I hope they have spiced this up considerably, because I have seen a similiar demo with an interactive flight from space down to some icy mountain terrain as early as 1996.
(If my memory serves me right, this was on an Onyx2 Infinite Reality, at the CeBit expo, with the Graphics Boards just flown in the day before (or so the salesperson told me))
First you write about SGI having needed to dump MIPS for IA32 and later you criticize them for building a PC? This doesn't make sence, since you can't change to IA32 and cheaper one-chip graphics hardware without building something that comes pretty close to a PC.
In fact, the first IA32-based machines from SGI weren't quite PCs, they were so peculiar that SGI needed to provide a special Hardware Abstraction Layer for NT to run. which, I suspect, was just one more point why those machines weren't as highly acclaimed as SGI hoped. (The performance also didn't quite meet the high expectations SGI had built up before the launch.)
I think, however, had Linux then already enjoyed the broad success it has now, SGI could have fared much better by making Linux its IA32 graphics workstation OS and trying to keep its customer base from wandering off to Windows NT.
It doesn't make much sense to compare number crunching supercomputers to high-end (more or less general purpose) servers, as the former are much too specialized.
However, to make Opterons come in 106-processor configurations, you indeed need what the original poster lauded Sun for, experience in designing systems.
Don't forget that the Sun 386i was a dismal failure, though.
I certainly hope Sun will do better with their Opteron boxes, and will make them real Sun boxes. So far the x86-based Sun boxes are just OEM machines with a Sun logo slapped on it.
Yeah, right, Checkpoint on Linux on IP330. Get the half-assed Linux support of Checkpoint together with the sub-par performance of the overpriced IP330. Now THATs a real good point...
Every SSI has some amount of intercommunication it needs to keep itself in a consistent state. At a certain number of CPUs this communication grows so large, that adding more CPUs doesn't yield results. This is the scalability limit I meant.
That said, please decide what you want. You can have large CPU Single System Image scalability, go to SGI and pay the price. As I wrote above in this thread, the interconnects for node coupling that is sufficiently fast for SSI operation isn't cheap.
So, maybe your question should rather read: "Gee, wouldn't it be cool if you had a massively scalable system that was as cheap as a blade server but used a single system image like an Origin?" To which I'd say: "Yes, would be cool, but won't ever be there."
And now what if I don't quite fit on a 12 CPU box? Wait for Sun bringing the V1680 and watch them duplication their "Midframe" line in the V-Class?
I could understand the V880 as an E450 successor with processors internal and lots of internal disk, but I can't really understand V1280 with its not-quite-Uniboard CPU boards, except as a stopgap measure forced by competition. Who will now buy F4800 anymore?
(Boss, we need to upgrade from 2.6 to 8. - What, how do we know it will work! That's like upgrading WFW 3.11 to Win2000! The world will end! Can't we just upgrade to 3.5?? I hear we have that on one of our servers...)
Just tell him you need to upgrade from SunOS 5.6 to SunOS 5.8. Simple, isn't it?:)
Well, the Fire 15K scales up to 96 CPUs, it even has something like blades - it is called Uniboards.;-) Sure the price for base configuration it still high. Same with SGI - just they have a litte bigger granularity with their CPU Bricks, as they like to call them. You see, modularity is there.
The problem is that the high-speed interconnects that acually scale up to several hundred CPUs are hideously expensive. Gigabit Ethernet just doesn't cut it for these applications and that is what common blades usually have.
1) This heterogenous blade system is INCREDIBLY cool!!! You can have Intel blades, Sparc blades, encryption blades, network caching/acceleration, blades, etc. etc. etc. all in one frame. The fact that Sun (late as they are) is the first out of the blocks with this idea is remarkable.
They aren't, HP was first. You can have x86 blades and PA-RISC blades in the same chassis. What I'd find REALLY INCREDIBLY cool would be if the hardware vendors would standardize on a common blade platform. x86, SPARC, PA blades in a chassis with features like the one from IBM, that would be cool.
That said, I also find it a bit lacking that there is no word about what tech acually is in the load balancing and SSL accel blades.
(But you can have SSL-style CAs for PGP, as Thawte demonstrated.)
Not only Thawte (which I remember more as a PGP-style CA for SSL (X509) certs), but many others too act as central CA für PGP keys, for example the german publisher Heise or the german research network.
