Thats very cool to hear that its stompin NT left and right, but I'm suprised that they wouldn't be porting to OpenBSD or FreeBSD first due to networking speed and security issues. Linux is neat and all, but I would still rather have a Solaris box on my front lines than a Linux box for a FW. Right now I don't do either, I've got a FreeBSD box handling 5 networks, all running at 100Mbps at 50-60% capacity (network bandwidth) so I'm pretty happy. I wasn't able to accomplish that with Linux on the same hardware.
To tell you whats truly bizzare, is that Nokia's FW-1 Appliance is based on some kind of BSD, so porting the actual FW to Free/Net/OpenBSD would be a snap. Hopefully they'll see the light and port soon.
Regardless, porting the CP GUI to Linux is great because its so well emulated on other platforms. Solaris, Open/Net/FreeBSD all can run Linux X86 binaries. I'd still rather have a native binary, but any binary is better than none.
Just about every other OS I know of (except for NT) includes this. Having a Kernel Debugger, Kernel Core Dump, and a few other tools available over the past few *YEARS* has saved me a lot of hassle. If Linux hasn't had this till now, I'm sooooooooooo sorry. Thats really dissapointing. *BSD, Solaris, Dynix, and bazillions of other OS'es have had this ever since they were created.
This is a really neat thing to find, especially since so many people are putting hard work into the desktop environment. So far Linux and NetBSD have put quite a bit of effort into this arena. NetBSD was one of the first FreeNixes to impliment USB support (that was usable) and they're working on quite a few other things. NetBSD recently ported some of the Linux DVD extensions, and we'll probably see that wrapped into the other BSD's sometime soon.
Re:*BSD's vs. Linux for Clusters?
on
Which BSD?
·
· Score: 2
Unfortunately it seems that clustering support has been lacking for *BSD so far. I've seen a few attempts, but nothing really good that I've seen so far. It would be nice if someone ported Beowulf over, because that would give researchers an easy way to migrate over, or use in parallel, or whatever with existing clusters.
Yes, *BSD, and FreeBSD in particular have pretty advanced TCP/IP stacks, so far they've been the only ones capable of driving the Myrinet cards past the gigabit range.
Unfortunately the 2GB file size is still there in 2.2.12. FreeBSD is able to handle files this size and larger on the X86 platform, as well as Net and OpenBSD. Solaris handles it without a hitch as well. What really bugs the hell out of me is the swap size limit with Linux, not being able to use 2GB of swap (or larger) is a pain in the ass. Granted it sucks to be even 500MB into swap, it happens sometimes.
Wonder what they're up to this time. They're running their site offa *single* FreeBSD 3.x box so chances are they're going to be Open-Source OS oriented. Or, they're a Wine Auction house:)
Selling a secure webserver and calling it a turnkey package is a load of BS. If that were the solution, life would be easy. There are already buttloads of SSL capable OSS webservers out there. Hell, one of the best secure webservers out there (stronghold) is basically Apache.
The company I work for now has Doug MacEachern of mod_perl fame:) And I've seen Cliff Skolnick and Brian Behlendorf still around the bay area, so they don't have all of them yet. IBM has dumped a *lot* of money into Apache, and into creating ties with companies that will use their HW/SW platform with Apache. They did help sponsor ApacheCon and had a large presentation there.
I've felt the same way, the FreeBSD installer is just plain out *EASY*. Few have come close enough to duplicate its ease of use. Actually, the BSD/OS 4.0 installer is just a whizbang deal. Super easy to install (mostly due to limited choices:) and its just plain out fast as all getout. Might be from using PAX instead of tar/cpio/etc.
They toed the PWC company line, said that only NT could deliver 100% uptime with some type of mirroring, and tried to downplay HP-UX, AIX, Solaris, and MVS, since they could only claim 99.995% availability.
Thats pretty a pretty interesting figure, especially since MS can't even do it. I've checked out quite a few versions of 'Wolfpack' and it fails to do even 99% uptime in most situations. NT also requres *4* machines do get there, and if you're running SQL7, you're SOL because it doesn't support it yet, *AND* its not instantly available either. That DB has to startup, check through possibly corrupted data, and then become available. That could take up to half an hour for a large database... It was very interesting explaining this to the manager of the DB department where I used to work at MSN. He said it was all BS, but gee, when he released his pet project, it flopped HARD and didn't even work because the DB response was so bad. (Maybe a whole slashdot article on its own) Ohwell, too bad for him:) I quit about 3 weeks before.
I think part of the problem that you'll be encountering is people not wanting to put something weird on X86 hardware. I've had very tough times telling people that yes, free operating systems can provide really great service. One thing I've seen is that companies are not as afraid of SCO or BSD/OS as they are of Linux, might have something to do with the longstanding corporate backing, but X86 OS'es other than NT are tough to push. If you are at a point where BSD/OS is acceptable, you might suggest FreeBSD or even OpenBSD to get in the door. You can make the claims that since they're based on 4.4BSD (lite-2/etc), that you have a similarly stable code base, remind them of the last time they saw big bad things about BSD/OS, FreeBSD or even OpenBSD. You can remind them that Wells Fargo, the US Government, Hotmail, Yahoo, CDROM.COM, and countless others use BSD based products. Its not perfect, but its a start. Getting free software in the door is the biggest step, once you accomplish that, you're almost assured a success.
Its not really the amount of CPU's thats the only factor. Its how much ram it can handle, its how many processes it can handle, its how fast its file system is. NT Beats Linux at this because NT gets a bit sloppy to get its speed, rather than with superior coding. FreeBSD beats Linux in Memory and Filesystem handling, however NT is still reigning in terms of speed on the X86 architechture because of the way its designed - to allow one process to monpolize all CPU time, whereas Un*x doesn't want you to do that.
Its like nobody on Slashdot cares about meaningful things. A 10k donation to OBSD is like a 100k donation towards Linux. Linux already has the name recognition. If you say 'Linux' to the average IT guy, they'll sit there and look like they're stoned for a lil bit. OpenBSD just gets "Huh? Is that like a Bondage thing?" I hope that people who see this realize what that donation is going to do. Its personally making me very partial to the IDS system that the donator sells, and gosh, it seems to run *GREAT* on OpenBSD why not use that... And yes, I need to do Gig-E IDS because I damn well feel like it!
Chances are they have had more than one NC at NCI. I wouldn't be suprised if they had StrongARM based units, but I believe the inital run was done with FreeBSD. When I was first in the know about this, it was late 1997/1998, so things may have changed quite a bit. Its quite likely that NCI switched to NetBSD to switch architechtures easily.
CacheFlow's product is similar to NetApp's NetCache product, however the CacheFlow seems to be eating the NetCache alive in terms of hits per second and simultaneous connections. I would reccomend Squid for a 'home' or office installation because its cheap for client use. For Acceleration you'll definately want a CacheFlow. I setup an inverse-web proxy for one of the sites I worked on, and squid couldn't handle the hits per second, it had 0 problem with massive amounts of connections being open which was ok, but not my primary objective.
Choosing Oracle is probably the way to go, mostly because if its what you're already using, you should stick with it. However one problem you might run into if you run it with Linux is the memory data structure size, and the file size. FreeBSD 3.X allows you to have larger than 2GB files around, so does Solaris, and Net/OpenBSD.
There has been some interesting success in putting Oracle 8i on FreeBSD, it seems to run faster and a bit smoother under FreeBSD, but I can't confirm that personally because I haven't put it into production. I am going to be installing it today on my workstation to give it a shot.
To be honest though, you should really be rolling your databases over before they hit filesize limits if you're only growing by 40GB per year. I would invest well in FCAL raid and see what gives you the best performance. Linux's FCAL support sucks so far, we had to discontinue using it and switched for FreeBSD on one installation I was working on. Just a couple of bumps
They are using FreeBSD, I have it from a reliable source inside of NCI. We've talked about this at the BAFUG (Bay Area FreeBSD Users Group) Meetings quite a bit.
This article doesn't suprise me at all. Oracle has been pushing this way for a while. They started pushing that way with NCI (They use FreeBSD) making an NC that runs under Linux wouldn't be that difficult. What Ellison really needs is the clueless IT manager going "oh! Linux Network computers! great *peanutbrainwhirring* linux == good *drool*"
Its what most IT managers do now with NT, soon it'll be Linux, after that who knows what.
Point being is that Oracle isn't stupid, they've already had a lot of the NC work done by NCI, now they just have to port the stuff over. They already have Oracle 8 ported to Linux, chances are they have their 'embedded' Oracle 8 ported over as well. Spending a few hundred man hours to port all their crap onto Linux based NC's is a no brainer. Its free money if it takes off.
Linux needs to take a *big* point from the BSD's on updating. FreeBSD, OpenBSD and NetBSD have implimented the use of CVSup. Its a totally painless update process. I will admit that it won't be that simple for the Linux crowd because there are many many distributions that are totally separate of eachother. It would be at least a start if RedHat started using CVSup to keep their security fixes available. Write a dinky lil script that CVSup's the latest stuff, then installs it, tells you to restart/whatever. For what its worth it seems that the Linux community has focused more on installation rather than maintenance. But then again its new, and the NT communities learned that a long time ago.
I will definately agree that Mr. Glass is quite a fool. I had a bit of a tangle with him as I manned the DOC booth selling the Defcon 7 - FreeBSD shirts. He came and argued with me about commercialization/GPL/etcetc. He was pretty unclued on the state of events, although he seemed impressed with the utter lack of Linux *anything* at the event. Actually that was pretty much the most amazing thing about the event, even with all the Skr1pt kiddiez at the event, there were *no* linux CD's being sold, and everything there was *BSD oriented. FreeBSD and NetBSD were being represented by the DoC (Myself, Dover, Cyber/etc) and Mike Smith (and a friend of his who's name slipped my mind). OpenBSD was being sold by Theo's cohorts.
Point 4 is kinda correct actually, RedHat builds a distribution, and they distribute it. I don't believe that they charge for any of the actual contents of the CD. I'm sure I could be wrong, but that seems to be how they would get around licensing issues. I need to re-read all the licenses again, I should know better what I'm talking about.
He's also sorta right about point 6, I'm sure that you *could* hide stuff in source, but its so pointless if its open source anyways:)
Regardless of this, I think Mr. Glass is a first class twit-of-the-media and should be debunked as often as possible, and as publicly as possible.
All of you are being ridiculous. Ever used a Sun box? *GHASP* they have a hardwired host identifier built into them!!! AND OHMYGOD YOU CAN TYPE A COMMAND AND SEE THE IDENTIFIER!!!! OHMYGOD YOUR ETHERNET HAS A HARDWARE IDENTIFIER TOO!!!!
This kind of paranoia only matters if you're using a browser/app that will send back that identifier on request. I'm going to doubt that Netscape will, I'd be pretty assured that MSIE will, and I'm *positive* people will come up with ActiveX tools to get that Host identifier. And the BIGGEST thing is that it will probably only affect windows users aversely because they can't get source code to their OS...
Yep, it is an MS based option, but you don't have to be using NT to do it. Its supported in most HW dialin servers, and with FreeBSD. I use it all the time actually, its pretty easy. Its not just a nice thing, it makes support *really* easy. lUsers don't even have to type in anything other than their username/password and voila!
Actually MSN has some *very* large unix installations, however they don't use Linux. They have a couple *thousand* FreeBSD boxes, Probably 20-50 BSD/OS boxes, a few hundred Sun boxes, and a couple SGIs around. Microsoft uses a lot of Unix in every day use. I'm not too suprised someone fell for the Linux Hype and decided to install it, although I'm curious if they can actually use it since they are using MS-Proxy server and I wouldn't be suprised if there wasn't a Linux client for it...
Thats a buncha bullpuckey:) I've regularly had Apache blasting away with over 600 httpd processes. Well over 200 new connections per second, queueing over 400-500 connections. You must have forgoten about the dinky lil maxprocesses tuning variable. It won't let you go higher than 256. You just have to have enough ram:) My 'big ass' box is a single PII-450 with 1GB Of Ram, Apache stripped down a lil bit. You just gotta have enough Ram and CPU to serve content:) Max blasting point I was running about 70% CPU Usage... Ohyeah, I was using FreeBSD, so YMMV
Slashdot Mirror
Thats very cool to hear that its stompin NT left and right, but I'm suprised that they wouldn't be porting to OpenBSD or FreeBSD first due to networking speed and security issues. Linux is neat and all, but I would still rather have a Solaris box on my front lines than a Linux box for a FW. Right now I don't do either, I've got a FreeBSD box handling 5 networks, all running at 100Mbps at 50-60% capacity (network bandwidth) so I'm pretty happy. I wasn't able to accomplish that with Linux on the same hardware.
