“Learn the intricacies of the AIX file system framework. Every operating system provides a native kernel framework that kernel developers have to understand and adhere to when developing a piece of a kernel component for that operating system. This article sheds some light on the AIX file system framework. You need to understand the framework in order to develop a new file system, or to port an existing file system to the AIX operating system.”
I don’t know if I just have had bad luck with in the environment I work in but AIX is full of empty “feature” promises. I will gladly take Solaris, HPUX and Linux over AIX anyday. Any unix that sports a “registry” is not my idea of unix
Solaris is a fantastic operating system. You should see if you can migrate to it. Very stable, easily customised, easy to use, feature rich (even exclusives) and very well supported.
OMG, could you please post some more canned “AIX topic” responses?
I a part of a team of about 20 Unix admins who adminstrate almost 400 Unix systems for a Fortune 500 company. There is an even mix of Linux, Solaris, and AIX. AIX has its strengths, and the ODM is not anything like the Windows registry.
Actually, one of AIX’s strengths is its ease of administration. Anyone who tries to tell you Solaris is easier to administer than AIX is trying to sell you something. AIX has more administrative luxuries than any Unix out there. Always has, AFAIK.
I am not a huge AIX fan, but I’m not a huge Solaris fan either. I’m pretty impartial, as long as it’s not Windows, but really, it baffles and annoys me how many people keep posting this “AIX is Unix w/a registry” drivel.
Hence the reason I said a couple of posts back in a topic that this forum has become over run with Linux fanboys who migrated in the last two years, and try to claim they’re a UNIX expert.
This registry crap has been repeated so many times, I tend to skip over when idiots start to mention it; the problem with Windows isn’t the idea of the registry but the fact that they had no developer policies in place on how to interact with it, and worse still, its all in binary format. As far as I remember, most operating systems have a registry of some sort, be it as plain text or XML; the settings have to be saved somewhere – and thats a registry.
I don’t know anything about AIX but I would find it rather unusual if IBM was going out and throwing settings on mass into a giant binary file. IBM does some stupid things, but I doubt a binary registry would be one of them.
AIX easy to administrate? I guess if you ONLY used AIX then it wold be easy. Myself, after using Solaris, Linux, BSD, HPUX etc, that moving to an aix box is no where near as easy as transversing through the others are. Sure they all have their hang ups but when seasoned administrators can’t tell you what the command line version of the SMIT option their using is, thats rough. I would also like to professionaly disagree with you on the ODM not being anything at all like a windows registery, at the the end of the day they both are a db that contains system configuration information. Sure, the winreg is way more intrusive but what got me hooked on unix in the first place was the whole “wow, just edit this file and thats it! awesome” Especially given the tools and its billions cmdline options to try to querey ODM its devices, its attributes etc.
I will however agree with you that it does has it strenghts but once again, I don’t think ODM is one of them. Its ability to scale inside the environment, Its ability for LPAR is awesome and coming vituralization is great too. I guess another tipping point for me is support. When it comes to big metal, SUN has always provided great service to us, IBM so far has been very spotty with hit and miss experiances.
I will agree with you 100 percent on the “as long as it’s not windows”. I rather run my enterprize on saltine crackers than on windows.
What have you been smoking?
Don’t want to start a flamewar or anything. I happend to be a unix admin in telco company, and have pretty decent amount of servers to support including HP and IBM. First i was supporting only HP-UX then i started working with AIX and if i could choose i would never go back to HP. The difference between AIX and HP-UX is simple, the guys in IBM had planned the OS very well, the design is very robust and well thought! Every command in AIX has it’s purpose, it’s there to do the job and do it good. AIX is by far the most advanced unix we have these days, and to get the advanced features you have, you will need ODM and all the other things that are tightly integrated in to the OS, comparing ODM to registry is plain stupid. You can do whatever you want with the ODM you have the cmd tools but you have to know what you are doing, becouse you can render the OS useless if you mess up. IBM is one step ahead of the competiton and that is proven by the new release of the Power CPU! Please do show me ONE single empty feature promise!! How is it with HP? I tell you, you have config files scatered in all sorts of places! HP are trying to implement all the stuff AIX has for ages, and they do it badly. Sometimes i have the feeling the HP-UX developers from different parts of the OS don’t comunicate to each other and you get what HP-UX is these days, the new 11.31 version is no difference! Do you know how you create a VG in HP-UX? I will tell you, it starts like this:
mkdir /dev/somevg
mknod /dev/somevg/group c 64 0x030000
nice isn’t it? Even the stupid linux doesen’t do that!!
In HP-UX online FS resizing is a feature that costs extra?!?! What server OS is one that can’t resize a stupid FS online? Mirroring the vg00? That costs extra too…In the new version they give those for Free (WOW that’s a real bonus you know). Did you know in AIX you can flash the firmware of the SCSI drives in the server ONLINE (they recommend doing it while you have less I/O to the drives) good luck doing that in HP-UX.
And about the seasoned administrator that can’t tell what command they run in smitty? Right like you do know all the switches in every command you use…that’s pretty silly don’t you think? Smitty is there to help you, i can’t say that for sure about SAM. The new version of sam in 11.31, called shm is a step in the right direction, time will tell.
