Announcement

Collapse
No announcement yet.

Ubuntu 19.10's ZFS TODO List Goes Public - A Lot To Of Work Left

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Ubuntu 19.10's ZFS TODO List Goes Public - A Lot To Of Work Left

    Phoronix: Ubuntu 19.10's ZFS TODO List Goes Public - A Lot To Of Work Left

    We've been quite eager to see what happens around Ubuntu 19.10's ZFS support with their plumbing this out-of-tree file-system into their new desktop installer and a lot of other Ubuntu happenings around ZFS. There is now at least a public TODO list/board outlining some of their ZFS work for the Ubuntu 19.10 Eoan Ermine cycle...

    http://www.phoronix.com/scan.php?pag...-ZFS-TODO-List

  • k1e0x
    replied
    Originally posted by gclarkii View Post
    Yes, in most cases compression would be a help. But for whatever reason, with this EB-1033, if I enable compression, and it does not matter what type, access to that data set or pool, depending on where I enabled it at, heads for the basement.

    The the ASUS file server is actually new, being less than 2 weeks old. That hexacore I mentioned earlier had this position before.
    I really need to sit down and do a some serious tuning, upgrading and bench marking. But as it works right now for what we need with ZERO sign of lagging or hesitation, I've just been too lazy to get it done. But thanks for pointing it out anyway...
    Hmm that is interesting. Only on really old processors should that be the case.. or on algorithms that compress incompressible data like gzip. Well, if you tested it you tested it.. it's an oddity tho. If it's used for media it is a waste of time to compress.

    Leave a comment:


  • gclarkii
    replied
    Yes, in most cases compression would be a help. But for whatever reason, with this EB-1033, if I enable compression, and it does not matter what type, access to that data set or pool, depending on where I enabled it at, heads for the basement.

    The the ASUS file server is actually new, being less than 2 weeks old. That hexacore I mentioned earlier had this position before.
    I really need to sit down and do a some serious tuning, upgrading and bench marking. But as it works right now for what we need with ZERO sign of lagging or hesitation, I've just been too lazy to get it done. But thanks for pointing it out anyway...

    Leave a comment:


  • DoMiNeLa10
    replied
    Originally posted by k1e0x View Post
    Hurd is functional and it boots.. Start improving it.
    It's far from it. Even in a virtual machine, I never managed to get it to shut down properly. At this point, I think the best thing to do would be to pick some better microkernel and start from there.

    Leave a comment:


  • k1e0x
    replied
    gclarkii Just a note here, Compression on ZFS (with LZ4 and maybe Zstd) is a performance boosting feature because the processor is so much faster than the disk and the way it's compressed. It actually give you more cache hits too because it's compressed in ram as well.

    Originally posted by DoMiNeLa10 View Post

    I just want a proper GNU operating system. That's it. I wish I could just say GNU instead of saying GNU/Linux.
    Why? For me I want to move technology forward because I believe the OS and platform isn't a finished problem and we can improve.

    Your concern here is the name? They could call it ButtmunchOS for all I care so long as it was highly functional. I don't believe it's about the name.. You want to force your ideas of "Free Code" on everyone else.. You are free.. to do whatever I say maybe? lol That isn't the way man, dictatorships are not helping the people. Freedom is allowing and accepting that people may have different ideas than you do.

    Take FreeBSD for example.. It can be used for commercial products. (and IS, PS4 and Nintendo Switch are FreeBSD And tons of other things)
    Though I like MPL style the developers are ok with it and being business friendly has it's advantages. Juniper and NetApp took FreeBSD code as well and they forked it and now they own the mess.. A smart business will upstream.. Like Netflix. You don't always *need* teeth to protect it. forking alone is dangerous enough and business know it.

    And you already HAVE a Gnu OS. Hurd is functional and it boots.. Start improving it.
    Last edited by k1e0x; 06-07-2019, 12:59 PM.

    Leave a comment:


  • gclarkii
    replied
    Originally posted by linuxgeex View Post
    God help them if they implement /boot on ZFS because it will hugely increase the kernel/module load size before the Linux kernel takes over driving the IO. As is it already takes minutes to bring up a Power9 machine, even on SSD.
    Well don't blame ZFS for this. I operate a FreeBSD 12.0 based ZFS file server running Samba to export disk to a couple of Windows 10 work stations and two other FreeBSD boxes. This is a ASUS EB-1033 with a D2700 atom, 4GB, 500 GB sata internal and at this point 5 USB3 disks, three 2TB in a triple mirror and two 4TB in a dual mirror I've got a semi-modern 16GB usb flash plugged into one of the USB2 ports acting as L2ARC. It runs headless stashed in an OLD huge ATX case, 4 5.25 external bays and 3 3.5 internal bays. The power supply that is in there is a 5V/3V3 heavy 400W ps, which might tell you how old it is. It's been recapped and I use a boost converter to produce the 19V the EB-1033 requires from the ATX 12V. Three of the drives are 2.5" SATA drives that do not requiring external power, even though that USB3 hub is powered. The other two drives are 3.5" and so require external power, all supplied by the ATX power supply.
    The EB-1033 has two USB3 ports on different buses from what I can tell. Right now I'm seeing 400MB/s on the USB3 based drives, 300MB/s SATA2 from the internal drive and 40MB/s from the USB2 Cruzer Drive.

    Once tuned, it boots to kernel load in about 40 seconds to about 60. From there it screams and in less than a total of about 180 seconds I can ssh into the machine. It has a Gig link to a ASUS GB switch.

