|
D. Ebdrup posted:By the sounds of it it's as simple as this oneliner. well, I meant without having to use zfs and multiple devices, etc.
|
# ? Apr 17, 2020 21:24 |
|
|
# ? May 16, 2024 23:57 |
H110Hawk posted:Just use urandom instead. I also think they've improved random materially in the last decade to make exhaustion less likely. If you're going to write a TB of random data, you're going to need something better than what's in Linux. Also, so far as I know, Linux doesn't have good concurrent random generation on multiple threads. FreeBSD does. ChiralCondensate posted:well, I meant without having to use zfs and multiple devices, etc. Probably there are other ways of doing it, but none have been documented as of what I've seen?
|
|
# ? Apr 17, 2020 21:45 |
|
D. Ebdrup posted:But I'm on FreeBSD, there's no point in using a symlink. I meant in Linux re: urandom. I didn't realize that Linux has a global lock on (u)random so it can only read by one process at a time. Reading wikipedia around it it looks like it's to guarantee that no two processes get the same random data, and generally applications needing more than a certain amount are likely mis-designed. (As our SMR beater script is obviously mis-designed, it's an abuse script.) How does FBSD handle the two processes potentially getting the same random data? Or does it assume that is too outside of a chance to bother with? Really /dev/zero should be sufficient for these tests, no? SMR doesn't do anything with the actual data like compress it or similar? The randomness of the data may actually be slowing down the script here by causing the kernel to think harder regardless of which kernel it is.
|
# ? Apr 17, 2020 21:57 |
|
xarph posted:My zpool replace operation completed successfully sometime overnight. Pool is fine. I saw that article go up. The drives do perform at the specified 180 MB/s write speed. I have a theory that the drives use the whole disk as CMR cache, which would mean that if you write more than half the drive's free capacity in one go you would find out what the SMR performance hit is. A real issue for the tech review sites to look into (rather than an LTT video).
|
# ? Apr 17, 2020 22:11 |
|
On Linux if you really really really want lots of random bytes with no contention with other processes the way to do it is to use /dev/urandom to generate a seed value and feed that into your own prng. For a 1-liner you can use openssl to run aes on /dev/zero with the /dev/urandom bytes as the key
|
# ? Apr 17, 2020 22:37 |
H110Hawk posted:I meant in Linux re: urandom. The commit cited in the commit I linked posted:I think the primary difference is that the specific sequence of AES keys will differ if READ_RANDOM_UIO is accessed concurrently (as the 2nd thread to take the mutex will no longer receive a key derived from rekeying the first thread). However, I believe the goals of rekeying AES are maintained: trivially, we continue to rekey every 1MB for the statistical property; and each consumer gets a forward-secret, independent AES key for their PRF. And you still don't get camdd, so no queue depths to utilize all the threads your system has which means you're gonna need to make the oneliner more complex by doing foreach - except oh wait! You're not on csh so that won't work, as won't the backtick syntax I used for getting the number of CPUs, so you'll have to deal with undefined variables that bash loves so much when invoking environment variables. Still means you're going to be overloading Linux' CSPRNG while simultaneously performing offloads of Galois fields in finite field theory to the applicable hardware in the CPU, as well as doing hardware offload for fletcher checksums. What I'm saying is, the CPU is gonna be quite toasty. Relying on /dev/zero and writing to /tank/tub/null.bin assumes that what I remember about ZFS object ranges doesn't apply to strings of zeroes. Again, ZFS is pretty drat smart, so I believe it might even be capable of specifying a range of zeroes - in effect deduplicating it for free - instead of writing all of them, irrespective of the inline compression used. My brain is too tired to go source spelunking now, though. BlankSystemDaemon fucked around with this message at 22:52 on Apr 17, 2020 |
|
# ? Apr 17, 2020 22:46 |
|
I’ve got 4, 4TB Reds, they are the EFAX version so likely SMR, sitting in boxes next to my server. I was going to set them up in RaidZ as an upgrade for my home server. I should just return them to Amazon, right? I’m still within the return window.
|
# ? Apr 18, 2020 14:39 |
|
LordOfThePants posted:I’ve got 4, 4TB Reds, they are the EFAX version so likely SMR, sitting in boxes next to my server. I was going to set them up in RaidZ as an upgrade for my home server. Yes. Select the not as described option.
