive connectivity over USB.
       
 (DIR) Post #B5FqRLSVRMJelSWvJY by antsu@p.antsu.net
       0 likes, 0 repeats
       
       @quad Makes perfect sense. What are you connecting it to?
       
 (DIR) Post #B5FqRLfGfuWZP2V7Cq by quad@akko.quad.moe
       0 likes, 0 repeats
       
       @antsu I have an HPE Microserver Gen10 Plus.Great box, but I've simply outgrown it.It originally had a Pentium G5420 and 8GB of RAM.I've upgraded it to an i3-9100f and 32GB of RAM, which is about as high as it goes. (There's some Xeons it supports, but they cost a ton and basically only let me squeeze another 100-200 MHz out of it)Currently it has 4x4TB drives in its bays, in the PCI-E slot I've installed an adapter so it holds 2x1TB SATA M.2 SSDs and its boot drive is a 1TB USB SSD zip-tied to the frame inside and connected to the internal USB-A port.
       
 (DIR) Post #B5FqmFU4qhdKJdETh2 by quad@akko.quad.moe
       0 likes, 0 repeats
       
       @antsu Only thing I could really do was upgrade the drives to bigger ones, but it won't help with my performance issues, because having all four drives in a single raid-z1 vdev just doesn't perform great. Not to mention the fact that a 9th gen i3 isn't exactly top tier stuff either. But now I have 9 bays to go around and I'll keep the microserver for compute. (actually 13 bays, but the 4 in the server won't stick around once I replace it)So now I'm prepped to buy some more HDDs once the prices fall, and replace the compute once that starts getting cheaper. Who knows when that'll happen though
       
 (DIR) Post #B5Fsn3Vj3EjhNXW3ZA by antsu@p.antsu.net
       0 likes, 0 repeats
       
       @quad I almost went for an older model of HPE Microserver (the cube-shaped ones) when I was looking for a new server a couple of years ago, but it was out of my budget at the time. As much as a hate HP for their crappy end-user hardware, their enterprise stuff is pretty cool.>USB SSD zip-tied to the framelol can you even say you have a homelab if there isn't something zip-tied, duck-taped or super-glued together?
       
 (DIR) Post #B5Fsn3iUHmwc17UFSS by quad@akko.quad.moe
       0 likes, 0 repeats
       
       @antsu The Microservers were great. My first server was a Microserver Gen8 (I still have it, it's just backup storage now). They were good and cheap. However they fucked up the pricing on them after the Gen8, so they're no longer worth it imo.I got my Microserver Gen10 Plus through a warranty. I had a Qnap TS-963X, but the 10 Gbit port on it died after 11 months, since it wasn't manufactured anymore they couldn't give me a new one, so I was told I could pick whatever from their store with an equivalent price or lower. So I said fuck it and got a proper Microserver Gen10 Plus instead.It's been great, but I don't really recommend that anyone buy the newer ones unless you find them used at a good price. The Gen7 and Gen8 cubes were absolutely amazing, proper server with an IPMI and everything for real cheap. But the later ones cost too much and are now priced as actual servers.>lol can you even say you have a homelab if there isn't something zip-tied, duck-taped or super-glued together?the funniest part is that the internal port is only USB 2.0. linux is linux of course and doesn't give a shit, everything important to the OS gets cached in RAM anyways. but despite my / being an SSD, whenever i do anything "demanding" like ""copying a file to it" or god forbid i update my system packages, the thing caps out at like 30 MB/s and iowait goes to hell
       
 (DIR) Post #B5GAgvSRsR07qBD4xU by fedops@fosstodon.org
       0 likes, 0 repeats
       
       @quad I started with 4 disks in raidz1 as I'll likely catch a failing disk early on. With raidz expansion I can always add more when the time comes. I'm more interested in capacity than performance though with a measly gigabit the CPU barely budges (on a Ryzen 7, not the Terramaster).
       
