|
Paul MaudDib posted:You mean apart from the obvious "subsequent backup runs will only move the changed data, potentially near-instantly"? You can also add compression, it transparently understands SSH/SFTP (vs having to loop through CIFS/Samba), and tons of other stuff. In other news, I don't dare hope for it, but will we get firmware-less (or at least observable/USEable) disk drives now that WD is moving from ARM to RISC-V?
|
#
?
Nov 29, 2017 22:26
|
|
|
|
| # ? Apr 26, 2024 18:38 |
|
|
D. Ebdrup posted:This is an incredibly small nitpick: ZFS will only ever transmit which individual blocks have changed; it can be quite a big difference if you're dealing with very large files. The question was rsync vs dircopy/xcopy type tools, if you are using zfs send you probably know about the merits of rsync.  While we're on the topic though - I used zfs send to move a backup to another ZFS pool. Let say we now have mainpool/myfs@snap1 and otherpool/myfs@snap1. I have made changes on mainpool/myfs, and I now have mainpool/myfs@snap2. How do I send the second snapshot over while preserving whatever context ZFS needs to know that snap1 and snap2 share blocks? Would that be zfs send -I mainpool/myfs@snap1 mainpool/myfs@snap2? And what would go in the rest of the recv command there? zfs recv otherpool?
|
|
#
?
Nov 29, 2017 23:21
|
|
|
D. Ebdrup posted:In other news, I don't dare hope for it, but will we get firmware-less (or at least observable/USEable) disk drives now that WD is moving from ARM to RISC-V? That press release is a giant buzzword salad written to impress dumb investors. Why do you think you would you see any difference? The instruction set shouldn’t be any more visible to you after this transition. Also, it’s not like ARM is some inscrutable custom architecture nobody can understand. Also, there’s nothing truly new here, it’s still just cpu cores running firmware. And actually I’m quite bemused by your apparent desire for a world in which a disk is no longer a black box. Disks are fixed-function embedded systems. High level abstraction at the interface is a good thing, it’s what lets us plug a SATA disk from any manufacturer into any SATA port and have a very high level of expectation that it will work. It would not be fun to have every disk running application code and I expect that even if WD tries to push the idea they will fail. What this is really about, once you read between the lines: WD is tired of paying royalties (no matter how small) to ARM Holdings for each disk controller chip they ship. They probably don’t need a super high performance core, particularly in the spinning disk business, and thus saw an opportunity to move to technically inferior RISCV cores that have a royalty cost of free-ninety five.
|
|
#
?
Nov 29, 2017 23:45
|
|
|
BobHoward posted:What this is really about, once you read between the lines: WD is tired of paying royalties (no matter how small) to ARM Holdings for each disk controller chip they ship. They probably don’t need a super high performance core, particularly in the spinning disk business, and thus saw an opportunity to move to technically inferior RISCV cores that have a royalty cost of free-ninety five.
|
|
#
?
Nov 30, 2017 00:44
|
|
|
Star War Sex Parrot posted:ding ding ding we have a winner! And there's nothing wrong with that. Just as ARM is eating Intel's pie for workloads that do not require massive computing power (even getting into servers now), so can RISC-V start eating into ARMs pie for things that don't require even that, and that are free to boot.
|
|
#
?
Nov 30, 2017 00:49
|
|
|
Volguus posted:And there's nothing wrong with that. Just as ARM is eating Intel's pie for workloads that do not require massive computing power (even getting into servers now), so can RISC-V start eating into ARMs pie for things that don't require even that, and that are free to boot. That may eventually bite them on power usage or silicon area in the long run. The voodoo erasure coding and signal massaging they have to do at ~200MB/sec requires some pretty impressive solutions, especially in the 1-2W thermal design envelope. Between that and the shiny new Microwave Assisted Magnetic Recording, WD is poised to have some interesting things show up to market in the next 2-3 years.
|
|
#
?