|
# ? Apr 18, 2020 14:44 |
|
LordOfThePants posted:I’ve got 4, 4TB Reds, they are the EFAX version so likely SMR, sitting in boxes next to my server. I was going to set them up in RaidZ as an upgrade for my home server. Well, what would you replace them with? Finding a solid 4TB non-SMR option right now seems challenging, unless there are 4TB whites in Easystores that are SMR (non verified AFAIK).
|
# ? Apr 18, 2020 15:35 |
|
DrDork posted:Well, what would you replace them with? Finding a solid 4TB non-SMR option right now seems challenging, unless there are 4TB whites in Easystores that are SMR (non verified AFAIK). IIRC easystores below 8tb are not SATA drives, they're straight USB.
|
# ? Apr 18, 2020 17:37 |
|
Buff Hardback posted:IIRC easystores below 8tb are not SATA drives, they're straight USB. Ah poo poo, I forgot about that. Welp. Sounds like a good impetus to bump up to 8TB drives, especially since the 8TB Easystores are usually cheaper than retail 6TB Reds, and only like $15 more than retail 4TB Reds. Sucks for corporate users and those poor souls out there who don't have a source of shuckable externals.
|
# ? Apr 18, 2020 18:04 |
|
Agree on the 8TBs.
|
# ? Apr 18, 2020 19:00 |
|
Did some double checking, looks like Easystores <=6TB are blues (so probably also SMR).
|
# ? Apr 18, 2020 19:08 |
|
DrDork posted:Ah poo poo, I forgot about that. Welp. Sounds like a good impetus to bump up to 8TB drives, especially since the 8TB Easystores are usually cheaper than retail 6TB Reds, and only like $15 more than retail 4TB Reds. I’ll have to get a whole new server since my ThinkServer only supports 6gb drives. Not a huge deal, aside from the cost. It’ll definitely make migrating easier since the ThinkServer only supports 4 drives. Sure glad I procrastinated on setting up that array.
|
# ? Apr 18, 2020 20:16 |
|
Does it really have a limit or is that just because 6TB was the max available when the server came out? The only controller limitation I know of is at 2TB, and that'd be for really, really old first-gen SATA/SAS controllers.
|
# ? Apr 18, 2020 20:23 |
|
Seagate Ironwolf would probably be my suggestion for that size of drive, otherwise 8TB+? Shucking WD externals all day long.
|
# ? Apr 18, 2020 20:36 |
|
IOwnCalculus posted:Does it really have a limit or is that just because 6TB was the max available when the server came out? The only controller limitation I know of is at 2TB, and that'd be for really, really old first-gen SATA/SAS controllers. It’s a Lenovo TS140, so it’s quite old. You may be right about that being the maximum available drive at the time of release. I looked earlier this year and there’s just not a ton of people trying that because the hardware is so old at this point. I’ve got to get 8gb drives anyway so I guess I’ll give it a shot. I like that form factor for the server so it’ll be nice if I can keep using it.
|
# ? Apr 18, 2020 21:36 |
|
Anyone familiar with FreeNAS here? I built a 2200G system, currently on a 256GB NVME and using an Easyshare 8TB for storage and running Plex Server on it. Two questions: Am I that bad off running the FreeNAS plug in instead of doing the command line install for Plex Server? The next question is I also have a few 2.5" SSDs around and I think my mom could use my NVME for her laptop. I'm pretty sure Plex is running off the HDD, which means it could be slower. Is there a way I can throw two 2.5" in, run FreeNAS off the the 40GB and run Plex off a 256GB SSD, and obviously still use the HDD for storage?
