NASA has a few things it can do for itself.... namely:
* Identify and correct any problems that can be fixed. * Resume flights as soon as feasible; * Ask Congress for a boatload of money; * Use boatload of money to design Shuttle2.
Line 1 is interesting because well, there are inherent risks in flying the shuttle. You absolutely can't guarantee safety; I mean, honestly, if a micrometeor hits the shuttle while in space, well, it's a problem.
*ALL* future shuttle flights should be equipped with a Canadarm, ISS docking ring, EVA packs, and enough fuel to get to the ISS.
No matter what. If that means we have to cut back on the payloads, well, too bad.
Even if we knew there were cracked tiles on Columbia in space, what could we have done for them? Not really very much.
We need a rescue system; some way to either get guys down without their vehicle, or a way to park 'em up there 'till we can get another vehicle in motion.
That should be Priority One. Next up, let's replace the shuttle with something more modern --- something that can carry as much payload, but more modern.
There's merit to teaching a token-based media access control mechanism.
There's also merit to teaching a shared media access mechanism.
If all we taught were the Most Popular (tm) results, music students would study Britney Spears, Lit majors would study Harry Potter, and CS would be nothing but how to use MS Word & Excel.
Sheesh.
Wait till you get to the Real World (tm). You'll grow up fast, I promise.
TR isn't that limited in use; lots of mainframe-type environments still use it. Hell, we still have SNA links....
Then you don't understand what a CS degree is good for.
My suggestion: Go to Chubb.
If you start thinking in the "That's not practical, who cares" mode, you belong in a trade school.
Sorry, I know that's not very politically correct, but it's the TRUTH.
Now, if you want to learn real computer SCIENCE, stick it out.
Learning assembly language for a theoretical computer is a great exercise -- you have to actually exercise that mush between your ears!
My favorite class in CS was Theory of Digital Machines.... designing AND, OR, NOT gates, building some theoretical microprocessors.... stuff that isn't "practical", but that theory means the world to you later on...
Again, if you want practical, go to Chubb. If you want to learn something, stick it out...
Many internships are not paid. Or you could say the pay is the experience. It might be that most I/T internships are paid, but maybe that is changing now.
Really? Many moons ago (~10 years ago) I did some internship work at Warner Music (Before we had this whole World Wide Web thing, and certainly before Napster was even a wet dream....) and I was paid for my time.:)
I'm surprised to hear that most internships aren't paid at all.... I figured they wouldn't pay WELL, but they'd pay at least something!
So it can't outperform some other OS on a machine that 99.9% of companies will never be willing to buy? Big fat hairy deal.
Yeah, it sure is a big deal. Means it's useless to me.
Most of the world isn't on the absurdly high end. Even the part of it is has many people that would rather solve their problems with a collection of cheaper machines.
That's not true at all. Sure, the people who read the trade rags that say "Come on man, try some Linux, everyone's doing it..." think so, but people who have _real_ data to protect? No way.
Sorry, no CIO worth his salt is going to trust their important financial system to a filesystem driver written by a 13-year-old Yugoslavian kid named Konchenko.
I didn't mention the "E12K". There is no such system. There's an Enterprise 10000, which is a Cray descendant, and there's a Sun Fire 12000 and Sun Fire 15000 which are pure Sun designs. Sure, they inherit some of the E10k's technology (OK, a lot of it) but they're still Sun-engineered and designed.
And I disagree that they're desinged for what Beowulf clusters do nicely.
Have you ever built a Beowulf? I have.
Have you ever built a cluster of E10k's? I have.
Beowulfs are fine for home-grown applications for which performance is THE key, and reliability & security take a backseat.
For real-world use, they're not all that good outside the scientific community.
And yes, I know Sun techs don't show up in suits. That's not what I meant. What I meant was that some suits show up to pacify management while the technicians get the work done. Never, ever underestimate how important that is.
Ever.
I get the feeling that most of you who think that a bunch of Dells running Linux is a replacement for the high-end systems haven't ever used one.