Nov 30, 2017 11:32
|
|
|
Methylethylaldehyde posted:That may eventually bite them on power usage or silicon area in the long run. The voodoo erasure coding and signal massaging they have to do at ~200MB/sec requires some pretty impressive solutions, especially in the 1-2W thermal design envelope. If I were them I'd be doing signal path compute in fixed function accelerator blocks or DSP cores rather than more general purpose CPUs.
|
|
#
?
Nov 30, 2017 12:27
|
|
|
BobHoward posted:If I were them I'd be doing signal path compute in fixed function accelerator blocks or DSP cores rather than more general purpose CPUs. Oh, I guarantee that every single function more mathematically complex than the queue scheduler is run in fixed function stupidly optimized and pipelined custom silicon, but there is a lot of SATA specific stuff that you'd run in firmware vs. custom fixed function blocks, and the overhead there is what's gonna make the difference. ARM's stupidly well done memory controller vs. whatever 3rd party RISC memory controller they buy/develop, etc. Gonna be interesting to see what errata and weird gently caress failure modes they get on a brand new controller architecture vs. the mature one they have now.
|
|
#
?
Nov 30, 2017 13:34
|
|
|
BobHoward posted:That press release is a giant buzzword salad written to impress dumb investors. I did say that I didn't dare believe it, didn't I? I know it's because WD doesn't want to pay ARM royalties, but at the very least vendor backing for RISC-V isn't gonna hurt it and may help it. The reason I don't like firmware is that I've encountered or heard about way too many situations in which the lack of observability actually made it impossible to debug why something went wrong. For example: Bryan Cantrill tells a story in one of his presentations about how a motherboard firmware (UEFI, I believe - but it could've been BIOS all the same) was actively supressing interrupts for the ECC DIMM telling the OS that it'd corrected an error, which meant that they didn't get any advance warning before the ECC memory could no longer cope and the DIMM started getting uncorrectable memory errors - which, obviously, is quite bad on something where data integrity and availability is very important. For an example related closer to WD and disks, look no further than a few posts back to the post in which I mention the Toshiba disks that "exponential decay of the writes per second before it magically resets", as Bryan put it. If firmware let us ask questions about what goes wrong when it inevitably does go wrong, I wouldn't be so against it - but what firmware does that? Why yes, I do talk about Bryan a lot, don't I.
|
|
#
?
Nov 30, 2017 13:35
|
|
|
Regardless of what architecture they're using for the controllers, if you seriously think that WD is going to let you fully inspect their system, you're fooling yourself. They'll bake firmware in for part of it somewhere, they'll refuse to release the block diagrams, whatever, and we'll be little better off than we are now in terms of understanding precisely how and why bits get from the SATA port onto the platter. They'll find a way.
|
|
#
?
Nov 30, 2017 13:41
|
|
|
DrDork posted:Regardless of what architecture they're using for the controllers, if you seriously think that WD is going to let you fully inspect their system, you're fooling yourself. Seriously, what incentive do they have for anything else?
|
|
#
?
Nov 30, 2017 15:04
|
|
|
DrDork posted:Regardless of what architecture they're using for the controllers, if you seriously think that WD is going to let you fully inspect their system, you're fooling yourself. They'll bake firmware in for part of it somewhere, they'll refuse to release the block diagrams, whatever, and we'll be little better off than we are now in terms of understanding precisely how and why bits get from the SATA port onto the platter. They'll find a way. Hilariously enough, releasing that kind of info can get the State Department to crawl up your rear end about it. The special sauce to tons of industrial or computing technologies are on the export control list, which means publishing them online, where Iran and North Korea can see them counts as directly exporting it to those countries, which can get you in a world of poo poo. It's why you probably won't ever see a fully open source hardware, firmware and OS stack published for any kind of high performance computing system.
|
|
#
?
