Dude, when Itanium came out, it was the fastest FP core on the planet. It came out how many years late?
Had it come out on time, it would have changed processor design. There was a large debate about whether to go superscaler or the itanium route. Superscaler one this time, but it is hard in a different way, and we've already reached some of its limits (hence the move to multiple cores), and it is a lot harder to write software to use superscaler and multicore designs than an itanium approach. Itanium required really good compilers, which it would have had, had it come out on time.
You are calling Intel stupid, but saying that they should try to move in on IBM's turf? I just hope that they'd be smart enough to implement an atomic check and set instruction so that one could do a spin-lock without disabling interupts. Anyways, RISC is quite a bit less efficient than CISC with memory bandwidth, and Intel doesn't have the greatest history with making memory fast and wide. CISC does have some definite advantages outside of the core, which is why it has worked so well. x86 is just a particuarly weird implementation.
You have some very valid points, but I think you are selling Intel and the Itanium short. Itanium isn't really VLIW either, it's weirder than that, though VLIW is the easiest way to describe it. It was designed to solve a lot of the problems with VLIW. My opinion is that they should have first shipped it without any of the x86 stuff as a pure supercomputing processor, and write a solid C and fortan compilers for it. After that, they could add x86 emulation to break into the server market. I can understand why they wanted to go all in though.
While it may have failed, their first superscaler chip failed horribly as well. They'll be pillaging the Itanium IP for decades to come.
My uid is artificially high from the Great Accidental Account Deletions of 199something.
I think that there is probably a hardware problem with the box, but the company will not take it seriously. I really don't have anything good to say about them except that they are cheap.
I'm really sure sure how you can call the transfer speeds exceptional.
These are with Jumbo frames enabled. I honestly can't remember if the infrant numbers were done directly connected or through a switch.
Xserve to Xserve via AFP 254.6 Xserve to NAS via NFS 44.48 Xserve to NAS via AFP 50.23 Xserve to NAS via SMB 49.89 (all speeds in Mb/s)
OSX is pretty bad at filling pipes, but I don't think it can be blamed here. We've seen similar speeds with the Linux box backing up to it.
All RAIDs loose drives? You must be using cruddy hardware as we have not seen this elsewhere. We have a number of vendors, and none of them lose drives. I'm not sure that Infrant does hardware RAID, they mention "Hardware Accelerated RAID", but the logs point to md. We've retired a number of Dells into lesser service using either md or geom, and they work flawlessly. For 3 or 4 drives, software RAID is in a lot of ways superior to HW RAID. Anyways, the Infrant won't let you rebuild without a reboot, which is silly. Their AFP support is just netatalk, same as almost anyone else.
I poo-pooed the cgi because it doesn't work. Same thing with their detection software.
I think the general consensus is that NetApps are the best NAS out there. An ultrasparc would be better than the infrant. Businesses want stuff that work and fail gracefully. The infrant is neither
Infrant stuff sucks ass. We bought one at work to try it out, and it is horrid. We reboot it once a week because it will lock up. Transfer speed is horrible. NFS support is horrible.
Infrant tech support has flaked us off, and claims that a drive is bad, even though we can find no evidense of it. It will "lose" drives (usually 2 or 4, sometimes 1) for no reason, which will be fixed by a reboot.
And, under the hood it's a cheap SPARC running hacked up linux and md. You'd be far better off building a whitebox running md and lvm instead of dealing with the infrant crap. They keep promising to give ssh access, but I'm thinking that I should just hack through their shitty security instead. Picking linux without using LVM to its potential was stupid anyways. They obviously are not a fan of openness, so they'd have been better served by FreeBSD-Stable running gvinum. I'm really not sure how they got NFS to run so slowly. It's almost as bad as their CGI coding skills.
Makes sense. I don't have a TV, but I'm thinking about getting a media server, so I forget about some of the "normal" features. This still doesn't discount the mini or a hacked aTV with an external converter, but that is less than elegant.
I concluded that there aren't any products out there that do what I want, so I'm not going to buy any, which is fine by me.
Another route that you could take would be to get either a mini, or a slim linux box (if VLC's quality doesn't bother you) as your set top box, and then have a server in the closet with a TV card serving it out via NFS. Also less than elegant, but it at least gives some advantages as a trade.
When I do decide to get one, I'll probably build it myself, since all the ones I've seen are missing half of what I want and have a whole lot of extra crap.
