Fwd: The Morning Paper: NOVA - A log-structured file system for hybrid volatile/non-volatile main memories

Peter Jeremy peter at rulingia.com
Mon May 9 06:26:34 UTC 2016


On 2016-May-08 14:08:47 +0300, Lev Serebryakov <lev at FreeBSD.org> wrote:
>Hello Poul-Henning,
>
>Sunday, May 8, 2016, 1:47:39 PM, you wrote:
>
>> The problem with that kind of hardware is that either you specialize
>> 100% for that vendors product, with resulting lock-in, or you satisfy
>> yourself with a generic solution which works an anything sensible.
>
>> I don't think FreeBSD has the bandwidth for anything but the second option.
>
>  Really, Intel is not "that vendor" now. It is THE Vendor. FreeBSD has a LOT
>of specialization for this vendor (all i386 code, NVMe, and other), and many
>thing from Intel (or which becomes widespread in Intel-based systems, like
>PCI, which is not exactly "from Intel", of course) becomes industrial
>standard.

There's a difference between FreeBSD having i386/amd64 as the main Tier-1
platform and supporting hardware that is only supported by some (not even
all) x86 CPUs manufactured by Intel.  There are a number of other vendors
that supply i386/amd64-compatible CPUs.  Whilst we do have CPU-specific
code (for CPU initialisation and hwpmc), there's a big step between that
and supporting a filesystem that requires vendor-specific hardware and a
vendor-specific CPU.

> I don't think such type of memory will be very-Intel-specific for long. HP
>works on something comparable, and other companies too.

"Something comparable" doesn't mean it'll be compatible - Intel have a
vested interest in locking customers into their CPUs.  And the hype
surrounding "The Machine" has been toned down significantly - suggesting
that HP are having a harder time productizing their NVM than they expected.

-- 
Peter Jeremy
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 949 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-fs/attachments/20160509/1a6a145a/attachment.sig>


More information about the freebsd-fs mailing list