Nov 30, 2017 15:06
|
|
|
IOwnCalculus posted:Seriously, what incentive do they have for anything else? Methylethylaldehyde's notes are non-trivial. There are also concerns about divulging information or techniques that competitors might be able to lift for their own purposes. Allowing for full inspection also invites people to identify vulnerabilities and weaknesses--which, while such identification can eventually result in better overall products, tend to produce a lot of bad PR in the meantime, and thus most companies are pretty damned gun-shy. Also tech companies generally exist in a cloud of "we ain't tellin' no one nothin'!", and that sort of institutionalism is hard to change. Basically there's virtually no up-side for them, but obvious potential down-sides.
|
|
#
?
Nov 30, 2017 15:30
|
|
|
I was agreeing with you, there's no benefit for WD or anyone else to offer more visibility. Certainly won't make a difference in sales volume.
|
|
#
?
Nov 30, 2017 15:39
|
|
|
Yeah, absolutely. An article headline of "WD drives' controller found to have fatal bug!" would be way more damaging to their sales than the occasional person posting an annoyed review on NewEgg about a drive that mysteriously died and needed to be RMA'ed.
|
|
#
?
Nov 30, 2017 15:54
|
|
|
Paul MaudDib posted:The question was rsync vs dircopy/xcopy type tools, if you are using zfs send you probably know about the merits of rsync. Sending the second snapshot the same way you did the first should only transmit the delta (as long as the receiving FS has the first, obs). https://github.com/jimsalterjrs/sanoid/ (syncoid, partway down the readme) might help make the process easier than dealing directly with zfs send - honestly wish I had found that/it had existed last time I dealt with snapshot replication across servers.
|
|
#
?
Nov 30, 2017 15:55
|
|
|
There's always going to be bugs on the manufacturer's end that you can't really do anything about. Even if you could inspect the firmware, there's still the possibility of hardware bugs.
|
|
#
?
Nov 30, 2017 16:07
|
|
|
Main Paineframe posted:There's always going to be bugs on the manufacturer's end that you can't really do anything about. Even if you could inspect the firmware, there's still the possibility of hardware bugs. Something like half the firmware updates are workarounds to hardware errata. Hell, look at AMD's AGESA updates, every time they update it, they give the hardware better memory compatibility and improve and address feature issues like the IOMMU bug they launched with. It's why old rear end or longer lived platforms tend to be less crap when they're actively maintained, you eventually quash all the low hanging bugs and beat some stability into it.
|
|
#
?
Nov 30, 2017 18:06
|
|
|
I'm going to finally retire my old NAS and do a new build. I want around 30-40TB of space, right now I'm using like 12TB but when I look at the size of 4K UHD stuff I want to futureproof as much as I can. Part of me just wants to go with Synology or something but I hate depending on package maintainers for stuff like Radarr and I need decent transcoding performance for at least 3-4 streams. Most of their stuff seems to be 2000 passmark rated Celeron/Atom crap unless I'm missing something. Any recommendations for a 5+ bay case that will sit in my basement? Something with a SATA/SAS backplane with a decent amount of space. I had a Silverstone DS380B awhile ago and drives were very hot so I ended up just using a desktop case I had laying around instead. I was looking at the uNAS stuff but they seem really annoying to cable and work inside of.
|
|
#
?
Nov 30, 2017 18:25
|
|
|
I use a Fractal Design Define R5. No hot-swap but the eight bays are on sleds. I have 8x6TB drives in mirrored vdevs in a FreeNAS VM for 24TB of storage. Alternatively you could use a Supermicro or Chenbro tower with hot-swap bays. If it's in the basement you shouldn't have to worry too much about noise (or heat?)
|
|
#
?
Nov 30, 2017 18:44
|
|
|
phongn posted:I use a Fractal Design Define R5. No hot-swap but the eight bays are on sleds. I have 8x6TB drives in mirrored vdevs in a FreeNAS VM for 24TB of storage. Hah that's funny, that's the exact case I'm using now. Maybe I'll just swap out the PSU, clean it out and keep it going. I'm weighing moving over to FreeNAS but the lack of expansion (I know its planned) sort of bothers me and at home I want my NAS to be as hands off as possible after setup.
|
|
#
?
