[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: Ten Reasons To *NOT* Use ZFS:



Hi Keith,

I must say that reading your post convinced me that some of the /.
trolls are present over here also. Let's take a look at a few of the
facts that you list over here. As a side note, I run RHEL and FC on my
servers. As Solaris still has a few down-sides, I am not considering
switching yet.

Keith G. Robertson-Turner wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
>
> In response the Josenildo Marques' message regarding Sun Microsystems'
> announcement of the release of their new "blazingly fast" filesystem,
> ZFS.
>
> Ten Reasons To *NOT* Use ZFS:
>
> 1. The License (CDDL) "has some complex restrictions that make it
> incompatible with the GNU GPL"
>
> http://www.gnu.org/philosophy/license-list.html
>   
Hmm, NTFS is proprietary, and we still have it in Linux. We also have
FreeVXFS and a few others. Clean-room ring any bells? It's also easier
considering that we have access to the source-code.
> 2 . ZFS does not support the necessary extended attributes and ACLs to
> enable the implementation of SELinux security. Instead Sun prefers the
> deployment of its own security software "Trusted Solaris", which is
> not FOSS and runs at a cost of "$995 per seat for the Standard Edition
> Desktop System to $79,495 for the Certified Edition Data Center
> Server."
>
> http://www.eweek.com/article2/0,1895,1641823,00.asp
>   
The part with the necessary extended attributes and ACLs is bull. The
reason is simple. It supports ACLs and it supports extended attributes,
it's just a matter of using that support in the Linux port, and
mapping-it to the corresponding Linux API. Since we have no real Linux
port yet, we have no acl and xattr. There are some filesystems that come
to mind right now that are available on Linux and still don't support
xattr and acl "the Linux way" (ReiserFS ring any bells?).
Going on to the second part: it's bull. Most of the Trusted Solaris
extensions have already been moved to OpenSolaris and Solaris 10, others
will follow. Sun is working hard to move the Trusted Extensions that
they had in their CDE desktop to GNOME. This seems like open-source to
me, and if we have open-source, you won't have to pay the $995-$79495
for the product.. The whole pricing is completely irelevant, as
OpenSolaris is still to have it's first release (Solaris 11, in the Sun
versioning). It's just like comparing the Fedora Core 0.9x releases with
RedHat 9. There were a lot of voices in those days that accused RedHat
of abandoning the community. Now Fedora Core has an even stronger
community than RedHat Linux ever dreamed of having.
> 3. This is primarily a Solaris product, and has nothing whatsoever to
> do with Linux. The possibility of future Linux support cannot be ruled
> out, however - at what cost - both in financial *and* license terms?
>   
Yeah, and XFS was primarily an Irix product. Guess what, it's
open-source and feature complete (should I bash ReiserFS here or not?)
> 4. This is a product for the server room, not the desktop. The main
> bottleneck on a desktop system is the underlying storage hardware
> itself, not the filesystem. If you have the budget and requirements
> for a Solaris server, then buy a Solaris server, otherwise there are a
> myriad of other (IMHO) better solutions. Remember, Trusted Solaris is
> not Open Solaris.
>   
Hmm... filesystem snapshots for the Documents folder. Add a hard-drive
and extend the file-system to that one. Those are features that we don't
need on our Home Desktop systems! Yeah right! In my opinion a tight
integration between LVM and EXT3, plus a few extra features to EXT3
would accomplish the same results, but without trolls like you bashing
Sun just because they have nothing better to do, just because it's Linux
that came up with that. NTFS was believed to be Server-ONLY a few years
ago also. Most systems these days come with Windows XP preinstalled on a
NTFS formated partition. Figure that out.
The Linux community has a lot of trolls that want to reinvent the wheel,
just because it wasn't them who came up with the ideea. The whole Kernel
Probes thing in Linux vs. DTrace! Dtrace still is better and faster by
miles. There are hundreds of other examples, but they are completely
ignored by most.
> 5. Linux already has a (arguably) superior filesystem in the form of
> Reiser4, which is much more extensible and uses a dancing trees
> system, which in itself is a great improvement over the old block tree