You just don't understand what a Regatta or SF15k are until you use them. And not just as an end-user, but as an admin. At 3am. When a CPU flakes out. When you have to do some massive upgrades and can't afford the downtime.
It's a whole different world. Take everything you know and pitch it out the window; we're not in Kansas anymore.
Too bad, you guys could have saved quite a bit of money. BTW, why all three? This seems quite silly unless managing complexity is something you enjoy doing.
Money's a LOT less important than reliability. And for the jobs some of my systems are running, there's really no easy way to do it besides one large honkin' server....
As for the 3-platform problem... we have 2 sysadmin teams; a Solaris team and an AIX team. The HP is a new venture, and right now the Solaris team is transitioning to manage it. Honestly, I would've rather have bought another SF15k, but it's political...........
RAC(Oracle 9i)/RHAS + Dell = (cost & performance. > Sun). Who cares if you have 48-106 slow USIII processors when you can have 32 fast Pentium processors.
Depends on the OS. Having 32 "fast" Pentium processors running what OS? Linux?
Linux doesn't scale nearly as well as Solaris to that many processors, no matter what you do.
Scalability is a lot more important than speed sometimes. Having a SF15k means that I can replace bad CPU's, memory, I/O, etc. without taking the node down. At all.
You don't know the meaning of SLICK AS HELL until you do a dynamic reconfiguration; just remove CPU/memory boards or hot-swap PCI components at will.
Very, very, VERY slick.
and that's not to mention the domain abilities, and things like Inter-domain networks......
If you don't believe me have a look at SPEC.org.
We all know the SPARC CPU isn't the fastest-ever-made. It wasn't designed for that; it was made to be scalable.... I have to say though, the 1.2GHz UltraSPARC III-Cu procs are wickedly fast.
And if all we cared about was speed, wouldn't we all be using Alpha processors (Our company just bought a pair of HP Marvel Alphaclusters too..... so we do really run the gamut of hardware:)
A Dell configuration can scale to 32 TB of storage is that enough?
Not always.
I have a 40TB Hitachi 9980 on the floor right now.
Keeping financial data for 10 YEARS at a clip can require _lots_ of storage.....
A Dell solution will also support.9999 uptime, but I know you don't need that much uptime since "suits" actually have time to get to your company when an outage occurs.
Really? Can you online/offline CPU's on the fly? Can you change out I/O boards with the system up?
Can you drain memory and remove it to replace failed components without bringing the OS down?
Can you add CPU's and memory to a running system?
Even more, can you REMOVE CPU's and memory from a running system to create another [LPAR|Domain|Partition]?
Now things have turned the other way. Every new Unix application is available for Linux. The old ones that still matter are being ported rapidly. It's getting so that Linux is the only "no-brainer" deployment. Everything else requires thought. Is AIX supported? Is Solaris supported? Who knows - just use Linux because you know it will work.
Stuff from freshmeat doesn't count.....
Anything that matters runs on either Solaris, AIX, or HP-UX. Full stop. Nothing else scales.
It may have potentially been Winders 95 chit, but I don't know; I had been Converted by then.
As for Winders 3.1; no, it certainly wasn't Winders 3.1 stuff. Winders was simply another application that ran on top of DOS. What I mentioned above were DOS drivers and executables; none of it was Winders-based....
There was a Winders 3.x Banyan client at the time, but we found it to be such a POS that we didn't use it... we just stuck with the same stuff that had been working for years....
Somehow I doubt Warner Music is still using Banyan these days......:)
1. Ok, well, yeah, but the drives cost more than 1U dual PIII servers.
That's right, and a Mercedes costs more than a Fiat.
2. Yeah, you can, but try upgrading the firmware on an A3500FC on the fly. Would you now trust hitting the disk array with your database at the same time? I wouldn't....
True enough, but anyone worth their salt knows the A3500 was a flaming piece of....... that Sun OEM'd from LSI.
Now, take an E10k. I can dynamically add/remove processors, memory, SBus cards, PCI cards, etc.