To tell you whats truly bizzare, is that Nokia's FW-1 Appliance is based on some kind of BSD, so porting the actual FW to Free/Net/OpenBSD would be a snap. Hopefully they'll see the light and port soon.
Regardless, porting the CP GUI to Linux is great because its so well emulated on other platforms. Solaris, Open/Net/FreeBSD all can run Linux X86 binaries. I'd still rather have a native binary, but any binary is better than none.
Just about every other OS I know of (except for NT) includes this. Having a Kernel Debugger, Kernel Core Dump, and a few other tools available over the past few *YEARS* has saved me a lot of hassle. If Linux hasn't had this till now, I'm sooooooooooo sorry. Thats really dissapointing.
*BSD, Solaris, Dynix, and bazillions of other OS'es have had this ever since they were created.
This is a really neat thing to find, especially since so many people are putting hard work into the desktop environment. So far Linux and NetBSD have put quite a bit of effort into this arena. NetBSD was one of the first FreeNixes to impliment USB support (that was usable) and they're working on quite a few other things. NetBSD recently ported some of the Linux DVD extensions, and we'll probably see that wrapped into the other BSD's sometime soon.
Unfortunately it seems that clustering support has been lacking for *BSD so far. I've seen a few attempts, but nothing really good that I've seen so far. It would be nice if someone ported Beowulf over, because that would give researchers an easy way to migrate over, or use in parallel, or whatever with existing clusters.
Yes, *BSD, and FreeBSD in particular have pretty advanced TCP/IP stacks, so far they've been the only ones capable of driving the Myrinet cards past the gigabit range.
Unfortunately the 2GB file size is still there in 2.2.12. FreeBSD is able to handle files this size and larger on the X86 platform, as well as Net and OpenBSD. Solaris handles it without a hitch as well. What really bugs the hell out of me is the swap size limit with Linux, not being able to use 2GB of swap (or larger) is a pain in the ass. Granted it sucks to be even 500MB into swap, it happens sometimes.
Wonder what they're up to this time. :)
They're running their site offa *single* FreeBSD 3.x box so chances are they're going to be Open-Source OS oriented.
Or, they're a Wine Auction house
Selling a secure webserver and calling it a turnkey package is a load of BS. If that were the solution, life would be easy. There are already buttloads of SSL capable OSS webservers out there.
Hell, one of the best secure webservers out there (stronghold) is basically Apache.
The company I work for now has Doug MacEachern of mod_perl fame :) And I've seen Cliff Skolnick and Brian Behlendorf still around the bay area, so they don't have all of them yet. IBM has dumped a *lot* of money into Apache, and into creating ties with companies that will use their HW/SW platform with Apache. They did help sponsor ApacheCon and had a large presentation there.
I've felt the same way, the FreeBSD installer is just plain out *EASY*. Few have come close enough to duplicate its ease of use. :) and its just plain out fast as all getout. Might be from using PAX instead of tar/cpio/etc.
Actually, the BSD/OS 4.0 installer is just a whizbang deal. Super easy to install (mostly due to limited choices
They toed the PWC company line, said that only NT could deliver 100% uptime with some type of mirroring, and tried to downplay HP-UX, AIX, Solaris, and MVS, since they could only claim
:) I quit about 3 weeks before.
99.995% availability.
Thats pretty a pretty interesting figure, especially since MS can't even do it. I've checked out quite a few versions of 'Wolfpack' and it fails to do even 99% uptime in most situations.