> method employed by ZFS. Yes Reiser4 does not currently support SELinux
> extensions either, but then the book's not closed on Reiser4
> development, and it *is* Free, after all.
>   
Superior my ***. Ext2/3 is also extensible, but only ReiserFS makes a
whole marketing campaign out of that. The dancing trees thing is
completely useless for me. Let me explain why. If I want performance,
I'll go to ext3 and get SCSI/SAS instead of (S)ATA. But what ReiserFS 4,
after some time since it appeared can't offer me is support for
xattr/acl, "the Linux way", like any other normal Linux file-system
(ext/jfs/xfs). It seems like I'm not the only one over here. If Reiser4
was that good the kernel engineers at RedHat would be all over-it
implementing the missing features, and we'd have-it in Rawhide in a few
weeks. When the next RHEL release features also ReiserFS, I'll switch.
The more extensible thing is crap.
> 6. ZFS is a product of Sun Microsystems, developers of Sun Java. The
> distribution license for Sun Java has caused so much controversy over
> at Debian that they're threatening to pull the package from non-free
> and even dissociate with Sun's legal team, the SPI. Sun Java is
> subject to draconian export restrictions imposed by the US government,
> against any country not favoured by the US, and its SDL dictates that
> "you do not combine, configure or distribute the Software to run in
> conjunction with any additional software that implements the same or
> similar functionality or APIs as the Software"
>
> http://download.java.net/dlj/DLJ-v1.1.pdf
>
> This would preclude distributing GNU GCJ and Sun Java on the same
> system. Do you really want Sun dictating to you what software you have
> on your computer?
>
> Do you really want software from this company on your system? Frankly
> you might just as well install Microsoft Windows and be done with it.
>
>   
I fail to see the parallel with Debian. It's only political moves. The
Java product is one issue, the ZFS is another, OpenOffice is another,
NFS4 is another, X11 is another, and GNOME is another. All of them are
Sun products or have a lot of Sun code in them. Without Sun, the GNOME
desktop that FC defaults to would be a different story. They added some
order into chaos (as a lot of RedHat engineers will testify). What about
X11, you say? Ask Mike Harris about the contributions brought to Xorg by
Sun (Alan Coopersmith rings any bells?).
> 7. Sun is claiming all kinds of performance benchmarks, but the
> "blazing performance" has yet to be independently verified, and few
> comparisons exist between ZFS and any other high performance
> filesystem. Ultimately Sun are motivated by money, not integrity, so
> if their motivations are questionable, then so are their claims.
>   
That's marketing. Learn to live without-it. Set your own standards for
performance, price and features, conforming to your needs/budget. If a
product fits, go for it, if not, don't listen to markeing. Apple said 2
years ago, that 64bit is a must and the future, yet, the new Apple
systems are 32bit. It seems to them that Intel was more important than
64bit. Apple said that Intel is a performance gain, but looking at the
setiathome stats, I still see Power in the top, followed by Opteron.
Thus, marketing shouldn't really be relevant to an engineer.
> 8. Despite their rivalry and even previous court battles, Sun is
> rather too close to Microsoft for comfort.
>
> "A year ago you could say we were moving from the courtroom and
> entering the computer lab," Ballmer said, summing up the rivals'
> progress. "Twelve months later I think we're poised, thanks to the
> work of hundreds of engineers on both sides, to leave the computer lab
> and enter the market place together."
>
> http://www.internetnews.com/dev-news/article.php/3504976
>   
That's what Wall Street needs. Wall Street needs the Sun Enterprise
servers that they have running Solaris, to go along well with their
Sun/HP/IBM/etc. department servers running Windows and with their
Windows XP/2000 workstations.
> 9. After SCO made ridiculous claims about being the owner of Linux IP,
> Sun apparently agreed with them, since they entered into a business
> agreement to license Unix from SCO.
>
> http://www.newsfactor.com/perl/story/21894.html
>
>   
No one, contested the Unix IP that SCO has. Sun didn't sign for the
rights to distribute Linux, but to incorporate UNIX technology into
Solaris. Guess what!?! Solaris IS UNIX (as certified by The Open Group).
> 10. 64 bit architectures have been around for years, and are in fact