|
# ? Apr 19, 2020 01:32 |
|
LordOfThePants posted:It’s a Lenovo TS140, so it’s quite old. You may be right about that being the maximum available drive at the time of release. I looked earlier this year and there’s just not a ton of people trying that because the hardware is so old at this point. While 6TB drives were the largest Lenovo would ship them with, there's pretty much no reason they wouldn't support arbitrarily large drives--they just weren't reasonable options when Lenovo bothered to validate drive options. There are plenty of people on storage forums discussing their TS140 builds full of 8TB drives, for example. VulgarandStupid posted:Anyone familiar with FreeNAS here? Lot of us use it, yeah. To answer your questions: (1) The primary downside to the Plex plug-in is that it is consistently behind (sometimes by months) the current release. If this doesn't bother you, then no, the plug-in works fine and makes everything super easy. Frankly Plex hasn't added much that really matters lately, and if you really get antsy you can always drop into the jail and manually update it from there, I believe. I never bothered because it Just Worked well enough that I didn't care to tinker. (2) If you're itching for space you can keep Plex where it is in the jail and then do a simlink for the Metadata folder (that's the one that eats all the space) and have it actually live on the HDD. That said, having Plex run off the HDD probably isn't going to make all that much of a difference performance-wise if you've got 16GB or more RAM for your system and only are using it for one stream at a time. But, yeah, FreeNAS will let you happily install it on one SSD, have jails live on another SSD, and bulk media storage on a HDD. It don't give a poo poo. You'd just have to make mount points in the Plex jail for the HDD media folder, which is pretty trivial and can be done in the FreeNAS GUI with a couple of clicks.
|
# ? Apr 19, 2020 03:03 |
|
Looks like I got lucky with my 4TB Reds being EFRX
|
# ? Apr 19, 2020 03:18 |
|
LordOfThePants posted:It’s a Lenovo TS140, so it’s quite old. You may be right about that being the maximum available drive at the time of release. I looked earlier this year and there’s just not a ton of people trying that because the hardware is so old at this point. I've got a ts430, which is the previous generation. It's got 8x 8TB white labels on the SAS card, perfectly happy. Ditto with my ts440 (which is the same gen as your 140).
|
# ? Apr 19, 2020 04:35 |
|
for the two source of random, 'random' will lock and 'urandom' will never lock. random will lock whenever it runs out of 'entropy' quote:Without the assistance of external hardware RNG's, Linux uses naturally occurring chaotic events on the local system to generate entropy for the kernel pool. These events, such as disk IO events, network packet arrival times, keyboard presses, and mouse movements, are the primary sources of entropy on many systems. this is why applications to create certs or password managers will sometimes will ask you to move your mouse or hit random keys. urandom will also suck down this entropy but when it runs out, it will switch to psuedo-random but will be 'enough' in the cases above. it will never fail to return random values ever. i agree that using disk fill will be better to test gzip compression than all 0's but gzip compression only works well with files that use ASCII , pictures, and anything not just pure binaries like apllications. if you are testing random gzip compression... its not ever going to good...
|
# ? Apr 19, 2020 20:30 |
EVIL Gibson posted:for the two source of random, 'random' will lock and 'urandom' will never lock. All of the BSDs had support for at least hifn(4), which is from 2000, ie as soon as the U.S munitions export laws were relaxed - and before then blocking wasn't a thing because 4.2BSD didn't do it, and neither does HP-UX. Modern CSPRNGs (for example Fortuna, from 2003) are designed to never run out of entropy as long as they're well-seeded by at least one source of entropy, be it RDRAND from Intel or its equivalent from AMD, ARM, POWER, or RISC-V CPUs, as well as the hardware devices mentioned above. That's why FreeBSD lets you supply harvest sources as you see fit at runtime and by defualt uses a mix of software interrupts, hardware interrupts, net_ng (netgraph nodes, if used), net_ether (ie all the ethernet-level noise that's present on any network with more than one device, available thanks to BPF), as well as the small timing differences for when you move your mouse or between key-strokes on the keyboard, plus device attach events, as well as the information cached in /entropy which is the last thing to get written to during shutdown, in addition to those hardware PRNGs in the CPU I mentioned above. The idea that it does block is because Linux for a long time didn't have rngd from rng-tools until 2004 and Linux mainline kernel didn't support the hardware to seed it until 2006 in order to seed information from hardware devices into /dev/random because Linus thought he was gods gift to programming and he still hasn't fully conceded that he has no loving idea what he's talking about and just need to trust domain experts. You might also question why /dev/random needs to be seeded from userspace, but that's a mystery even I can't answer. BlankSystemDaemon fucked around with this message at 21:29 on Apr 19, 2020 |
|
# ? Apr 19, 2020 21:14 |
|
D. Ebdrup posted:Linus thought he was gods gift to programming and he still hasn't fully conceded that he has no loving idea what he's talking about and just need to trust domain experts. Why I never. Shocked I am. Shocked!