At what resolution? 1080 might still be a problem. Last time I checked, non apple h264 blew pretty hard (I'm looking at you, VLC (and others)), so why not get get a mini or an apple TV?
Translation is hard, really hard. Especially when the languages aren't very similar. Any translation will have a strong bias of the translator. So, while the basics may not change, the inferred meanings certainly have.
The KJV was rewritten to be more poetic, and isn't a very good 'literal' translation.
A lot of verses have been added or changed to suite the times.
That being said, I generally like the bible, I just think that it is used in foolish ways. It's an imprecise work of man from which we can gain some wisdom. Or we could use it to act stupid.
But, the books is the major thing. Is Luther a prophet? How about the church who picked the other books? At best it's, "The word of GOD (edited by man)
Last time I looked, there was one for JACK. I used Ardour a bit with my Delta1010LT and it worked fine. Jack can output to Core Audio, so most things should work. (Except maybe protools crap)
You should do what you think is right. "just doing my job" isn't an exuse. If your job is incompatible with your beliefs, you should quite (i.e. pharmacist who won't hand out birth control) but that does not seem to be the case here. I doubt they'd fire you for putting your foot down, and if so, you'd win in court.
I agree what NIS and NFS in osx suck. I do understand why they picked LDAP over NIS though, and given their track record, I really doubt that they'll fix it. There really is not good reason for NFS to suck so bad. (though, NFS on linux sucked until recently (and NIS on Linux still has some problems last time I heard))
OS X works fine with/etc/fstab, and it has init scripts, they're just weirder. Why not hang out at 10.4.6? OS X doesn't seem ready for your enterprise environment, but that doesn't mean that it isn't ready for all or even most. I might be wrong, but I'd guess that Kerberos has a lot more penetration than NIS.
If you had Kerberos, things aren't great, but they start to look a lot better. You could try to get the macs to auth against samba. It isn't pretty, but I've seen it work. You could also install netatalk on the linux boxes and serve out afp, which works pretty well. You could also either build or buy a NIS->LDAP gateway to keep the Macs happy. While annoying, it's similar to what you did for Windows, just new and different.
Multicast asr dude. It's the shiznit. Use it for any mass installs and then netinstall or a FW HD + radmind for the rest and the Win admins are going to cry.
If you really really really love packages, use ARD and be a twit, but at that size, you really need radmind. If you need per seat licenses, either write some code to do it in postscripts or write code to automate the builds. Really though, you should be able to negotiate with the vast majority of your vendors into being able to use KeyServer. Radmind is many, many times better than SMS. Even ARD+OpenLDAP is right up there with SMS if done right.
Why don't you use the group policy in LDAP? You can push settings to (almost?) every plist on the system. 60 minutes is an insane refresh rate. 2 weeks is more reasonable.
You guys aren't looking for an OS X admin, are you?
They are pretty good about supporting OSX. By pretty good I mean that their products are just as shitty here, but they bought all of the competition, so we deal.
Sure. You should be able to stick everything inside the DMG. If you really plan ahead, you should be able to do it with only one image. That way, launchd just needs to call asr and reboot afterwords.
1: Create a diskless netboot that has disk utility and terminal 2: Prep both your final images (including radmind for the OSX one) for ASR multicast 3: Start both asr multicast jobs
Then on each computer, boot into the netboot image, reformat and then drop down the images with asr. If you're slick, this can all be scripted into launchd.
Currently no, we'll probably put it in this summer. We aren't large enough to need it from a security standpoint and we've been debating how to organize it. (for us it is more of a matter of how to separate different types of accounts)
One way to accomplish this (there are certainly others, but this comes to mind and is similar to what we're going to do, but we have different goals) is to have uid=jblow,dc=dept,dc=users,etc. This way, acls could be put on at the dc=dept level. The bind accounts to authenticate users could either look to users, or the department depending on what you needed. (or you could use groups and similarly ACL them)
We've never had any trouble with replicas, but we don't have any WAN links. I don't know if OpenLDAP can do what you describe.
You should poke at Directory Access. My guess is that however eDirectory organizes data can be explained to Directory Access. OpenLDAP and DA are quite extensible tools. I imagine that you can bend them to fit your organizational structure.