> virtually mainstream these days, and yet Sun (having released a 64 bit
> version of Java) have yet to release a 64 bit Mozilla plugin for it, to
> enable Java Applets on Gecko based browsers. The Linux AMD64
> self-extracting file of Sun Java JDK (jdk-1_5_0_07-linux-amd64.bin) is
> some 41.76 MB ... the missing plugin (libjavaplugin_oji.so) is only
> *77 bytes* !!!
>
> And *why* does Sun refuse to include a tiny 77 byte Mozilla plugin
> with their 64bit Java? Because, apparently, they think that there are
> no 64bit versions of any Gecko browsers available, or to be more
> precise, because Mozilla does not distribute a 64bit binary
> upstream???
>   
The Sun JDK is a Sun supported product, thus they only give plugins for
supported products. This means Mozilla.org builds for Mozilla. The
source is available should you want to recompile the plugin for 64 bit.
Should I also mention that all other "must-have" plugins are 32bit only?
See flash, shockwave, acrobat, quicktime, windows_media. I am sure that
they will be ported to 64bit when mozilla feels that 64bit is worth
having an aditional supported platform.
> A company that employs such retarded logic, and has such a slow
> response to technological change, cannot be trusted to keep pace with
> changing hardware trends in other sectors of the market, such as
> storage hardware for example. Imagine that you had paid a considerable
> amount for new storage solutions, either at home or at work, only to
> discover that Sun's "blazingly fast" filesystem was completely
> unsupported on that hardware. How long might you have to wait for your
> ($995 - $79,495) Trusted Solaris with ZFS to be redeployed?
>   
Well, RHEL is only supported on some hardware and I see no complaints
about that. I haven't seen Apple PPC versions of RHEL, although it does
support the IBM PowerG5 equivalent, and the Anaconda patches are
available for some time now. Actually, RHEL has the kernel package split
into kernel-$ver-$arch and kernel-unsupported-$ver-$arch. 25% of the
Linux drivers can be found over there. Talk about unsupported. The
wonderfull part of OpenSolaris is that you can add support for another
storage solution. Although Sun does not have or support the PowerPC
arch, OpenSolaris has been ported to it (see the polaris branch of the
genunix cvs). You mention ReiserFS. Is that supported in RHEL4? What
about 5? It's not in rawhide, thus it's safe to assume that it won't be
in the soon to be released beta. Does Red Hat have a retarded logic?
> I used to be under the impression (or at least ever hopeful) that Sun
> might one day wake up to the FOSS trends in the industry, and
> reconsider their licensing and methods, but with each new announcement
> from Santa Carla they just get worse.
>   
It's not easy to move something as big as Solaris, that is full of IP
protected code, into OpenSource. They considered the GPL, but they
didn't find-it enough. They said that they might reconsider when the
GPL3 arrives.
> So to anyone considering posting here, on a Fedora mailing list of all
> places, announcements from Sun Microsystems, I say that IMHO any news
> from Sun is unwelcome, unless that news is the wholesale GPL
> re-licensing of their entire product catalogue.
>   
Hmm... OpenOffice is GPL, GNOME is GPL, their latest CPU is GPL (and
it's a fast one, see the heanet.ie experiment against Itanium and Xeon),
Xorg is something compatible (I'm not sure). Sun wrote a lot of the code
out there that you're using. I am disappointed with people like you. The
world is not split into black and white. There's always a gray zone. Sun
is making huge efforts to please everyone. Sun is moving into the FOSS
zone, but it can't happen over night.

The only real reasons not to use ZFS that you gave were that it didn't
support xattr and acl and that it's not open-source. The rest were Sun
trolling. Keep in mind that ZFS IS BY ALL DEFINITIONS OPENSOURCE. Keep
in mind that a clean-room re-write is doable and that it might actually
be the only solution technical solution also, since Linux and Solaris
have different internals. And ZFS supports xattr and acl, but not the
Linux way since it's not yet ported.
> - --
> Regards,
> Keith G. Robertson-Turner.
>
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.3 (GNU/Linux)
> Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org
>
> iD8DBQFEmgGbi+i8fEQ9xnoRAu+kAKCRaYxkLkNalnzfphkpmqj7uFqAHACeLY8B
> aHo2njLLA77E5qqmYRzmN0Q=
> =eiMo
> -----END PGP SIGNATURE-----
>
>   



[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]