|
# ? Apr 19, 2020 22:50 |
|
I'm interested in building a NAS but I have no idea what hardware to go with. I don't know whether to build a modern AMD system (R3 3200G, Athlon 3000G, R5 1600), a modern Intel system (Pentium G5400) or an older Intel system (Ivy Bridge Xeon E3s on LGA 1155.) My main uses for this NAS would be Plex, running apps I don't want to keep on my laptop 24/7 (e.g. Sonarr/Radarr), and serving as a network backup hub. My Plex needs seem pretty light for now as I'm not seeing major CPU usage when using Plex internally and I haven't bothered to get my Plex server working for external users. I'd be running unRAID for this system. My number one priority would be noise. For reference I think I'd get a Synology DS418play if I were to go the prebuilt NAS route.
|
# ? Apr 20, 2020 04:46 |
|
I'd do the Xeon, but that's as much because if you get a server board you should also have IPMI, and that's a huge quality of life improvement if you can't stash a keyboard / monitor with it for those times poo poo goes sideways.
|
# ? Apr 20, 2020 05:00 |
|
Godzilla07 posted:I'm interested in building a NAS but I have no idea what hardware to go with. I don't know whether to build a modern AMD system (R3 3200G, Athlon 3000G, R5 1600), a modern Intel system (Pentium G5400) or an older Intel system (Ivy Bridge Xeon E3s on LGA 1155.) The ASRock Rack X470D4U (or X570D4i for ITX) is really slick and has IPMI so you can safely pick a Zen2 CPU if you go that route. Get a 3600 or one of the new 3300X quad cores - video encoding is one of those tasks that Zen2 is wildly better at than the Zen1/Zen+ chips you listed, it’s around 60% faster per core than Zen1 at video encoding. There are diy NAS chassis like the U-NAS or DS380 although they definitely tend a bit towards the chintzy. My U-NAS has had the lights burn out on 3 of the 8 drive trays after like two years of operation. Or you can just get a a Fractal Design R6/R7 and stack in a bunch of drives on the tray mounts. Paul MaudDib fucked around with this message at 09:06 on Apr 20, 2020 |
# ? Apr 20, 2020 09:03 |
|
IOwnCalculus posted:I'd do the Xeon, but that's as much because if you get a server board you should also have IPMI, and that's a huge quality of life improvement if you can't stash a keyboard / monitor with it for those times poo poo goes sideways. Seconded, but mostly for price. You can get an older Xeon (v3 or v4) for a song these days, and server boards with IPMI are readily available. That said, if you don't like the idea of going used, the ASRock Rack Paul mentioned is a pretty nifty piece of kit that I'd totally have gone for myself if I hadn't stumbled upon a hilariously good eBay deal. Unless you're thinking about transcoding multiple 4k streams at once, you probably don't really need to worry about performance much. I don't know how much Plex experience you have, but if you're just playing 4k stuff on a 4k TV, you're probably not even transcoding in the first place.
|
# ? Apr 20, 2020 15:13 |
|
EVIL Gibson posted:urandom will also suck down this entropy but when it runs out, it will switch to psuedo-random Both random and urandom are always psuedo-random regardless of platform. This is fundamental to this slap fight between the various unixes.
|
# ? Apr 20, 2020 15:19 |
|
If you're throwing money at the problem, not rolling your own and not having IPMI is very dumb, IMO. ECC support doesn't hurt, either.
|
# ? Apr 20, 2020 15:27 |
|
Crunchy Black posted:If you're throwing money at the problem, not rolling your own and not having IPMI is very dumb, IMO. ECC support doesn't hurt, either. Ipmi is very much a cost benefit thing. Knowing what I know about those $100 cards* (not the Dell idrac $500 cards) I would choose to hook up a monitor and keyboard the one time every few years I needed to do something that the booted os couldn't do. Even if it's "free" with the chassis I don't think I would set it up. If you buy a dell/HP with the $500 level card go hog wild those things are tanks. * based on my sample size professionally (20k+ of those things) those cards barely work from the factory and can't be relied on long term. Especially on the used markets they have a near 100% failure rate by 5 years in.
|
# ? Apr 20, 2020 16:00 |
|
The Supermicro boards I use have it built in and I've never seen a reliability problem with it.