I currently work with AIX5.3, maybe you work with AIX version 3?
And don’t get me started on the Backup tools AIX and HP-UX have, the guy that made Ignite-UX did a crime against the humanity.
Have you read a single HP document and then some AIX documents? I don’t think you have. Please do stop by http://www.redbooks.ibm.com and look around.
After you work sometime with Linux, HP-UX and AIX you get to appreciate what AIX offers you. Some peapole think that the harder it is for you to do the daily tasks you have, the more l33t you are. NEWSFLASH: YOU ARE NOT. You are just wasting your time with stuff that isn’t suppose to be that hard to do, instead of learning new technologies and working with them, the world is progressing and it’s moving forward not the other way around!
NOTE: i know this post sounds like i’m some kind of AIX fanatic but i’m not, i’m just regular unix admin thats all.
NOTE1: edited becouse of typos.
Edited 2007-06-01 14:12
In HP-UX online FS resizing is a feature that costs extra?!?! What server OS is one that can’t resize a stupid FS online? Mirroring the vg00? That costs extra too
I guess this is the downside of HP OEMing Veritas VM and FS. So far as I know, this attaches a fixed license fee to every CPU you license for HP-UX, and of course features cost extra to keep Veritas (Symantec) happy. I wonder if the free Veritas SF release helped HP to negotiate the free LVM features increase?
Not that I am saying that I think integrating the Veritas code was a bad idea–actually, I think it was a reasonable move and made a lot of sense in a time when open source was not especially on the radar. But I am pretty sure that Solaris is the only UNIX which was free enough of third party licensed code to be open sourced reasonably, and perhaps ‘staying independent’ was the better strategy.
On the AIX side, I am under the impression that the original volume management code came from OSF through Veritas and is thus also encumbered and requires a fixed fee to Veritas for every CPU (I think this was also the case with Tru64). Is this so?
I’m not familiar with the history of the volume managers. But from this link: http://www-03.ibm.com/servers/aix/products/aixos/whitepapers/lvm_ve…
I assume IBM developed the LVM by them self. Not 100% sure at all. The IBM FS (JFS2) is developed by IBM.
http://en.wikipedia.org/wiki/IBM_Journaled_File_System_2_(JFS2)
Thought the IBM JFS2 and HP-UX JFS are totaly different.
I’m not familiar with the history of the volume managers. But from this link: http://www-03.ibm.com/servers/aix/products/aixos/whitepapers/lvm_ve…..
I assume IBM developed the LVM by them self. Not 100% sure at all. The IBM FS (JFS2) is developed by IBM.
Well, I have seen those ‘translation’ docs for HP-UX LVM versus VxVM too, so I don’t know if that shows much. What I did find that is interesting is this old thread:
http://www.osnews.com/comment.php?news_id=13874&limit=no&threshold=…
In which one of the posters presents a somewhat convincing argument that the LVM in OSF was from IBM. Of course this is around 1990 and so printed sources are going to be the problem. At any rate, I do believe IBM has licensed Veritas IP, regardless of what the LVM looks like. But I am much more sure that this is the case with Tru64 (and of course HP-UX) than with AIX.
The IBM FS (JFS2) is developed by IBM.
http://en.wikipedia.org/wiki/IBM_Journaled_File_System_2_(JFS2)
Thought the IBM JFS2 and HP-UX JFS are totaly different.
Agree 100%. What I am not totally clear on is how close Linux JFS is to IBM JFS2.
“What I am not totally clear on is how close Linux JFS is to IBM JFS2.”
JFS2 was developed for OS/2 and ported to AIX and later opensourced and ported to Linux.
Sure, the winreg is way more intrusive but what got me hooked on unix in the first place was the whole “wow, just edit this file and thats it! awesome” Especially given the tools and its billions cmdline options to try to querey ODM its devices, its attributes etc.
ODM is unlike the registry in that it actually works 🙂
It seems text files in general are on their way out in modern unixes, at least for managing services. AIX has had the SRC for years, Solaris 10 has SMF with its xml repository that should only be edited through commands and OSX has launchd.
ONLY IF YOU HAVE A HANDFULL of AIX boxes is it “easy”(read gui) to admin.
Try taking the settings you point and clicked into your AIX system and automatically replicate those to many boxes and you will find the total flaw with AIX.
Having configs stored in a binary “registry” like AIX does totally destroys the ability to automate systems configuration and makes you HAVE TO login to each box and admin them BY HAND EVERY TIME….just like Windows.
It sucks…and is a unix for amatures in small shops with no vision of real configuration management.
Point and click? You using the WebSM to config the unix? Don’t want to be rude/sarcastic but there are cmd tools for that? And if you want to clone a box NIM is useful. You can define custom scripts that will do whatever you want after the bos_inst. There is updated draft document of using NIM (if i understand you correctly about the mass instalation of many boxes) really up to date. In ODM you store mainly device related configurations. As someone mentioned, there are text config files too. If you don’t understand how AIX works or how to work with it, don’t say it’s stupid and deny it. “HAVE TO login to each box” please do tell how you admin the unix you support without logging? And you can press “Esc+6” or F6 in most cases in smitty to see what command does the smitty run behind the scene. It’s funny how people deny anything that they don’t understand or are too lazy to read about. I don’t say AIX is the perfect unix, it has some bugs to iron out. But at the end of the day, when you have to support some OS it’s a matter of “Administrator Friendly”, so you can do your job and be done with it, and not have to remember or do a 2 or 3 page instruction on a simple mirror of vg00 (check out HP-UX procedure for doing that) when you can do it with 2 cmds.
Just my 2e cents.
I did some googling around on these interfaces, and didn’t see any provisions for asynchronous filesystem IO. I didn’t look too hard, but is this feature missing or am I just not seeing something?
If asynchronous file IO is not present, then even lowly Windows might beat out AIX for many kinds of server tasks.
Dude, I think you need to do a little more homework before starting your FUD spasm. AIX has had asynchronous IO for years…
http://inetsd01.boulder.ibm.com/doc_link/en_US/a_doc_lib/aixprggd/k…
Not trying to spread FUD. I honestly did not know the answer to my question, so I asked it. I shouldn’t have added the little snipe about Windows, but I was hoping that it would encourage someone (like butters who undoubtedly knows or has access to all the details) to respond with some details.
The link you sent me seems to suggest that the AIO is queued and dispatched by a layer higher than the file system the (AIO server). Is this the way it works? This is the reason I asked the question. The filesystem interface itself does not seem to contain the requisite functions to handle AIO directly.
I’m wondering why I was modded down, btw. My post was not off-topic or personally insulting. Maybe it was spam or an advertisement? Not sure.
Yes, that is the way I understand it. IO requests are queued by aioserver processes, and they perform the transaction on your behalf. They can handle requests for either file systems or raw logical volumes. You configure how many aioserver processes are running by manipulating attributes of the aio0 device with the chdev command.
There seems to be some good technical information in section 6.3.1 here:
http://www.redbooks.ibm.com/redbooks/pdfs/sg247463.pdf
I would think there’s a more in-depth treatment of the subject out there somewhere, but I can’t find it.
Yeah… When it comes to things like graphics, AIX is lacking. But when it comes to memory management and I/O, AIX is as sophisticated and advanced as any server OS, if not clearly the best. It runs on 128-core systems with TBs of RAM and SANs that push upwards of 10Gbps (maybe more, not sure). Not only does it have asynchronous I/O, but also multipath I/O and soon virtual I/O in AIX 6.
To be fair, there are server tasks at which Windows clearly bests AIX. Exchange, for example. 😉
I’m sure IBM really wants to run Microsoft Exchange on their systems, so that’s a huge design issue for them.
AIX may not be fancy on the outside but it works well enough on a text terminal and has some freakishly great throughput because of the fancy coding underneath it all.
You already have Virtual I/O with AIX 5.3 and Virtual I/O Server ! (check the IBM site)
How could virtual IO work through this filesystem interface? It seems like there’s no way to prioritize IO’s in the vn_rdwr call and it doesn’t make sense to me that higher layers would have the Virtual Cluster Number information needed to implement VIO?
Is there another interface to filesystems on AIX besides this one? It looks like this interface might be a second-class citizen API for compatibility. I’d like to be proven wrong, so don’t take my questions as an insult.
How could virtual IO work through this filesystem interface? It seems like there’s no way to prioritize IO’s in the vn_rdwr call and it doesn’t make sense to me that higher layers would have the Virtual Cluster Number information needed to implement VIO?
Is there another interface to filesystems on AIX besides this one? It looks like this interface might be a second-class citizen API for compatibility. I’d like to be proven wrong, so don’t take my questions as an insult.
The filesystem layer doesn’t care about how a storage device is attached! Or even how to access it!
IO is very complicated. There are more layers than just the filesystem. Briefly, underneath the filesystem is a lot of SCSI code. Then multiple device drivers (SAN via fibre channel, iSCSI, internal parallel SCSI disks, VIO attached storage, etc.).
As the other poster pointed out, there are several layers here:
LFS->VFS->FS->LVM->VSCSI<–>VIOS->LVM->SCSI-&g t;disk
Or something very close to that. It’s sophisticated enough that you’ll want to think of it as a black box that works like magic.
Sometimes in a highly-abstracted software system, you lose the ability to do smart things on a high level and expect them to remain smart at the bottom of the stack. For instance, filesystems that optimize for spatial locality of reference are easily thwarted by striping in the LVM. That carefully allocated contiguous area might actually be on two different disks on two different controllers.
That doesn’t mean that abstraction is bad. It means that abstraction should be done in a smarter way. For example, the above situation wouldn’t be an issue if the LVM exported a set of logical stripes instead of a flat block device. But then the filesystem wouldn’t be able to manage a raw block device without the LVM. Where you draw the line is often a tradeoff between performance and flexibility.
because they aren’t hip like Linux, OpenSolaris or FreeBSD.
They are purely server oriented operating systems and who the hell wants to develop anything sexy for servers?