We did it how we did it for two reasons, a general distrust of Apple, and the amount of customization we wanted to put in the tree. Apples inability to inform the world about release dates and features makes us uncomftorable having our infrastructure run on XServes. And we don't like paying the Apple tax when we don't need to. Things like DNS, NTP, LDAP and mail are run on linux (mostly Dell. I wish I knew another company to buy servers from. I'm really thinking about Sun actually. I'm sick of never knowing what you're really getting with Dell)
We've hacked up users to fit our organization, but groups are left fairly untouched. We keep track of a lot of nonstock data in out LDAP tree. WGM sees the stuff, but doesn't mess with it. I think, in the end, we looked at if more as what does running it on OSX get us than the other way around. We're a heterogeneous shop though, had we been all OSX, we'd have run it on OSX.
I've never worked with eDirectory, but can't ACLs do what you're describing? We have a tools set up that require people be members of a certain group in order to run by having ACLs on certain attributes and sections of the tree. What more does eDirectory do? (It'd probably be easier on eDirectory though)
Come on, be friendly. And please, don't tell me what I'm doing, I have a pretty good clue how my network is set up.
We set up OD initially to steal its schema and then modified it some. It isn't that hard and there are docs out there. The server talks to the ldap server, not WGM. LDAP is configured in Directory Access (both primary and backup). WGM can connect to any server or workstation, and then to the ldap through that.
We have hundreds of computers doing it. WGM just calls the ldap password change. It views them as crypt passwords, but they are ssha1 in ldap. Everything works fine. We don't use kerberos, but that'd probably work just as well. We have a legacy system for managing printers, but we do the rest in WGM (part of why I keep it around). Likewise, groups are easier to manage with dscl and/or WGM.
You might want to look at dscl. WGM is nice for affecting one user, but dscl is about the closest thing to scripting WGM.
A lot of things are possible when you don't preclude them. The built in tools are just a start, grab vim and a scripting language of your choice. I've been using ruby lately, our netadmin likes python; both work well. That way your tools will fit your tree, which is much more powerful than WGM, especially with cron.
Slashdot Mirror
Dude, when Itanium came out, it was the fastest FP core on the planet. It came out how many years late?
Had it come out on time, it would have changed processor design. There was a large debate about whether to go superscaler or the itanium route. Superscaler one this time, but it is hard in a different way, and we've already reached some of its limits (hence the move to multiple cores), and it is a lot harder to write software to use superscaler and multicore designs than an itanium approach. Itanium required really good compilers, which it would have had, had it come out on time.
You are calling Intel stupid, but saying that they should try to move in on IBM's turf? I just hope that they'd be smart enough to implement an atomic check and set instruction so that one could do a spin-lock without disabling interupts. Anyways, RISC is quite a bit less efficient than CISC with memory bandwidth, and Intel doesn't have the greatest history with making memory fast and wide. CISC does have some definite advantages outside of the core, which is why it has worked so well. x86 is just a particuarly weird implementation.
You have some very valid points, but I think you are selling Intel and the Itanium short. Itanium isn't really VLIW either, it's weirder than that, though VLIW is the easiest way to describe it. It was designed to solve a lot of the problems with VLIW. My opinion is that they should have first shipped it without any of the x86 stuff as a pure supercomputing processor, and write a solid C and fortan compilers for it. After that, they could add x86 emulation to break into the server market. I can understand why they wanted to go all in though.
While it may have failed, their first superscaler chip failed horribly as well. They'll be pillaging the Itanium IP for decades to come.
> DO practice your English writing skills, this will help you in your career too.
DO NOT take this as an insult at your nationality. The parents advice is universal; native speakers should heed this advice as well.
That's pretty damn cool.
It runs fine on my first gen G5. I think something may be wrong with your setup.
My uid is artificially high from the Great Accidental Account Deletions of 199something.
I think that there is probably a hardware problem with the box, but the company will not take it seriously. I really don't have anything good to say about them except that they are cheap.
I'm really sure sure how you can call the transfer speeds exceptional.
These are with Jumbo frames enabled. I honestly can't remember if the infrant numbers were done directly connected or through a switch.
Xserve to Xserve via AFP 254.6
Xserve to NAS via NFS 44.48
Xserve to NAS via AFP 50.23
Xserve to NAS via SMB 49.89
(all speeds in Mb/s)
OSX is pretty bad at filling pipes, but I don't think it can be blamed here. We've seen similar speeds with the Linux box backing up to it.
All RAIDs loose drives? You must be using cruddy hardware as we have not seen this elsewhere. We have a number of vendors, and none of them lose drives. I'm not sure that Infrant does hardware RAID, they mention "Hardware Accelerated RAID", but the logs point to md. We've retired a number of Dells into lesser service using either md or geom, and they work flawlessly. For 3 or 4 drives, software RAID is in a lot of ways superior to HW RAID. Anyways, the Infrant won't let you rebuild without a reboot, which is silly. Their AFP support is just netatalk, same as almost anyone else.