|
# ? Apr 20, 2020 16:15 |
|
IOwnCalculus posted:The Supermicro boards I use have it built in and I've never seen a reliability problem with it. Maybe there is some fundamental difference to one that is fully integrated vs a small daughter board. Asus, Quanta, and Intel BMC cards are pure poo poo. Also from a security standpoint, which I realize is inside your perimeter, you should treat them like any other Internet Of poo poo device and assume they are riddled with security issues - never expose them to the internet.
|
# ? Apr 20, 2020 16:27 |
|
IOwnCalculus posted:The Supermicro boards I use have it built in and I've never seen a reliability problem with it. Same. I have zero experience with add-in IMPI cards, but I've never had a single problem with a motherboard that had it built-in. Which makes sense, considering what IMPI is designed for: if you build it in, it already has access to everything it needs to and can Just Work. If it's an add-in card it has to figure out wtf a given board might have decided to do with implementing various functions.
|
# ? Apr 20, 2020 16:36 |
|
yeah, what in the gently caress lol What kind of off the shelf mobo these days has an optional IPMI that isn't iDRAC or iLO, etc. ?
|
# ? Apr 20, 2020 18:32 |
|
DrDork posted:Same. I have zero experience with add-in IMPI cards, but I've never had a single problem with a motherboard that had it built-in. Which makes sense, considering what IMPI is designed for: if you build it in, it already has access to everything it needs to and can Just Work. If it's an add-in card it has to figure out wtf a given board might have decided to do with implementing various functions. The addons are dedicated parts manufactured to spec by the motherboard manufacturer. It's in theory literally the same chip. For example it can still be a tagged vlan on the onboard ethernet ports. For example: https://www.intel.com/content/www/us/en/server-management/intel-remote-management-module.html / https://ark.intel.com/content/www/us/en/ark/products/91525/remote-management-module-4-lite-2-axxrmm4lite2.html these are plugged into something like this, from the HCL in Ark: https://ark.intel.com/content/www/us/en/ark/products/192606/intel-server-board-s2600bpbr.html Maybe this is something Supermicro got fundamentally correct in their sea of crap server products. Crunchy Black posted:yeah, what in the gently caress lol Anything that isn't Dell or HP. Those are also $500/shot and built to way different standards. Dell also offers a $100 option but I haven't tested it out. (For all I know it's literally the same physical card and a license upgrade.) In the used market those iDRAC cards are great if you want to deal with ipmi, they even support html5 video consoles like the Intel cheapy modules do when they work. H110Hawk fucked around with this message at 18:40 on Apr 20, 2020 |
# ? Apr 20, 2020 18:38 |
|
H110Hawk posted:Maybe this is something Supermicro got fundamentally correct in their sea of crap server products. Presumably this. The Supermicro IPMI works well, works every time, and is included on-board for a lot of their servers. I find it kinda wild that Dell sells an iDRAC controller for twice the price of a lot of motherboards. Then again, I've installed my fair share of >$50,000 1U Dell servers wondering how the hell they get away with charging what they do for the hardware involved and questionable software support.
|
# ? Apr 20, 2020 19:41 |
|
I would love some sort of universal pcie IPMI card.
|
# ? Apr 20, 2020 20:33 |
|
Moey posted:I would love some sort of universal pcie IPMI card. I once sat down and started working out what it would take to use a Raspberry pi to control some relays to be able to remotely power on/off the box, and use it as a serial terminal. I stopped when I realized the BOM was already close to the cost increase of just buying a proper SM board. I think just about all Supermicro boards past the X34xx / X55xx generations have IPMI standard. I mean it'd be cool as hell - there's no reason a PCIe card couldn't present a super-low-end GPU over the PCIe bus, a jumper to a USB header to provide keyboard/mouse/storage, and jumpers to the power/reset switches. You'd need to provide it with its own power separate from the main power supply. But the quantity of people who would buy that, probably would keep the price for such a device well above the cost increase of buying actual server parts.
|
# ? Apr 20, 2020 20:56 |
|
|
# ? May 16, 2024 23:57 |
|
We use a simple network capable relay board to hit the power/reset pins on motherboards (just have to connect them together, ie grounding the signal input) and it works good for that. Having a vga and other ports relay would be useful too sometimes, but mostly the power cycling is all we need for doing power/reset testing.
|
# ? Apr 20, 2020 20:59 |