NT also requres *4* machines do get there, and if you're running SQL7, you're SOL because it doesn't support it yet, *AND* its not instantly available either. That DB has to startup, check through possibly corrupted data, and then become available. That could take up to half an hour for a large database... It was very interesting explaining this to the manager of the DB department where I used to work at MSN. He said it was all BS, but gee, when he released his pet project, it flopped HARD and didn't even work because the DB response was so bad. (Maybe a whole slashdot article on its own) Ohwell, too bad for him
I think part of the problem that you'll be encountering is people not wanting to put something weird on X86 hardware. I've had very tough times telling people that yes, free operating systems can provide really great service. One thing I've seen is that companies are not as afraid of SCO or BSD/OS as they are of Linux, might have something to do with the longstanding corporate backing, but X86 OS'es other than NT are tough to push. If you are at a point where BSD/OS is acceptable, you might suggest FreeBSD or even OpenBSD to get in the door. You can make the claims that since they're based on 4.4BSD (lite-2/etc), that you have a similarly stable code base, remind them of the last time they saw big bad things about BSD/OS, FreeBSD or even OpenBSD. You can remind them that Wells Fargo, the US Government, Hotmail, Yahoo, CDROM.COM, and countless others use BSD based products. Its not perfect, but its a start. Getting free software in the door is the biggest step, once you accomplish that, you're almost assured a success.
Its not really the amount of CPU's thats the only factor. Its how much ram it can handle, its how many processes it can handle, its how fast its file system is. NT Beats Linux at this because NT gets a bit sloppy to get its speed, rather than with superior coding. FreeBSD beats Linux in Memory and Filesystem handling, however NT is still reigning in terms of speed on the X86 architechture because of the way its designed - to allow one process to monpolize all CPU time, whereas Un*x doesn't want you to do that.
Its like nobody on Slashdot cares about meaningful things. A 10k donation to OBSD is like a 100k donation towards Linux. Linux already has the name recognition. If you say 'Linux' to the average IT guy, they'll sit there and look like they're stoned for a lil bit. OpenBSD just gets "Huh? Is that like a Bondage thing?"
I hope that people who see this realize what that donation is going to do. Its personally making me very partial to the IDS system that the donator sells, and gosh, it seems to run *GREAT* on OpenBSD why not use that... And yes, I need to do Gig-E IDS because I damn well feel like it!
Chances are they have had more than one NC at NCI.
I wouldn't be suprised if they had StrongARM based units, but I believe the inital run was done with FreeBSD. When I was first in the know about this, it was late 1997/1998, so things may have changed quite a bit. Its quite likely that NCI switched to NetBSD to switch architechtures easily.
CacheFlow's product is similar to NetApp's NetCache product, however the CacheFlow seems to be eating the NetCache alive in terms of hits per second and simultaneous connections. I would reccomend Squid for a 'home' or office installation because its cheap for client use. For Acceleration you'll definately want a CacheFlow. I setup an inverse-web proxy for one of the sites I worked on, and squid couldn't handle the hits per second, it had 0 problem with massive amounts of connections being open which was ok, but not my primary objective.
Choosing Oracle is probably the way to go, mostly because if its what you're already using, you should stick with it. However one problem you might run into if you run it with Linux is the memory data structure size, and the file size. FreeBSD 3.X allows you to have larger than 2GB files around, so does Solaris, and Net/OpenBSD.
There has been some interesting success in putting Oracle 8i on FreeBSD, it seems to run faster and a bit smoother under FreeBSD, but I can't confirm that personally because I haven't put it into production. I am going to be installing it today on my workstation to give it a shot.
To be honest though, you should really be rolling your databases over before they hit filesize limits if you're only growing by 40GB per year. I would invest well in FCAL raid and see what gives you the best performance. Linux's FCAL support sucks so far, we had to discontinue using it and switched for FreeBSD on one installation I was working on. Just a couple of bumps
They are using FreeBSD, I have it from a reliable source inside of NCI. We've talked about this at the BAFUG (Bay Area FreeBSD Users Group) Meetings quite a bit.
This article doesn't suprise me at all. Oracle has been pushing this way for a while. They started pushing that way with NCI (They use FreeBSD) making an NC that runs under Linux wouldn't be that difficult. What Ellison really needs is the clueless IT manager going "oh! Linux Network computers! great *peanutbrainwhirring* linux == good *drool*"
Its what most IT managers do now with NT, soon it'll be Linux, after that who knows what.