I poo-pooed the cgi because it doesn't work. Same thing with their detection software.
I think the general consensus is that NetApps are the best NAS out there. An ultrasparc would be better than the infrant. Businesses want stuff that work and fail gracefully. The infrant is neither
Infrant stuff sucks ass. We bought one at work to try it out, and it is horrid. We reboot it once a week because it will lock up. Transfer speed is horrible. NFS support is horrible.
Infrant tech support has flaked us off, and claims that a drive is bad, even though we can find no evidense of it. It will "lose" drives (usually 2 or 4, sometimes 1) for no reason, which will be fixed by a reboot.
And, under the hood it's a cheap SPARC running hacked up linux and md. You'd be far better off building a whitebox running md and lvm instead of dealing with the infrant crap. They keep promising to give ssh access, but I'm thinking that I should just hack through their shitty security instead. Picking linux without using LVM to its potential was stupid anyways. They obviously are not a fan of openness, so they'd have been better served by FreeBSD-Stable running gvinum. I'm really not sure how they got NFS to run so slowly. It's almost as bad as their CGI coding skills.
Makes sense. I don't have a TV, but I'm thinking about getting a media server, so I forget about some of the "normal" features. This still doesn't discount the mini or a hacked aTV with an external converter, but that is less than elegant.
I concluded that there aren't any products out there that do what I want, so I'm not going to buy any, which is fine by me.
Another route that you could take would be to get either a mini, or a slim linux box (if VLC's quality doesn't bother you) as your set top box, and then have a server in the closet with a TV card serving it out via NFS. Also less than elegant, but it at least gives some advantages as a trade.
When I do decide to get one, I'll probably build it myself, since all the ones I've seen are missing half of what I want and have a whole lot of extra crap.
I was referring to the visual quality, not the processor usage.
At what resolution? 1080 might still be a problem. Last time I checked, non apple h264 blew pretty hard (I'm looking at you, VLC (and others)), so why not get get a mini or an apple TV?
Translation is hard, really hard. Especially when the languages aren't very similar. Any translation will have a strong bias of the translator. So, while the basics may not change, the inferred meanings certainly have.
A ramaic-Words/dp/0060619953/ref=pd_bbs_sr_1/002-015 3266-8695263?ie=UTF8&s=books&qid=1179440604&sr=8-1
The KJV was rewritten to be more poetic, and isn't a very good 'literal' translation.
A lot of verses have been added or changed to suite the times.
Wikipedia is a pretty good place to start, and the straight dope had a good article on the history of the bible. If you want something weirder, check out: http://www.amazon.com/Prayers-Cosmos-Meditations-
That being said, I generally like the bible, I just think that it is used in foolish ways. It's an imprecise work of man from which we can gain some wisdom. Or we could use it to act stupid.
But, the books is the major thing. Is Luther a prophet? How about the church who picked the other books? At best it's, "The word of GOD (edited by man)
The books that comprise the bible have changed a number of times.
Last time I looked, there was one for JACK. I used Ardour a bit with my Delta1010LT and it worked fine. Jack can output to Core Audio, so most things should work. (Except maybe protools crap)
You should do what you think is right. "just doing my job" isn't an exuse. If your job is incompatible with your beliefs, you should quite (i.e. pharmacist who won't hand out birth control) but that does not seem to be the case here. I doubt they'd fire you for putting your foot down, and if so, you'd win in court.
Productively, you might want to look at http://sassafras.com/.
I agree what NIS and NFS in osx suck. I do understand why they picked LDAP over NIS though, and given their track record, I really doubt that they'll fix it. There really is not good reason for NFS to suck so bad. (though, NFS on linux sucked until recently (and NIS on Linux still has some problems last time I heard))
/etc/fstab, and it has init scripts, they're just weirder. Why not hang out at 10.4.6? OS X doesn't seem ready for your enterprise environment, but that doesn't mean that it isn't ready for all or even most. I might be wrong, but I'd guess that Kerberos has a lot more penetration than NIS.
OS X works fine with
If you had Kerberos, things aren't great, but they start to look a lot better. You could try to get the macs to auth against samba. It isn't pretty, but I've seen it work. You could also install netatalk on the linux boxes and serve out afp, which works pretty well. You could also either build or buy a NIS->LDAP gateway to keep the Macs happy. While annoying, it's similar to what you did for Windows, just new and different.
