4 ms·

The USB 3.0 ports should give 100MB/s (Benchmarks show up to +300MB/s), and you can mirror two disks for RAID-1. With 4GB of RAM, you'd even be able to use ZFS

The USB 3.0 ports should give 100MB/s (Benchmarks show up to +300MB/s), and you can mirror two disks for RAID-1. With 4GB of RAM, you'd even be able to use ZFS.

morphle 7y ago
not these USB 3.0 ports. Connected with shared 4 Gbps, they only run at 4/5th, 1/2 or less of the full 5 Gbps speed.
- tmikaeld 7y ago
  Which means, 2 ports and two drives (not ssd) should give at least 100MB/s each. Maybe I wasn't clear enough on that. - 1Gbit ethernet - 1Gbit USB 3.0 port 1 - 1Gbit USB 3.0 Port 2 Total: 3Gbit of 4Gbit, having some margin to spare.
  - fanf2 7y ago
    The Ethernet is directly connected to the SoC, it doesn't share the 4Gbit PCIe with the USB ports
    - dotancohen 7y ago
      On the older RPIs the ethernet was connected via the USB ports, so I understand GP's stance.
  - morphle 7y ago
    It has: 1 Gbit Ethernet over RGMII 4 Gbit PCIe Gen 2 shared by 2 USB 3.0 (so 1x 4 gbps max or 2 x 2 Gbps shared) Total: 5 Gbits, no margin to spare
    - kllrnohj 7y ago
      > or 2 x 2 Gbps shared Only if both drives are saturated at the same time. But since the context here is a NAS usage, there's a max 1gbps upstream/downstream in the first place so you could comfortably do a 4 drive mirror NAS off of this and not have any bottlenecks on the USB 3.0 -> SATA side of things.
chongli 7y ago
With 4GB of RAM, you'd even be able to use ZFS. Does it use ECC memory? If not, you should not use ZFS. Without ECC memory, ZFS carries the nasty risk of writing good data with a bad checksum, leading to data loss which would not occur in other filesystems (those that do not try to correct errors on the fly like ZFS).
- nabla9 7y ago
  Most other filesystems don't do checksum at all, so you can't even discover that you have a data loss/or corruption. Only difference is that other filesystems may have better tools to rescue damaged filesystem after metadata is corrupted.
- mclehman 7y ago
  In the case of a traditional filesystem, similarly unfortunate memory corruption would just affect the actual data directly. Forgoing ZFS throws out the happy path where the checksum does catch a data error and repairs it transparently. Edit: To add on to this, ZFS doesn't do parity-based repair at the checksum level. In the mentioned case (good data, bad checksum) no copy of the block will match the checksum and there will be no supposedly-good block to copy over.
- TheAceOfHearts 7y ago
  As I understand it, this is just a myth. Here's a post [0] from Matthew Ahrens, one of the co-founders of ZFS who has remained active in its development: > There's nothing special about ZFS that requires/encourages the use of ECC RAM more so than any other filesystem. If you use UFS, EXT, NTFS, btrfs, etc without ECC RAM, you are just as much at risk as if you used ZFS without ECC RAM. Actually, ZFS can mitigate this risk to some degree if you enable the unsupported ZFS_DEBUG_MODIFY flag (zfs_flags=0x10). This will checksum the data while at rest in memory, and verify it before writing to disk, thus reducing the window of vulnerability from a memory error. [0] https://arstechnica.com/civis/viewtopic.php?p=26303271#p26303271 https://arstechnica.com/civis/viewtopic.php?p=26303271#p2630...
  - flyinghamster 7y ago
Maledictus 7y ago
I tried ZFS with 4 GiB RAM in the past and it was not fun. Does it work for you?
- tmikaeld 7y ago
  I have several 4GB machines that's used in production (ZFS on Linux), however, it's not problem free and require lot's of tuning, the correct workload and smaller sizes of disks. A guide: https://wiki.freebsd.org/ZFSTuningGuide https://wiki.freebsd.org/ZFSTuningGuide

