Originally posted by Cthulhux
View Post
Announcement
Collapse
No announcement yet.
PC-BSD Rolls Into A Rolling Release Distribution
Collapse
X
-
Last edited by 0xBADCODE; 13 March 2013, 12:52 PM.
-
Hey, CthuIhux, (the fake one, not the real one), evading the questions and trolling around?
I know that you are somewhat slower than other people when it comes to comprehension and thinking for yourself, so I will repeat it for you:Is there a valid reason you used an already existing username, slightly modified in a way to resemble the look of the username?essentially most of the BSDs projects are a nothing more then a very bad stand up comedy. For example, thier BSD license
Oh, and this one:extremely bad system design
Leave a comment:
-
Originally posted by CthuIhux View PostThe BSD fuckers put google up there to fool people into thinking that even open source companies support them.
Originally posted by CthuIhux View PostYou see, right now AMD GPUs support in FBSD appears to be real crap.
That said: I have an ATI Mobility Radeon X1300. Performance in FreeBSD is almost as good as it is in GNU/Linux. BUT!! I can still boot into the latest FreeBSD kernel and get full 3d and resolution. If I try in Linux's latest kernel I have to disable KMS to even boot. So I get stuck in VESA
Leave a comment:
-
Originally posted by CthuIhux View PostBSD has been always about being proprietary company's bitch (a lot similiar to a sex slave but BSD loves self harming and degration and so they enjoy it).
Leave a comment:
-
Originally posted by 0xBADCODE View PostCorporate guys would not invest men or money if they can't expect return.
Leave a comment:
-
Originally posted by Cthulhux View PostAlso, (actually) FreeBSD does not really have much manpower.
And why should we as users care about the market?Last edited by 0xBADCODE; 13 March 2013, 09:02 AM.
Leave a comment:
-
Originally posted by intellivision View PostSo, what exactly is it that UFS2 lacks that EXT4 has?
Specifics please, not buzzwords, and correct spelling. Wow, you're even worse than that BSDbullshitDistro and SolarisCrap guy.
Then we come to next page in filesystems history and stuff like journalling. It has turned out that classic filesysterms are tradeoffs in engineering, again. In classic journalling filesystem design if you want full journal, first you commit data+metadata to journal area. This is "transaction intent". Then you transfer them to main data area. This is "transaction commit" (the similar notion also used some databases). If system crashes when "transaction intent" is not completely written to journal, it's completely discarded and it's rather "rollback". "All or nothing" principle returns "nothing". Filesystem is going to old state. If crash occurs after whole "transaction intent" is in journal, system will be able to complete crashed write using data from journal. Filesystem goes to new state, transaction completes. "All or nothing" returns all. This is basic theory. It works for databases and so on. But there is one major annoyance: in this ideal approach data have to be written TWICE. One time data+metadata go journal and another time they go to main area. Needless to say it haves huge write speed penalty. So actual filesystems are usually don't do things like this. They journal only metadata this way but not file data. As a result upon crash filesystem state is logically correct either way. However ... it's not applied to file data. It's possible to have half-old and half-new file in this way. And since there is no data in journal area, there is no way to complete partially written transaction for DATA. Only metadata would be correct. It saves a lot of time on fsck but does not really warrants files integrity and predictable states on crash. This is major problem of classic jornalling designs. In fact it's quite a bad tradeoff. Either you have poor write speed or sometimes you can face bad files.