Multicast asr dude. It's the shiznit. Use it for any mass installs and then netinstall or a FW HD + radmind for the rest and the Win admins are going to cry.
Ugh, bloody ACs.
Use. Radmind. Now.
If you really really really love packages, use ARD and be a twit, but at that size, you really need radmind. If you need per seat licenses, either write some code to do it in postscripts or write code to automate the builds. Really though, you should be able to negotiate with the vast majority of your vendors into being able to use KeyServer. Radmind is many, many times better than SMS. Even ARD+OpenLDAP is right up there with SMS if done right.
Why don't you use the group policy in LDAP? You can push settings to (almost?) every plist on the system. 60 minutes is an insane refresh rate. 2 weeks is more reasonable.
You guys aren't looking for an OS X admin, are you?
We do it with Open LDAP on linux. I've heard of other places doing it with AD. All you need is the right schemas. LDAP is LDAP.
At least they aren't hypocritical.
They are pretty good about supporting OSX. By pretty good I mean that their products are just as shitty here, but they bought all of the competition, so we deal.
Sure. You should be able to stick everything inside the DMG. If you really plan ahead, you should be able to do it with only one image. That way, launchd just needs to call asr and reboot afterwords.
1: Create a diskless netboot that has disk utility and terminal
2: Prep both your final images (including radmind for the OSX one) for ASR multicast
3: Start both asr multicast jobs
Then on each computer, boot into the netboot image, reformat and then drop down the images with asr. If you're slick, this can all be scripted into launchd.
Currently no, we'll probably put it in this summer. We aren't large enough to need it from a security standpoint and we've been debating how to organize it. (for us it is more of a matter of how to separate different types of accounts)
One way to accomplish this (there are certainly others, but this comes to mind and is similar to what we're going to do, but we have different goals) is to have uid=jblow,dc=dept,dc=users,etc. This way, acls could be put on at the dc=dept level. The bind accounts to authenticate users could either look to users, or the department depending on what you needed. (or you could use groups and similarly ACL them)
We've never had any trouble with replicas, but we don't have any WAN links. I don't know if OpenLDAP can do what you describe.
You should poke at Directory Access. My guess is that however eDirectory organizes data can be explained to Directory Access. OpenLDAP and DA are quite extensible tools. I imagine that you can bend them to fit your organizational structure.
We did it how we did it for two reasons, a general distrust of Apple, and the amount of customization we wanted to put in the tree. Apples inability to inform the world about release dates and features makes us uncomftorable having our infrastructure run on XServes. And we don't like paying the Apple tax when we don't need to. Things like DNS, NTP, LDAP and mail are run on linux (mostly Dell. I wish I knew another company to buy servers from. I'm really thinking about Sun actually. I'm sick of never knowing what you're really getting with Dell)
We've hacked up users to fit our organization, but groups are left fairly untouched. We keep track of a lot of nonstock data in out LDAP tree. WGM sees the stuff, but doesn't mess with it. I think, in the end, we looked at if more as what does running it on OSX get us than the other way around. We're a heterogeneous shop though, had we been all OSX, we'd have run it on OSX.
I've never worked with eDirectory, but can't ACLs do what you're describing? We have a tools set up that require people be members of a certain group in order to run by having ACLs on certain attributes and sections of the tree. What more does eDirectory do? (It'd probably be easier on eDirectory though)
Come on, be friendly. And please, don't tell me what I'm doing, I have a pretty good clue how my network is set up.
We set up OD initially to steal its schema and then modified it some. It isn't that hard and there are docs out there. The server talks to the ldap server, not WGM. LDAP is configured in Directory Access (both primary and backup). WGM can connect to any server or workstation, and then to the ldap through that.
We have hundreds of computers doing it.
WGM just calls the ldap password change. It views them as crypt passwords, but they are ssha1 in ldap. Everything works fine. We don't use kerberos, but that'd probably work just as well. We have a legacy system for managing printers, but we do the rest in WGM (part of why I keep it around). Likewise, groups are easier to manage with dscl and/or WGM.
You might want to look at dscl. WGM is nice for affecting one user, but dscl is about the closest thing to scripting WGM.
A lot of things are possible when you don't preclude them. The built in tools are just a start, grab vim and a scripting language of your choice. I've been using ruby lately, our netadmin likes python; both work well. That way your tools will fit your tree, which is much more powerful than WGM, especially with cron.