The advantage of X509 certs is that you can have a hierarchy of CAs. The basic HTTPS cert doesn't use that though, so the PGP model could probably work there.
This means that unlike IBM and MSFT who can rest assured that if there is a problem with their software their users will have to call their expensive support lines, users of OSS can simply fix the code or ask on a newsgroup and get a faster and sometimes better response from some enterprising hacker than from some tech support flunkie
Yeah, but can you bet your business on news group replies, which you may or may not get? I've asked enough questions on Usenet I didn't get any reaction to at all, and I've seen way too much "stop whining, fix it yourself" flames that I could comfortably rely on volunteer support via news groups or mailing lists. Sure, if a problem isn't critical, I too ask on a news group and try to figure out the solution myself, but when I really need something fixed yesterday, I'd rather have a service contract that guarantees me that my problem gets the appropriate attention. (depending on what service level I'm willing to pay for)
Isn't the Samba team based in au? I thought Australia gave much better protection to reverse engineering than the US.
Yeah, that's a very interesting question. Some of the Samba developers appear to be in Germany, too, where reverse engineering for the sake of compatibility is allowed. Would the resulting product still be redistributable in the US?
Re:The Strange Case Of The Video Card Industry
on
Goodbye, Number Nine
·
· Score: 1
Yeah, what happened to Tseng Labs? I recall the ET4000/W32p being the first common chipset with 2D acceleration and awesome transfer rates. It really must have rocked those days. (I couldn't afford one and was stuck with my lousy Paradise:-)
But what happened to them after? I also remember them bringing out the ET6000, beating the Matrox Millennium in 2D performance, but I think they too missed the 3D train and have gone the same way as #9.
The aren't gone or deleted. It will not allow the user to run or save them. If you later change your security policy you can save/run them any time you like. The data is always there.
Yeah. Most users will change the security policy exactly *one time* to the relaxed setting and leave it there. Great security advancement.
Um, you _can_ administer NT via SSH from home. We've got several NT boxes here which we can administer via SSH (run on a Linux box, of course) using VNC. Works like a dream.
I guess it is a bit sluggish over a modem line, isn't it?
Perhaps it is acceptable for your WebDesigners to use MS Access as a frontend to the MySQL database through ODBC drivers, so that the actual data is still stored in MySQL. We use that setup with our Sybase databases and our Web development people are quite happy with it.
Slashdot Mirror
I hope they have spiced this up considerably, because I have seen a similiar demo with an interactive flight from space down to some icy mountain terrain as early as 1996.
(If my memory serves me right, this was on an Onyx2 Infinite Reality, at the CeBit expo, with the Graphics Boards just flown in the day before (or so the salesperson told me))
You are contradicting yourself in this post.
First you write about SGI having needed to dump MIPS for IA32 and later you criticize them for building a PC? This doesn't make sence, since you can't change to IA32 and cheaper one-chip graphics hardware without building something that comes pretty close to a PC.
In fact, the first IA32-based machines from SGI weren't quite PCs, they were so peculiar that SGI needed to provide a special Hardware Abstraction Layer for NT to run. which, I suspect, was just one more point why those machines weren't as highly acclaimed as SGI hoped. (The performance also didn't quite meet the high expectations SGI had built up before the launch.)
I think, however, had Linux then already enjoyed the broad success it has now, SGI could have fared much better by making Linux its IA32 graphics workstation OS and trying to keep its customer base from wandering off to Windows NT.
It doesn't make much sense to compare number crunching supercomputers to high-end (more or less general purpose) servers, as the former are much too specialized.
However, to make Opterons come in 106-processor configurations, you indeed need what the original poster lauded Sun for, experience in designing systems.
We have boatloads of E450s. You can still do better with a 4 way Intel/AMD.
Well, nowadays you have to compare with a V440.
Don't forget that the Sun 386i was a dismal failure, though.
I certainly hope Sun will do better with their Opteron boxes, and will make them real Sun boxes. So far the x86-based Sun boxes are just OEM machines with a Sun logo slapped on it.
Siemens a major original member of OSF?
This is odd, considering that SINIX, Siemens' own UNIX, has always been rather plain System V.
write me an iptables rule that stops all GIF images from being loaded from an arbitrary website.