Some bright heads invented really bright thing with all these journals: let the whole filesystem area to be "journal". No dedicated main data areas. So no commits to that area. No double writes. No speed losses. Problem solved! And it still retains full journalling properties: if write crashes, crashed record completely discarded. Filesystem is in old state. If write completes, record is here and filesystem in new state. And let overwrites not to destroy old data but hit new place instead. Then actual state of file is composed from the chain of logs (or equivalent structure). Hence it earned it's name: Copy On Write. This design copies written data to new place instead of rewriting them on old place. Yet another page in filesystem design history has been written. New designs started to appear. So called "log structured", "version-based" and other similar subflavours of technology started to appear. Some DBs also adopted this approach for similar reasons. ZFS has been one of early implementers of this technique. And you see, this technique really different. Ironically, generic journalling things from BSD guys never expected this kind of designs. They were meant for classic filesystems with classic journal approach or no journal. Ironically enough ZFS does not needs any kind of such subsystems and does CoW on it's own. Most new filesystems designs recognized that it could work better and adopted this approach. This way of doing things also makes it easy and straightforward to implement snapshots. Such filesystem inherently keeps multiple states. To travel back in time this design just have to ignore log records newer than X and voila - you can see pre-X state of things. Easy and cool. So snapshots are native part of this design. Sure, there are some drawbacks as well. It increases fragmentation and requires quite complex garbage collector to prevent volume from being filled by infinite older versions of previous states. But overall design is quite promising and most new designs usually center around CoW techniques.
So you see, FBSD guys essentially solved journalling problem in complex, generic ways. Only UFS could use it. And it has turned out not to work for newer CoW based designs which are quite custom and benefit from custom ways of doing journal-related things themselves.
Result? Awesome amounts of manpower wasted with almost zero user-visible results. Epic fail in project management, guys.
Also, AMD KMS support is being worked on
and will ship with FreeBSD 10, which isn't really an issue since most people use Intel or Nvidia chipsets these days anyway.Last edited by 0xBADCODE; 13 March 2013, 08:50 AM.
Leave a comment:
-
Also, (actually) FreeBSD does not really have much manpower.
And why should we as users care about the market?
Leave a comment:
-
Originally posted by 0xBADCODE View PostNope, I don't. You see, right now AMD GPUs support in FBSD appears to be real crap. So if you happen to have more or less recent AMD GPU, it's likely to be very crappy experience. Sure, you can find some ancient Linux to beat it. But why BSD guys always try to compete with some crap and weaklings? Come on, market does not needs losers and weaklings. Can't you get this simple thing?
Nope, I don't. I've seen in their mailing list how they developed it. They told they don't have resources for major disk structures overhaul due to lack of manpower. So UFS2 is almost the same as UFS in terms of actual on-disk structures. It even lacks extents. While EXT4 got those. Just as each and every modern design around. Sure, it required major disk structures rework and partial loss of compatibility. But at least Linux guys could afford enough resources to rework their ancient block-based allocator to extent-based allocator. This in fact dramatically speeded up EXT4 over EXT3 on modern volumes, files and most workloads. But BSD guys proven once more they're pathetic losers. Instead, they rather created very generic jornalling subsystem. Sure, idea itself it not so bad in theory. But it took awful time and effort to implement that. And at the end of day only crappy and ancient UFS could use these services. So while UFS2 does not needs fsck on every crash, it's still slow and antique design inside. And virtually every benchmark shows it's crappy nature. You can't fool laws of physics after all. And LOL, ZFS is CoW-based design and it simply does not uses or needs such services as it does journalling equivalent on it's own. Guys created cool generic solution for ... one defective and old filesystem! Bah, that's impressive!
No, honestly, it's absolutely amazing example of how people could waste so many manpower with so little user-visible results at the end. Quite the same with clang replacement, etc. In fact it's a decent example of really poor project management.
Specifics please, not buzzwords, and correct spelling. Wow, you're even worse than that BSDbullshitDistro and SolarisCrap guy.
Also, AMD KMS support is being worked on and will ship with FreeBSD 10, which isn't really an issue since most people use Intel or Nvidia chipsets these days anyway.
Leave a comment:
-
Originally posted by Cthulhux View Post1. You mix up Linux with some of its distributions. To be fair, you should consider comparing it with PC-BSD.
2. You mix up UFS with UFS2.
No, honestly, it's absolutely amazing example of how people could waste so many manpower with so little user-visible results at the end. Quite the same with clang replacement, etc. In fact it's a decent example of really poor project management.Last edited by 0xBADCODE; 13 March 2013, 01:59 AM.
Leave a comment:
Leave a comment: