to mention any decent self respectin admin better be able to tool through some C to fix a dumb compile error.
Shell perl C at least a decent understanding..
Make -- dunno how many times ive had to fix makefiles to point to the right shit on various non linux platforms building OSS on them.
Understanding the computer from ground up has helped me tremendously throughout my career. This means understanding how CPU work how memory works how software interacts with software..
I dunno how you can get off saying an admin doesnt need to know how to program.. an admin that cant automate is gonna be a tired and angry admin:P
i.e. I have blocks 1 2 3 4 with data.. and i take a snapshot.. When block 1 is updated in the real volume the snapshot retains block 1's initial data at time of snapshot. blocks 2 3 4 are pointers to the actual data on the real vol. *least this is my albiet limited understanding of snapshotting* The goal here is to preserve state of a volume without having to have a snapshot volume of equal size.
you can take snapshots as often as you want.. and can "roll back" to a previous state.
i use evms on 2 prod servers.. it is verry nice. Infact soo nice that i migrated my home machine from lvm-> evms without a hitch. the guys at IBM making EVMS are doin a great job. I see alot of ppl complaining about lack of LVM in linux and its sad cause bot LVS and EVMS are stable adn quite usable. theres great docs and howto's for both. Really everyone should be using a LVM for any of thier systems. Add so much more power and configurability.
Being "stable" on your laptop means nothing. The question is about high-end Unix servers, very heavy load, mission critical complex apps.
I have been using ReiserFS on may systems running large arrays ( with LVM ) ~700+ GB arrays with at least 20 disks for last 2 years. I have never had an issue with the FS or LVM in terms of dataloss. These systems are heavy file servers. I have tested xfs on similar setups more recently without a hitch. Reiser has the most polished grow/shrink utilities.. which i use a lot in conjunction with LVM to re-allocate space where needed. I recently (last 8 months) tried to grow a 200GB logical volume running ext3 adn it was a pain in the ass compared to doin the same thing with reiser.
I have been running XFS on a few production machines ( one firewall one linux load ballancer ) and most of my desktops for the past 10 months or so without a hitch.. YMMV
besides i think that coda, intermezzo, and gfs are some verry interesting projects that will eventually supplant nfs for native *nix filesharing.. ( yes gfs is linux only;)...
"PS: NOTE - I'm not going to merge either EVMS or LVM2 right now as things stand. I'm not using any kind of volume management personally, so I just don't have the background or inclination to walk through the patches and make that kind of decision. My non-scientific opinion is that it looks like the EVMS code is going to be merged, but.."
...which resulted in a lengthy debate on the virtues and pitfalls of each of these volume managers.
Also a previous/. article mentioned that Intel will be pushing EFI on thier next gen boards.. I had a chance to play with EFI a feew years ago whith some beta ia64 boxes. It's a 1000% improvement over bios and feels much more like the proms most comercial unix vendors have on thier systems. Lack of PROM is really a x86 issue more than a linux issue, but i agree that its a feature that i wish was there.
It will when these big players are pushing thier wares with Linux.. IBM/SGI/SUN/HP all are pushing Linux machines.. and all of them are putting back into Linux. So the many Eyes works bcs we got the comercial eyes + the OSS eyes..
Besides look at what MS is doing.. consistantly pushing thier users to thier server platform with dumed down features XP/XP pro 2k 2k server 2k as 2k datacenter or whatever. How is this different from the Linux comunity having things like lindows and gentoo ??
Lindows == really dumbed down linux. gentoo/slack == linux as it should be (IMHO):P
Lower maintenance costs. Much of hardware can be dynamically allocated and/or shared, either in real-time or at least without rebooting.
what about over time ? I can sycle new Ssytems into and out of my cluster and redesignate them to desktops etc for way more longevity than that 72 proc system will stay top of the line. 100% of the hardware in a well designed cluster can be dynamically "allocate" add/remobve nodes at hearts content without effecting the system.
Much better I/O throughput between different computational tasks. Potentially back-plane / bus speeds as opposed to network speeds (order of magnitude faster inter-process communication).
I/O is deffinatly one place where the sinle box wins out, but depending on what type of cluster your building these limits can be over come ( assuming your not useing std network tech developign propritary interconnect technology).
Potentially lower hardware costs (related to first item). This depends a lot though... but in case of, say, Sun servers, it makes sense to buy couple of more CPUs instead of more systems.
might want to rethink this point a lil deeper. sure buying another sun box is more exspensiove then buying a new proc, but the total cost of say a sunfire is pretty high. For the same cost how many comodity machines can you buy and cluster together ? *A LOT* more than 72:P
Easier maintenance / administration (related to first item). Depends on tools, but it is generally easier to maintain single "big" box than multiple "small" ones.
debatable with ssh pubkeys, rysnc, , system imager, and most the hardware staying the same. Maintaining 100 machines is not that hard.
Slashdot Mirror
damn shouldnt post so early :P
to mention any decent self respectin admin better be able to tool through some C to fix a dumb compile error.
:P
Shell perl C at least a decent understanding..
Make -- dunno how many times ive had to fix makefiles to point to the right shit on various non linux platforms building OSS on them.
Understanding the computer from ground up has helped me tremendously throughout my career. This means understanding how CPU work how memory works
how software interacts with software..
I dunno how you can get off saying an admin doesnt need to know how to program.. an admin that cant automate is gonna be a tired and angry admin
they owe you what they have extened in the _code_ of your work.
you can build whatever you like "on top of" any GPL'd software.
Gentoo + MoL == a verry happy G4 owner. ( most my shit is intel, but this allows me to use that g4 instead of thik it was a nice play toy ).
also for you 1984 readers out there check out .. Brave New World (if you haven't already )
this book had a massive effect on me when i read it when i was 14 or 15..
yes am am free to lead a private life.
or
i am free to keep my information private
or
i am free to protect myself
etc.. etc..
if you like ports you should check out getnoo
gentoo.org
basically a linux dist that has implementened and extended the idea of ports.
#!/bin/sh
..
industrial strenght gui.. oh wait.. what does a unix server need a gui for ? none of my servers have X installed. ( X libs maybe, but not X )..
um why not just run the versions that are stable ???
how is this any different from a comercial vendor ?
like every patch that sun released never de-stabalized the system or its apps oracle
Typically a snapshot is a diff of changes..
:P
i.e. I have blocks 1 2 3 4 with data.. and i take a snapshot.. When block 1 is updated in the real volume the snapshot retains block 1's initial data at time of snapshot. blocks 2 3 4 are pointers to the actual data on the real vol. *least this is my albiet limited understanding of snapshotting* The goal here is to preserve state of a volume without having to have a snapshot volume of equal size.
you can take snapshots as often as you want.. and can "roll back" to a previous state.
in my expierience this is a good thing
i use evms on 2 prod servers.. it is verry nice. Infact soo nice that i migrated my home machine from lvm-> evms without a hitch. the guys at IBM making EVMS are doin a great job. I see alot of ppl complaining about lack of LVM in linux and its sad cause bot LVS and EVMS are stable adn quite usable. theres great docs and howto's for both. Really everyone should be using a LVM for any of thier systems. Add so much more power and configurability.
startx -- :1 :2 ....
startx --
etc etc..
Being "stable" on your laptop means nothing. The question is about high-end Unix servers, very heavy load, mission critical complex apps.
I have been using ReiserFS on may systems running large arrays ( with LVM ) ~700+ GB arrays with at least 20 disks for last 2 years. I have never had an issue with the FS or LVM in terms of dataloss. These systems are heavy file servers. I have tested xfs on similar setups more recently without a hitch. Reiser has the most polished grow/shrink utilities.. which i use a lot in conjunction with LVM to re-allocate space where needed. I recently (last 8 months) tried to grow a 200GB logical volume running ext3 adn it was a pain in the ass compared to doin the same thing with reiser.
I have been running XFS on a few production machines ( one firewall one linux load ballancer ) and most of my desktops for the past 10 months or so without a hitch.. YMMV
http://llg.cubic.org/tools/
add/remove scsi devices from linux on the fly.
NFS on Linux sux. Thats really the issue..
;)...
besides i think that coda, intermezzo, and gfs are some verry interesting projects that will eventually supplant nfs for native *nix filesharing.. ( yes gfs is linux only
from kernel trap on 2.5 and volume management:
.."
...which resulted in a lengthy debate on the virtues and pitfalls of each of these volume managers.
At the end of his message, Linus commented:
"PS: NOTE - I'm not going to merge either EVMS or LVM2 right now as things stand. I'm not using any kind of volume management personally, so I just don't have the background or inclination to walk through the patches and make that kind of decision. My non-scientific opinion is that it looks like the EVMS code is going to be merged, but
If i recall gnu mount supports partition labels. so you can ignore drive designation I can't recall the syntax atm..
isn't this also somthing thats being addressed with devFS ( or whatever its gonna be in 2.6)
Also a previous /. article mentioned that Intel will be pushing EFI on thier next gen boards.. I had a chance to play with EFI a feew years ago whith some beta ia64 boxes. It's a 1000% improvement over bios and feels much more like the proms most comercial unix vendors have on thier systems. Lack of PROM is really a x86 issue more than a linux issue, but i agree that its a feature that i wish was there.
It will when these big players are pushing thier wares with Linux.. IBM/SGI/SUN/HP all are pushing Linux machines.. and all of them are putting back into Linux. So the many Eyes works bcs we got the comercial eyes + the OSS eyes..
ya in the meantime the system was frozen.. ever had someone pull out a console cable and have it send a break to a e4500 ? eeek.. to the DB ;P
have a look at:
- system imager
- OSCAR
- System Instalation Suite:>
all of these are better replacements for jumpstart. again IMHOBesides look at what MS is doing.. consistantly pushing thier users to thier server platform with dumed down features XP/XP pro 2k 2k server 2k as 2k datacenter or whatever. How is this different from the Linux comunity having things like lindows and gentoo ??
:P
Lindows == really dumbed down linux.
gentoo/slack == linux as it should be (IMHO)
dont scale up scale OUT!
Lower maintenance costs. Much of hardware can be dynamically allocated and/or shared, either in real-time or at least without rebooting.
:P
what about over time ? I can sycle new Ssytems into and out of my cluster and redesignate them to desktops etc for way more longevity than that 72 proc system will stay top of the line. 100% of the hardware in a well designed cluster can be dynamically "allocate" add/remobve nodes at hearts content without effecting the system.
Much better I/O throughput between different computational tasks. Potentially back-plane / bus speeds as opposed to network speeds (order of magnitude faster inter-process communication).
I/O is deffinatly one place where the sinle box wins out, but depending on what type of cluster your building these limits can be over come ( assuming your not useing std network tech developign propritary interconnect technology).
Potentially lower hardware costs (related to first item). This depends a lot though... but in case of, say, Sun servers, it makes sense to buy couple of more CPUs instead of more systems.
might want to rethink this point a lil deeper. sure buying another sun box is more exspensiove then buying a new proc, but the total cost of say a sunfire is pretty high. For the same cost how many comodity machines can you buy and cluster together ? *A LOT* more than 72
Easier maintenance / administration (related to first item). Depends on tools, but it is generally easier to maintain single "big" box than multiple "small" ones.
debatable with ssh pubkeys, rysnc, , system imager, and most the hardware staying the same. Maintaining 100 machines is not that hard.
am i nutz isn't this part of what ACPI is for ?