Point being is that Oracle isn't stupid, they've already had a lot of the NC work done by NCI, now they just have to port the stuff over. They already have Oracle 8 ported to Linux, chances are they have their 'embedded' Oracle 8 ported over as well. Spending a few hundred man hours to port all their crap onto Linux based NC's is a no brainer. Its free money if it takes off.
Doesn't AutoRPM update and install and not tell you?
Linux needs to take a *big* point from the BSD's on updating. FreeBSD, OpenBSD and NetBSD have implimented the use of CVSup. Its a totally painless update process. I will admit that it won't be that simple for the Linux crowd because there are many many distributions that are totally separate of eachother. It would be at least a start if RedHat started using CVSup to keep their security fixes available. Write a dinky lil script that CVSup's the latest stuff, then installs it, tells you to restart/whatever. For what its worth it seems that the Linux community has focused more on installation rather than maintenance. But then again its new, and the NT communities learned that a long time ago.
I will definately agree that Mr. Glass is quite a fool. I had a bit of a tangle with him as I manned the DOC booth selling the Defcon 7 - FreeBSD shirts. He came and argued with me about commercialization/GPL/etcetc. He was pretty unclued on the state of events, although he seemed impressed with the utter lack of Linux *anything* at the event. Actually that was pretty much the most amazing thing about the event, even with all the Skr1pt kiddiez at the event, there were *no* linux CD's being sold, and everything there was *BSD oriented. FreeBSD and NetBSD were being represented by the DoC (Myself, Dover, Cyber/etc) and Mike Smith (and a friend of his who's name slipped my mind). OpenBSD was being sold by Theo's cohorts.
:)
Point 4 is kinda correct actually, RedHat builds a distribution, and they distribute it.
I don't believe that they charge for any of the actual contents of the CD. I'm sure I could be wrong, but that seems to be how they would get around licensing issues. I need to re-read all the licenses again, I should know better what I'm talking about.
He's also sorta right about point 6, I'm sure that you *could* hide stuff in source, but its so pointless if its open source anyways
Regardless of this, I think Mr. Glass is a first class twit-of-the-media and should be debunked as often as possible, and as publicly as possible.
All of you are being ridiculous. Ever used a Sun box? *GHASP* they have a hardwired host identifier built into them!!! AND OHMYGOD YOU CAN TYPE A COMMAND AND SEE THE IDENTIFIER!!!! OHMYGOD YOUR ETHERNET HAS A HARDWARE IDENTIFIER TOO!!!!
This kind of paranoia only matters if you're using a browser/app that will send back that identifier on request. I'm going to doubt that Netscape will, I'd be pretty assured that MSIE will, and I'm *positive* people will come up with ActiveX tools to get that Host identifier. And the BIGGEST thing is that it will probably only affect windows users aversely because they can't get source code to their OS...
Yep, it is an MS based option, but you don't have to be using NT to do it. Its supported in most HW dialin servers, and with FreeBSD. I use it all the time actually, its pretty easy. Its not just a nice thing, it makes support *really* easy. lUsers don't even have to type in anything other than their username/password and voila!
Actually MSN has some *very* large unix installations, however they don't use Linux.
They have a couple *thousand* FreeBSD boxes, Probably 20-50 BSD/OS boxes, a few hundred Sun boxes, and a couple SGIs around. Microsoft uses a lot of Unix in every day use. I'm not too suprised someone fell for the Linux Hype and decided to install it, although I'm curious if they can actually use it since they are using MS-Proxy server and I wouldn't be suprised if there wasn't a Linux client for it...
Thats a buncha bullpuckey :) :) My 'big ass' box is a single PII-450 with 1GB Of Ram, Apache stripped down a lil bit. :) Max blasting point I was running about 70% CPU Usage... Ohyeah, I was using FreeBSD, so YMMV
I've regularly had Apache blasting away with
over 600 httpd processes. Well over 200 new connections per second, queueing over 400-500 connections. You must have forgoten about the dinky lil maxprocesses tuning variable. It won't let you go higher than 256. You just have to have enough ram
You just gotta have enough Ram and CPU to serve content