    I don't enable much of anything, no compression, etc, etc. This is only a dual core/4 thread 2.17Ghz Atom with 4GB of ram. I use it as a VERY simple file server with a little bit of file redundancy and integrity. Once a day I snapshot both pools and send'em to one of the FreeBSD work stations, where it is stored to both an internal zpool HGTS 4TB I picked up refurbished on Amazon with a 4 year warranty and an another HGTS in an external eSATA enclosure to act as a basic backup. At the rate I'm using this drive, I've got easily at least a year or more of daily backups. It handles five or six Git repos, a sub repo, firmware archival duties plus it stores our library as a network mount. Note, this machine does use compression as it is a Hexacore AMD Athlon II with 32GB and so has power to spare. To be truthful, if I had not picked up that EB-1033 cheap as hell($35 on sale with the 4 GB memory and 500GB drive and a power supply I don't use), I would most likely be using this to serve files and be a design station.

    Yes, ZFS is the largest kernel module I could find on my 12.0 box weighing in at 3.1MB. That said, if your worried about loading something that small, are you sure you want to run ZFS? Except for those two zfs backup drives mentioned above, both of the workstations run UFS2 and not ZFS. In some ways UFS2 is much faster than ZFS. I just don't have large enough VDevs or the correct "HBA"s to really make the bits fly. I've seen systems with 10-30 SAS drives per VDev, combined into a simple mirror of two or more VDevs or a RaidZ with anywhere from 5 to 8 VDevs each with 2 to 30 drives in it, depending on what their trying to achieve. When the server has dual quad xeons, SSD main and cache drives, lots of HDD SAS ZFS drives and 96GB of ECC memory, It can really make the bits fly. I wish I had it, even though I really don't have a need...

    Damn, sorry about the rambling, I really need to go crash. My L433 sim run finished about 10 mins ago so I'm headed to bed.

    Leave a comment:


  • DoMiNeLa10
    replied
    Originally posted by k1e0x View Post

    The CDDL does the same thing without the virus nature.. CDDL code can not be stolen, if it could have been Oracle would have taken ZoL a long time ago and sold it, it has quite a few nice features not in Oracle ZFS(tm).

    Also you *do* realize that not everyone in the world wants to use the GPL right? Do they have no voice in your perfect world?

    Also you keep talking about the users in a "protect the children" kind of way. Far as I'm aware, the license dosen't apply to the users and users do not have to agree to it. It's for distribution. So you have MPL/CDDL that allow a free world with code that can be of any license.. and the GPL that enforces uniformity. If we want Linux to really really be the best, then we need to be more flexible on what code and technology it can accept. Or maybe... maybe the goal isn't to build great systems and a terrific OS.. Maybe the goal is zealotry and stoking that weirdo Stallman's ego? I mean what are we actually trying to accomplish here? - This is one of the reasons I left the Linux camp (for bsd). I wanted to work with an OS that was about the technology.. not the religion.
    I just want a proper GNU operating system. That's it. I wish I could just say GNU instead of saying GNU/Linux.

    Leave a comment:


  • k1e0x
    replied
    Originally posted by DoMiNeLa10 View Post

    Considering that GPL makes efforts to prevent from abuse of code, it makes sense to pick it as the license and use it to ensure nobody will abuse users with your code. I use GNU for the sake of using GNU (it's high quality software), and I want the ecosystem to be consistent. Having a single and strong license helps to prevent abuse, and the "viral" nature of GPL is a natural solution to enforce itself.
    The CDDL does the same thing without the virus nature.. CDDL code can not be stolen, if it could have been Oracle would have taken ZoL a long time ago and sold it, it has quite a few nice features not in Oracle ZFS(tm).

    Also you *do* realize that not everyone in the world wants to use the GPL right? Do they have no voice in your perfect world?

    Also you keep talking about the users in a "protect the children" kind of way. Far as I'm aware, the license dosen't apply to the users and users do not have to agree to it. It's for distribution. So you have MPL/CDDL that allow a free world with code that can be of any license.. and the GPL that enforces uniformity. If we want Linux to really really be the best, then we need to be more flexible on what code and technology it can accept. Or maybe... maybe the goal isn't to build great systems and a terrific OS.. Maybe the goal is zealotry and stoking that weirdo Stallman's ego? I mean what are we actually trying to accomplish here? - This is one of the reasons I left the Linux camp (for bsd). I wanted to work with an OS that was about the technology.. not the religion.
    Last edited by k1e0x; 06-06-2019, 12:59 PM.

    Leave a comment:


  • DoMiNeLa10
    replied
    Originally posted by k1e0x View Post

    So what you're saying is you don't care that the code is good or the license is acceptable, free and open source (more so than Linux itself really).. and you would *like* to see that technology be destroyed in order to protect the GPL for.. reasons... do you not realize why I liken you to a religious nut job?
    Considering that GPL makes efforts to prevent from abuse of code, it makes sense to pick it as the license and use it to ensure nobody will abuse users with your code. I use GNU for the sake of using GNU (it's high quality software), and I want the ecosystem to be consistent. Having a single and strong license helps to prevent abuse, and the "viral" nature of GPL is a natural solution to enforce itself.

    Leave a comment:


  • skeevy420
    replied
    Originally posted by Panda_Wrist View Post

    That's why I wrote " might be " and had quotation marks around gpl violation. Because some people and the gpl feel that it is. Wether it is or not a violation ( I don't think it is), I'm going to continue using zfs on linux.
    ZoL clearly follows every GPL requirement to the letter and the main difference between the two is allowing closed source build systems. When a CDDL project has a 100% open source setup with its build system and code then it is in compliance with the GPL -- that describes ZoL.

    If that isn't the case then all the MIT and BSD code in the kernel is at jeopardy if this ever does go to court. That line of thinking alone should be enough to scare off all the major players -- if they win, they could potentially screw themselves very hard because it a court could rule that GPL can only be used with GPL regardless of intent or compatibles between differing licenses.

    Leave a comment:

Working...
X