In fact, I just replaced 4 SBus I/O mezz's each on 2 of my E10k's with PCI ones. All while the system was up. And the database was running. And the data was processing. And not a single hiccup.
Now _that_ is what I call hot-swap hardware.
3. Solaris doesn't have 64 bit memory access. Its like 38 or 48 bit. Check their UltraSparc docs.
Errr.... check the Solaris docs.
4. Sure, and for things that need "decent" clustering, its one of a few options. Most things, however, don't need "decent" clustering.
OK, you tell me how to keep an Oracle database highly available without decent clustering.
Yes, a parallel DB is still technically a cluster.
5. No current Sun product supports 128 processors, and if you need a loaded E15k you have very specific needs indeed.
True enough, max. CPU in an SF15k (They're _not_ part of the Enterprise line), is 108.
However, it's not a "very specific" need; I see lots of places where running several domains on SF15k's would be ideal. I also have some E10k's that run balls-to-the-wall, 64 CPU's, 64gb RAM in one domain. We're trying to determine exactly what our performance gains would be if we migrated off of the pair of E10k's mentioned above to a single SF15k. Honestly, I don't think a single 15k would handle the load. The application in question seems to like more processors at a (relatively) slower speed than fewer procs at faster speed....
6. Again, how many UltraSparc II/III processors have failed on you in the past month? If you deal with lots of them, they die depressingly frequently. Especially considering the cost.
The US-II chips were very unreliable until the Sombra modules became available. They got seriously reliable after that.
Sun never introduced a Sombra-like module for the desktop-class equipment (E450 and below), _BUT_ they did replace the CPU's with IBM e-cache modules with CPU's with Sony e-cache modules. I haven't seen an e-cache parity error in a long, long time (And I support about 300 Sun machines, from Ultra 1's through SF6800's, and soon 15k's....)
7. Anybody can and does provide support like this. Sun premium support (gold/platinum) is really freaking expensive.
That depends on what it means to your business. If downtime costs you serious $$$, that contract is worth its weight in gold.
The RAID5 is a much better solution, since it can handle a single drive failure with no problems. The odds of two drives failing at the same time are really low. So as long as you are prompt about replacing failed drives, you can't go wrong.
Except that performance blows.
And it gets worse when (not if, but WHEN) a single disk fails. Parity recalculation is EXPENSIVE.
RAID0+1, or RAID 10. Mirror that stuff, don't look back.
And egads, seriously, 2 drives per IDE controller? Performance has to be in the toilet already.
Just buy a damned XServe RAID and be done with it.
ealise that most people don't want to hand-hack some poorly documented config file, they want to play games, browse the web, do their taxes and all the other things that Windows allows them to do (albeit sometimes in shades of blue). Don't diss people like my dad just because he's got better things to do.
You know, it's not even all about people like your Dad -- people like me want stuff to Just Work too.
I'm a Unix sysadmin by trade. (Mostly Solaris, but now some HP-UX as well) At work, I do nothing _but_ futz with configuration files and tweak things and make it go --- not so much because things are busted, but because I'm trying to make them run _better_. Things like setting the sd_max_throttle setting in Solaris, or various ndd commands to disable source-routed frames and so on...
Anyway, when I get home at the end of the day, the LAST DAMNED THING ON EARTH I want to be doing is tweaking config files, downloading Widgetx 0.234567 and finding out it's incompatible with the libc libraries in my OS, so I have to upgrade to glibc and now Widgetx still doesn't work because glibc broke my window manager.
I remember those days --- the libc-to-glibc migration... I still have nightmares about that.
March 24, 2001 I bought my first Mac. And I haven't looked back since...
Now stuff Just Works. I come home and I can read my email, play games, connect to work, and even watch DVD's and edit home movies and play with my digital camera....
Some folks can't see the forest from the trees --- The end-goal here should be for the system to just GET OUT OF MY WAY and let me work... and Linux unfortunately hasn't done that in many years.