 (DIR) Post #B5GAgvnMcfjOt2zmYi by quad@akko.quad.moe
       0 likes, 0 repeats
       
       @fedops Well I currently have the same setup. Four drives in a Z1.My main problem is mostly async IO clogs. if we pretend that hard links don't exist a moment for the sake of demonstration. Copying a 100GB folder containing a few seasons of anime to another location, the copy technically reports as "finished" pretty quickly and the command exits.But the next time zfs flushes async (is it every 5 seconds by default?) my HDDs will go into overdrive and be 100% active, causing iowait to go from like 10ms to 150ms for everything else on the system and my load average will exceed my number of CPU cores.The system then basically chugs like that for ages while ZFS figures out the actual work behind the scenes and everything else trying to access the pool suffers.With mdraid+<some filesystem> or bcachefs the copy command itself would actually work until the copy had really happened, but while the copy was happening the impact on the rest of the system would be a lot lower.
       
 (DIR) Post #B5GBGaBuQKVaJ9Tads by fedops@fosstodon.org
       0 likes, 0 repeats
       
       @quad interesting. I'm assuming an actual copy, not a zfs send/recv? That should be way less overhead.In general though I'm more concerned with copy onto/off the network and there the network is the limiting factor, by far.
       
 (DIR) Post #B5GBGaNFk9aAsKmeK8 by quad@akko.quad.moe
       0 likes, 0 repeats
       
       @fedops I did this exact thing like two days ago on rust-pool. I did a "cp -r" of a folder containing about 70-80GB of video because I was going to mess around with it a bit using ffmpeg.Sure, the "cp -r" command finished and exited in like 5 minutes. But behind the scenes ZFS was still doing its thing, meanwhile anything that wasn't already in ARC's MFU was borderline untouchable. For example if Sonarr wanted to import a few TV episodes, it does a scan first to list the filenames and then match them to a show, a process that typically takes maybe 10 seconds for a full season. Sonarr would now spend 5+ MINUTES instead trying to do the same task.My system has 4 CPU cores and 15 minute average basically held that constantly, 1 min went as high as 11 (might've been higher, SNMP only checks every 5 minutes).Basically took a whole hour for stuff to return to a usable state.
       
 (DIR) Post #B5GBrxwp3lkuhcFGwi by quad@akko.quad.moe
       0 likes, 0 repeats
       
       @fedops I'm no ZFS expert, but I suspect it's because ZFS does redundancy at the vdev level, but does most performance related decisions on the pool level. So when it's copying to and from a pool with a single vdev, it gets treated as a single disk, and the way it manages IO gets pretty bad. I also encounter similar performance chokes when I'm doing something like dumping a container to a tarball for backup.Even then a traditional filesystem seems to perform better when copying to/from a single drive at times, probably due to caching more in memory.This isn't just a thing in this particular server, I see this with friends when I help set up ZFS too, it seems ZFS really wants a minimum of two vdevs if you don't want your pool to regress to near single-drive performance during certain workloads. My personal experience with four drive ZFS pools is that if you want any performance whatsoever, your only option is making two mirrors, meaning sacrificing 50% of your capacity, otherwise mdraid /w raid5 will tend to outperform it by a pretty significant margin if you're both reading and writing to the pool.I'm sure some ZFS wizard can tune it to be usable, but I sure can't.
       
 (DIR) Post #B5GCBcXoLZ3c72v0hU by quad@akko.quad.moe
       0 likes, 0 repeats
       
       @fedops Also my home network is 2.5 Gbit now. So I'll be pushing my drives harder.I haven't been able to add a 2.5 Gbit NIC to my current server though, since there's no free PCI-E slot. But my server will be 2.5 Gbit (or possibly 10 Gbit) once prices fall so I can upgrade the compute part of my server setup.I'm sure my current drive would handle a sequential read or write at 2.5 Gbit with no issues as well though. So I'm not too worried about that part, it's usually tasks where I ask ZFS to read/write something to/from the same pool which exceeds the size of ARC.
       
 (DIR) Post #B5GCXocK8oGJvVc2ls by quad@akko.quad.moe
       0 likes, 0 repeats
       
       @fedops I forgot a screenshot of my pools. Referring to "rust-pool" makes little sense without it.