The USB 3.0 ports should give 100MB/s (Benchmarks show up to +300MB/s), and you can mirror two disks for RAID-1. With 4GB of RAM, you'd even be able to use ZFS.

morphle 7y ago

not these USB 3.0 ports. Connected with shared 4 Gbps, they only run at 4/5th, 1/2 or less of the full 5 Gbps speed.

tmikaeld 7y ago
Which means, 2 ports and two drives (not ssd) should give at least 100MB/s each. Maybe I wasn't clear enough on that. - 1Gbit ethernet - 1Gbit USB 3.0 port 1 - 1Gbit USB 3.0 Port 2 Total: 3Gbit of 4Gbit, having some margin to spare.
- fanf2 7y ago
  The Ethernet is directly connected to the SoC, it doesn't share the 4Gbit PCIe with the USB ports
  - dotancohen 7y ago
    On the older RPIs the ethernet was connected via the USB ports, so I understand GP's stance.
- morphle 7y ago
  It has: 1 Gbit Ethernet over RGMII 4 Gbit PCIe Gen 2 shared by 2 USB 3.0 (so 1x 4 gbps max or 2 x 2 Gbps shared) Total: 5 Gbits, no margin to spare
  - kllrnohj 7y ago
    > or 2 x 2 Gbps shared Only if both drives are saturated at the same time. But since the context here is a NAS usage, there's a max 1gbps upstream/downstream in the first place so you could comfortably do a 4 drive mirror NAS off of this and not have any bottlenecks on the USB 3.0 -> SATA side of things.

chongli 7y ago

With 4GB of RAM, you'd even be able to use ZFS. Does it use ECC memory? If not, you should not use ZFS. Without ECC memory, ZFS carries the nasty risk of writing good data with a bad checksum, leading to data loss which would not occur in other filesystems (those that do not try to correct errors on the fly like ZFS).

nabla9 7y ago
Most other filesystems don't do checksum at all, so you can't even discover that you have a data loss/or corruption. Only difference is that other filesystems may have better tools to rescue damaged filesystem after metadata is corrupted.
mclehman 7y ago
In the case of a traditional filesystem, similarly unfortunate memory corruption would just affect the actual data directly. Forgoing ZFS throws out the happy path where the checksum does catch a data error and repairs it transparently. Edit: To add on to this, ZFS doesn't do parity-based repair at the checksum level. In the mentioned case (good data, bad checksum) no copy of the block will match the checksum and there will be no supposedly-good block to copy over.
TheAceOfHearts 7y ago
As I understand it, this is just a myth. Here's a post [0] from Matthew Ahrens, one of the co-founders of ZFS who has remained active in its development: > There's nothing special about ZFS that requires/encourages the use of ECC RAM more so than any other filesystem. If you use UFS, EXT, NTFS, btrfs, etc without ECC RAM, you are just as much at risk as if you used ZFS without ECC RAM. Actually, ZFS can mitigate this risk to some degree if you enable the unsupported ZFS_DEBUG_MODIFY flag (zfs_flags=0x10). This will checksum the data while at rest in memory, and verify it before writing to disk, thus reducing the window of vulnerability from a memory error. [0] https://arstechnica.com/civis/viewtopic.php?p=26303271#p26303271 https://arstechnica.com/civis/viewtopic.php?p=26303271#p2630...
- flyinghamster 7y ago

Maledictus 7y ago

I tried ZFS with 4 GiB RAM in the past and it was not fun. Does it work for you?

tmikaeld 7y ago
I have several 4GB machines that's used in production (ZFS on Linux), however, it's not problem free and require lot's of tuning, the correct workload and smaller sizes of disks. A guide: https://wiki.freebsd.org/ZFSTuningGuide https://wiki.freebsd.org/ZFSTuningGuide