:)
Write me an INSPECT rule.
I can't see how this is done, except by using the Security Servers, which are Proxys. (and god-awful proxys, besides)
Yeah, right, Checkpoint on Linux on IP330.
Get the half-assed Linux support of Checkpoint
together with the sub-par performance of the overpriced IP330. Now THATs a real good point...
Sure they did. They made lots of money with it.
Every SSI has some amount of intercommunication it needs to keep itself in a consistent state. At a certain number of CPUs this communication grows so large, that adding more CPUs doesn't yield results.
This is the scalability limit I meant.
That said, please decide what you want.
You can have large CPU Single System Image scalability, go to SGI and pay the price. As I wrote above in this thread, the interconnects for node coupling that is sufficiently fast for SSI operation isn't cheap.
So, maybe your question should rather read: "Gee, wouldn't it be cool if you had a massively scalable system that was as cheap as a blade server but used a single system image like an Origin?"
To which I'd say: "Yes, would be cool, but won't ever be there."
Many CPUs, Cheap, SSI, pick two.
Indeed.
But there won't ever be, as single system images limit scalability.
Wait for Sun bringing the V1680 and watch them duplication their "Midframe" line in the V-Class?
I could understand the V880 as an E450 successor
with processors internal and lots of internal disk,
but I can't really understand V1280 with its not-quite-Uniboard CPU boards, except as a stopgap measure forced by competition.
Who will now buy F4800 anymore?
(Boss, we need to upgrade from 2.6 to 8. - What, how do we know it will work! That's like upgrading WFW 3.11 to Win2000! The world will end! Can't we just upgrade to 3.5?? I hear we have that on one of our servers...)
Just tell him you need to upgrade from SunOS 5.6 to SunOS 5.8. Simple, isn't it?
Well, the Fire 15K scales up to 96 CPUs, it even has something like blades - it is called Uniboards. ;-)
Sure the price for base configuration it still high.
Same with SGI - just they have a litte bigger granularity with their CPU Bricks, as they like to call them. You see, modularity is there.
The problem is that the high-speed interconnects that acually scale up to several hundred CPUs are hideously expensive. Gigabit Ethernet just doesn't cut it for these applications and that is what common blades usually have.
They aren't, HP was first. You can have x86 blades and PA-RISC blades in the same chassis.
What I'd find REALLY INCREDIBLY cool would be if the hardware vendors would standardize on a common blade platform. x86, SPARC, PA blades in a chassis with features like the one from IBM, that would be cool.
That said, I also find it a bit lacking that there is no word about what tech acually is in the load balancing and SSL accel blades.
The advantage of X509 certs is that you can have a hierarchy of CAs. The basic HTTPS cert doesn't use that though, so the PGP model could probably work there.
Maybe it would make more sense to adapt PGP signing to TLS and use the existing web of trust?
Yeah, but can you bet your business on news group replies, which you may or may not get?
I've asked enough questions on Usenet I didn't get any reaction to at all, and I've seen way too much "stop whining, fix it yourself" flames that I could comfortably rely on volunteer support via news groups or mailing lists.
Sure, if a problem isn't critical, I too ask on a news group and try to figure out the solution myself, but when I really need something fixed yesterday, I'd rather have a service contract that guarantees me that my problem gets the appropriate attention. (depending on what service level I'm willing to pay for)
Yeah, that's a very interesting question. Some of the Samba developers appear to be in Germany, too, where reverse engineering for the sake of compatibility is allowed.
Would the resulting product still be redistributable in the US?
I recall the ET4000/W32p being the first common chipset with 2D acceleration and awesome transfer rates. It really must have rocked those days.
(I couldn't afford one and was stuck with my lousy Paradise
But what happened to them after?
I also remember them bringing out the ET6000, beating the Matrox Millennium in 2D performance, but I think they too missed the 3D train and have gone the same way as #9.
Yeah. Most users will change the security policy exactly *one time* to the relaxed setting and leave it there.
Great security advancement.
Slashdot is like Usenet, only worse.
I guess it is a bit sluggish over a modem line, isn't it?
Perhaps it is acceptable for your WebDesigners to use MS Access as a frontend to the MySQL database through ODBC drivers, so that the actual data is still stored in MySQL.
We use that setup with our Sybase databases and our Web development people are quite happy with it.