Slashdot Mirror
NASA has a few things it can do for itself .... namely:
* Identify and correct any problems that can be fixed.
* Resume flights as soon as feasible;
* Ask Congress for a boatload of money;
* Use boatload of money to design Shuttle2.
Line 1 is interesting because well, there are inherent risks in flying the shuttle. You absolutely can't guarantee safety; I mean, honestly, if a micrometeor hits the shuttle while in space, well, it's a problem.
*ALL* future shuttle flights should be equipped with a Canadarm, ISS docking ring, EVA packs, and enough fuel to get to the ISS.
No matter what. If that means we have to cut back on the payloads, well, too bad.
Even if we knew there were cracked tiles on Columbia in space, what could we have done for them? Not really very much.
We need a rescue system; some way to either get guys down without their vehicle, or a way to park 'em up there 'till we can get another vehicle in motion.
That should be Priority One. Next up, let's replace the shuttle with something more modern --- something that can carry as much payload, but more modern.
--DM
Both.
....
There's merit to teaching a token-based media access control mechanism.
There's also merit to teaching a shared media access mechanism.
If all we taught were the Most Popular (tm) results, music students would study Britney Spears, Lit majors would study Harry Potter, and CS would be nothing but how to use MS Word & Excel.
Sheesh.
Wait till you get to the Real World (tm). You'll grow up fast, I promise.
TR isn't that limited in use; lots of mainframe-type environments still use it. Hell, we still have SNA links
--NBVB
Then you don't understand what a CS degree is good for.
.... stuff that isn't "practical", but that theory means the world to you later on ...
...
My suggestion: Go to Chubb.
If you start thinking in the "That's not practical, who cares" mode, you belong in a trade school.
Sorry, I know that's not very politically correct, but it's the TRUTH.
Now, if you want to learn real computer SCIENCE, stick it out.
Learning assembly language for a theoretical computer is a great exercise -- you have to actually exercise that mush between your ears!
My favorite class in CS was Theory of Digital Machines.... designing AND, OR, NOT gates, building some theoretical microprocessors
Again, if you want practical, go to Chubb. If you want to learn something, stick it out
--NBVB
Bah. One of my vendors took us for a cruise around Manhattan Island
I'm a diehard Yankee fan, but VIP seats are VIP seats.
..... and I suppose the mainframe is dead too, right?
Sorry, it doesn't matter how hard you try, this stuff matters in the real world.
Really? Many moons ago (~10 years ago) I did some internship work at Warner Music (Before we had this whole World Wide Web thing, and certainly before Napster was even a wet dream
I'm surprised to hear that most internships aren't paid at all.... I figured they wouldn't pay WELL, but they'd pay at least something!
--DM
Yeah, it sure is a big deal. Means it's useless to me.
That's not true at all. Sure, the people who read the trade rags that say "Come on man, try some Linux, everyone's doing it..." think so, but people who have _real_ data to protect? No way.
Sorry, no CIO worth his salt is going to trust their important financial system to a filesystem driver written by a 13-year-old Yugoslavian kid named Konchenko.
I didn't mention the "E12K". There is no such system. There's an Enterprise 10000, which is a Cray descendant, and there's a Sun Fire 12000 and Sun Fire 15000 which are pure Sun designs. Sure, they inherit some of the E10k's technology (OK, a lot of it) but they're still Sun-engineered and designed.
And I disagree that they're desinged for what Beowulf clusters do nicely.
Have you ever built a Beowulf? I have.
Have you ever built a cluster of E10k's? I have.
Beowulfs are fine for home-grown applications for which performance is THE key, and reliability & security take a backseat.
For real-world use, they're not all that good outside the scientific community.
And yes, I know Sun techs don't show up in suits. That's not what I meant. What I meant was that some suits show up to pacify management while the technicians get the work done. Never, ever underestimate how important that is.
Ever.
I get the feeling that most of you who think that a bunch of Dells running Linux is a replacement for the high-end systems haven't ever used one.