Nov 30, 2017 18:55
|
|
|
2nding the R5. Put some nice static pressure fans in it and plop it in the basement and it should keep things nice and cool. No hot swap, no backplane but its probably cheaper so you can pump that money into storage/CPU for encoding.
|
|
#
?
Nov 30, 2017 18:59
|
|
|
Is there a way/setting to make a normal SATA port on a regular consumer mobo support hot swapping or do you have to have some kind of hardware in between? Can't believe I've never thought to look into this before but hey first time for everything right?
|
|
#
?
Nov 30, 2017 20:13
|
|
|
Matt Zerella posted:2nding the R5. Put some nice static pressure fans in it and plop it in the basement and it should keep things nice and cool. No hot swap, no backplane but its probably cheaper so you can pump that money into storage/CPU for encoding. 3rding. Just found out yesterday that you can put the sled elsewhere in the case, I wonder if you can get two sleds in:  I want to go up to 16 drives not sure if I'll need a new case or not. Hughlander fucked around with this message at 20:54 on Nov 30, 2017 |
|
#
?
Nov 30, 2017 20:52
|
|
|
.
sincx fucked around with this message at 06:33 on Mar 23, 2021 |
|
#
?
Nov 30, 2017 21:41
|
|
|
Munkeymon posted:Is there a way/setting to make a normal SATA port on a regular consumer mobo support hot swapping or do you have to have some kind of hardware in between? The Gunslinger posted:I'm weighing moving over to FreeNAS but the lack of expansion (I know its planned) sort of bothers me and at home I want my NAS to be as hands off as possible after setup.
|
|
#
?
Nov 30, 2017 22:01
|
|
|
phongn posted:I believe that most Intel SATA ports can be set to hot-swap (may need to be set in the BIOS). You also need a proper backplane for electrical safety. Something like that might be prudent, but I've always just hot-plugged SATA drives without issue or extra setup. Even that one time when I hotplugged a SATA power cable in to a IDE connector nothing got broken. (I bought a 750 GB SATA drive from a local computer store and never actually looked at the connector before I saw sparks).
|
|
#
?
Nov 30, 2017 22:35
|
|
|
YMMV, but when I replaced all the drives in my array, I hot swapped all of them, without sleds or anything. Just unplugging cables and then plugging them into the new drives.
|
|
#
?
Dec 1, 2017 00:24
|
|
|
For almost any ready-built NAS the ports are configured hot swap. For most build systems the motherboard bios/uefi controls hotswap on each port.
|
|
#
?
Dec 1, 2017 03:03
|
|
|
I literally just rebuild an entirely new NAS every few years and use zfs send, then I sell off old drives. PSUs get old, and dust accumulates and so forth. Some might use the old NAS for some disaster recovery or something else but I’m finding that I just don’t care enough to do that. I may just buy a 48 bay monstrosity and manage my array and expect to keep the CPU and motherboard for 7 years. My Sandy Bridge Xeon machine has been on nearly 24/7 for 7 years now minus swapping motherboards after a weird incident that corrupted my BIOS after I put an M1015 variant to try to upgrade to 4 TB drives. I have a Lian Li A04 for my NAS as well as a UNAS NSC-800 I grabbed for experimenting with a successor while I got a new motherboard. Turns out the motherboard in that case has shorted out while I had it unplugged or after I fired it up after months of sitting as a backup array. I dunno, wtf is happening with the hardware in my life.
|
|
#
?
Dec 1, 2017 03:15
|
|
|
necrobobsledder posted:I may just buy a 48 bay monstrosity and manage my array and expect to keep the CPU and motherboard for 7 years. I keep thinking about that too. But if I did that, I'd feel obligated to add another 8 drive vdev every year until it was full, which would get cost prohibitive, AND I'd be replacing old drives at the same time every 2-3 years of their lifespan.
|
|
#
?