You just don't understand what a Regatta or SF15k are until you use them. And not just as an end-user, but as an admin. At 3am. When a CPU flakes out. When you have to do some massive upgrades and can't afford the downtime.
It's a whole different world. Take everything you know and pitch it out the window; we're not in Kansas anymore.
Be careful with 108528-16! It's got data corruption problems ............
Now 105181-32.... _THAT_'s a stable kernel!
--DM
Money's a LOT less important than reliability. And for the jobs some of my systems are running, there's really no easy way to do it besides one large honkin' server....
As for the 3-platform problem
Depends on the OS. Having 32 "fast" Pentium processors running what OS? Linux?
Linux doesn't scale nearly as well as Solaris to that many processors, no matter what you do.
Scalability is a lot more important than speed sometimes. Having a SF15k means that I can replace bad CPU's, memory, I/O, etc. without taking the node down. At all.
You don't know the meaning of SLICK AS HELL until you do a dynamic reconfiguration; just remove CPU/memory boards or hot-swap PCI components at will.
Very, very, VERY slick.
and that's not to mention the domain abilities, and things like Inter-domain networks
We all know the SPARC CPU isn't the fastest-ever-made. It wasn't designed for that; it was made to be scalable....
I have to say though, the 1.2GHz UltraSPARC III-Cu procs are wickedly fast.
And if all we cared about was speed, wouldn't we all be using Alpha processors (Our company just bought a pair of HP Marvel Alphaclusters too
Not always.
I have a 40TB Hitachi 9980 on the floor right now.
Keeping financial data for 10 YEARS at a clip can require _lots_ of storage.....
Really? Can you online/offline CPU's on the fly? Can you change out I/O boards with the system up?
Can you drain memory and remove it to replace failed components without bringing the OS down?
Can you add CPU's and memory to a running system?
Even more, can you REMOVE CPU's and memory from a running system to create another [LPAR|Domain|Partition]?
All without even a reboot?
--DM
Until it outperforms Solaris on a 106-way Sun Fire 15k, I'd say it doesn't scale enough.
Until it outperforms HP-UX on a Superdome, I'd say it doesn't scale enough.
Until it outperforms AIX on a Regatta, I'd say it doesn't scale enough.
"Most" people may use some desktop-class Intel PC-type servers, but the real money is in high-end systems.
At my job, I have all 3 of the above; Linux isn't even an option for us.
When you run Really Big Databases (tm), Linux isn't an option. When you require 4-9's or 5-9's of uptime, Linux isn't an option.
With Sun, HP & IBM, I get a bunch of suits who show up to soothe management every time there's an outage, large or small.
What do I get with Linux? Some 14-year old from the Czech Republic?
This isn't to say that Linux doesn't have a place in the world; it does. It just isn't on the high end.
--NBVB
Stuff from freshmeat doesn't count
Anything that matters runs on either Solaris, AIX, or HP-UX. Full stop. Nothing else scales.
It may have potentially been Winders 95 chit, but I don't know; I had been Converted by then.
....
...... :)
As for Winders 3.1; no, it certainly wasn't Winders 3.1 stuff. Winders was simply another application that ran on top of DOS. What I mentioned above were DOS drivers and executables; none of it was Winders-based....
There was a Winders 3.x Banyan client at the time, but we found it to be such a POS that we didn't use it... we just stuck with the same stuff that had been working for years
Somehow I doubt Warner Music is still using Banyan these days
--NBVB
Son, that's not Windows shit.
That's DOS shit.
And like I said, it was many years ago. And paid the bills then.
These days, I'm enlightened. I'm a Unix admin by day, a Unix (Mac OS X) user at night.
Uggggh ....
...
/i:protman.ini
/nc
..... it's been many years, but that stuff still haunts me :)
I really didn't need that flashback
I see horrible memories of BeyondSnail and IM III......
Yikes.
device=c:\banyan\protman.dos
@echo off
c:\banyan\e2131
ban
bind
yikes
OK, OK, so Sun considers the E450 a deskSIDE system
--DM
Fine, and my 60000 user mailserver running Sendmail works great too. It passes about a million messages/day.
And that's one one machine (With a standby node). So what is technically wrong with sendmail?
That's right, and a Mercedes costs more than a Fiat.
True enough, but anyone worth their salt knows the A3500 was a flaming piece of
Now, take an E10k. I can dynamically add/remove processors, memory, SBus cards, PCI cards, etc.
In fact, I just replaced 4 SBus I/O mezz's each on 2 of my E10k's with PCI ones. All while the system was up. And the database was running. And the data was processing. And not a single hiccup.
Now _that_ is what I call hot-swap hardware.
Errr.... check the Solaris docs.
OK, you tell me how to keep an Oracle database highly available without decent clustering.
Yes, a parallel DB is still technically a cluster.
True enough, max. CPU in an SF15k (They're _not_ part of the Enterprise line), is 108.
However, it's not a "very specific" need; I see lots of places where running several domains on SF15k's would be ideal. I also have some E10k's that run balls-to-the-wall, 64 CPU's, 64gb RAM in one domain. We're trying to determine exactly what our performance gains would be if we migrated off of the pair of E10k's mentioned above to a single SF15k. Honestly, I don't think a single 15k would handle the load. The application in question seems to like more processors at a (relatively) slower speed than fewer procs at faster speed....
The US-II chips were very unreliable until the Sombra modules became available. They got seriously reliable after that.
Sun never introduced a Sombra-like module for the desktop-class equipment (E450 and below), _BUT_ they did replace the CPU's with IBM e-cache modules with CPU's with Sony e-cache modules. I haven't seen an e-cache parity error in a long, long time (And I support about 300 Sun machines, from Ultra 1's through SF6800's, and soon 15k's....)
That depends on what it means to your business. If downtime costs you serious $$$, that contract is worth its weight in gold.
Fair enough .... but certainly not better than RAID1 :)
3 hits with a sledgehammer and you couldn't bust the cartridge?
Spent too much time playing with video games, I guess. Have your muscles _completely_ atrophed or are you just _that_ big of a wuss?
Except that performance blows.
And it gets worse when (not if, but WHEN) a single disk fails. Parity recalculation is EXPENSIVE.
RAID0+1, or RAID 10. Mirror that stuff, don't look back.
And egads, seriously, 2 drives per IDE controller? Performance has to be in the toilet already.
Just buy a damned XServe RAID and be done with it.
You know, it's not even all about people like your Dad -- people like me want stuff to Just Work too.
I'm a Unix sysadmin by trade. (Mostly Solaris, but now some HP-UX as well) At work, I do nothing _but_ futz with configuration files and tweak things and make it go --- not so much because things are busted, but because I'm trying to make them run _better_. Things like setting the sd_max_throttle setting in Solaris, or various ndd commands to disable source-routed frames and so on
Anyway, when I get home at the end of the day, the LAST DAMNED THING ON EARTH I want to be doing is tweaking config files, downloading Widgetx 0.234567 and finding out it's incompatible with the libc libraries in my OS, so I have to upgrade to glibc and now Widgetx still doesn't work because glibc broke my window manager.
I remember those days --- the libc-to-glibc migration
March 24, 2001 I bought my first Mac. And I haven't looked back since
Now stuff Just Works. I come home and I can read my email, play games, connect to work, and even watch DVD's and edit home movies and play with my digital camera
Some folks can't see the forest from the trees --- The end-goal here should be for the system to just GET OUT OF MY WAY and let me work
meh, i just ordered a 36TB Hitachi 9980 disk array at work ...
so that's 3.6LOC's then?
But three lefts do!
Dah-dum-dum.
Sun's good at that ... we do it too :)
.... :)
*you* convince a CIO who just spent $5m on a pair of E10k's that now he has to license Solaris
Hey, that's a list of people they've f'd THIs YEAR. :)
m om my/85
There used to be a list on my BBS way-back-when that started with Seattle Software....
Here's a partial list:
http://five2one.org/stdio/index.cfm/daddy/show/