Dec 1, 2017 04:16
|
|
|
Nah, I’d buy drives at inflection points of deals happening such as the recent Black Friday deals. I can probably whip up a script to guess these points based upon price data from deals sites. LightningDrops and such only notify based upon simple static price points, nothing based upon a function on time series data for forecasting like Holt-Winters or ARIMA. By what I’ve seen in the graphs I saw recently, I’d have bought them about twice this year and a couple times in the past 3 years (drive prices were super stagnant for years - lookzing for a sudden drop matters here), and that roughly matches how often I buy drives. Granted, capacity remaining and how long it’ll take to use more than 80% capacity may matter more.
|
|
#
?
Dec 1, 2017 05:02
|
|
|
I still use a ten year old https://www.newegg.com/Product/Product.aspx?Item=N82E16811119093 but I only bought one extra hard drive stack insert (comes with one, so I can get 8 in there safely), so unless I can find another on ebay or watever I have to gently caress with bay adapters to put more drives in. Other than the USB2 on the breakout box it's basically the perfect NAS case, though.
|
|
#
?
Dec 1, 2017 15:10
|
|
|
Is there a best solution for using a bunch of different sized drives? Where you plug everything in, say 4-10 drives, and the system figures out how to best split them up for redundancy. Basically want to use a bunch of extra hard drives I have on hand, running a Plex/file server on its own machine. I want the ability to add or replace drives as needed without losing the whole data pool. From my limited research Freenas with ZFS is not ideal for this type of storage.
|
|
#
?
Dec 1, 2017 20:00
|
|
|
You’re probably looking for something like Unraid it sounds like. It may be paid but the idea would be that you’re going to save substantially more with Unraid than by buying extra / matching disks and hardware to support another software RAID.
|
|
#
?
Dec 1, 2017 23:44
|
|
|
jeff8472 posted:Is there a best solution for using a bunch of different sized drives? Where you plug everything in, say 4-10 drives, and the system figures out how to best split them up for redundancy. Stablebit DrivePool is made for this exact situation. All of your mismatched drives get added to the pool and it takes care of figuring out what drive to put stuff on. It also offers folder level redundancy settings.
|
|
#
?
Dec 2, 2017 01:34
|
|
|
Krailor posted:Stablebit DrivePool is made for this exact situation. All of your mismatched drives get added to the pool and it takes care of figuring out what drive to put stuff on. It also offers folder level redundancy settings. Yep, Stablebit Drivepool was the real successor to Windows Home Server. The same devs even worked on the DrivePool. I have never had any real problems with my setup since WHS died and it expands and shrinks with extreme ease. Never mind being able to pull a drive out of the pool and read all the files in any windows computer for recovery. Downside is you get basically logical RAID 1 for a 'share' or triple redundancy.
|
|
#
?
Dec 2, 2017 02:01
|
|
|
Looks like Skylake Xeon D is happening next year. Maybe by then DDR4 RDIMMs will be priced about the same as when the Broadwell Xeon Ds were released. https://www.anandtech.com/show/12098/intel-to-update-xeon-d-in-early-2018-with-skylakesp-cores
|
|
#
?
Dec 2, 2017 04:51
|
|
|
I was actually checking memory prices earlier today, to see about adding another stick to my skylake xeon build. Yep, they're still almost double of what I paid in march 2016, and that was when there were almost no DDR4 ECC modules to choose from.
|
|
#
?
Dec 2, 2017 05:53
|
|
|
|
| # ? Apr 26, 2024 18:38 |
|
|
Alright. Finally going to upgrade my FreeNAS 9.3 up to 11. Hope poo poo works out. EDIT: Basics are all good it seems. Need to sort out what I was running in VMs in Virtualbox in a jail though since that is dead and gone. lurksion fucked around with this message at 22:53 on Dec 2, 2017 |
|
#
?
Dec 2, 2017 20:10
|
|
