Most beowulfs use a gigE interconnect... perhaps have two or three NICs per node, one for the system interconnect, and the other two for connection to the CCD. The CCD module could easily be broken down into a grid of virtual segments, each with its own cache and a gigE interconnect.
Another option is to wait for 10gigE (along with the rest of the supercomputing world) or go with Myrinet, which has recently broken the 1 gigabit barrier.
Watch for the browser distribution to shoot from 0.01% Mozilla to about 95% Mozilla!
But seriously... a Slashdotting usually takes the server from light load to an extreme load in a matter of seconds... remaining at max load for about 4 hours (~2 million unique visitors an hour) and then dropping back to normal over the next 12 hours.
We have two RIP servers in our office, both made and packaged by Colorbus. One is NT based and runs on what looks to be a rebadged Intergraph/SGI Zx10. The other is unix based and is a bit faster, it runs on an SGI O2 (although most of the work is being done by a PCI card). I have no idea what these cost, but I'd imagine they wern't cheap...
Now if only my.edu would build this excellent... um, tool!
SGI did this two years ago, as a *product*
on
IBM's Deep View
·
· Score: 2
SGI did the graphics cluster thing over two years ago and even released it as a product. Very, very similar to the IBM. They even ported several of their programming APIs and SDKs from InfiniteReality to the Linux Graphics Cluster. Not many were sold, however. Heck, Slashdot didn't even cover it. It was still pretty neat to see multiple spanned monitors and even composited high res projectors driven by a half rack of Linux PCs. Many of the demos were actually ported from SGI's big iron Onyx machines and worked just as well on the cluster. The basic setup was a stack of rackmount Linux boxes using nVidia AGP cards and custom PCI cards daisy chained together to provide sync for glSwapBuffers among other things. Also availble were gigE and Myrinet for networking the machines with something better than 100BT. A compositor (similar to what's used in InfinitePerformance) was also available.
I too wanted to do all of that with my primary workstation... and without sacraficing everything that I was already used to.
So, I bought a G4 with Mac OS X.
Granted, I had to add ram and buy a wireless 3 button mouse, but it's been working great and both my camcorder and Kodak digital camera worked perfect the first time I plugged them in. The next purchases on my list are a larger hard drive, Final Cut Pro (at my university computer store) and a USB jog/shuttle controller.
And I still handle my email with pine and fetchmail.
BMW installs fine seats in their automobiles... for firm support needed in performance driving. If I were to build a computer desk around an automobile seat, I would probably use something a bit more plush from a Lincoln, Cadillac, or Jaguar. Or better yet, get an overstuffed leather chair from a furniture store.
Why isn't the equipment wireless, using bluetooth or something similar for everything to communicate. Its not going to feel very realistic to me if I have a strand of wires attached to me.
SGI was showing off some examples of what you are describing. Basicly, the big iron (clusters, or large machines such as Onyxes) sit in the machine room, while the users have wireless webpads and such elsewhere. It's the only way we can currently tap the power of thousands of processors and dozens of 3D accelerators in a handheld using current technology.
SIGGRAPH exhibits closed on Thursday evening, but here are some of the biggest highlights:
SGI annouced their Infinite Reality 4 option for the Onyx series... comes standard with 1gbyte of texture ram and 2.5gbyte of buffer, expandable to 10gbyte of buffer for a total of 11gbyte of onboard gfx ram. Up to 16 IR4 subsystems can be installed in a single machine. Each subsystem can drive up to 8 monitors... or all subsystems can run in parallel for greater performance. The Virtual LA Urban Simulation project demoed part of their 3D LA using IR4 and the older IR3. They currently have over 1TB of texture and geometry data from Los Angeles, mostly in downtown areas... though they have 20,000 square miles mapped out, 4,000 of which are quite detailed.
Sun was showing off their XVR-4000 gfx option, a cardset that uses the IPA slot found in most Ultra-series machines. It has about 8x the geometry performance of IR3 and about 50% of the fill performance of IR4... for a fraction of the cost. 1gbyte of texture and 144mbyte of buffer. Different market targets, but interesting none the less.
All of my coworkers are excited about the SS51, I have a feeling it's going to be a huge seller... perhaps even one of the best selling pieces of PC kit this year. I wouldn't be too surprised if, over the next six months or so, we start seeing this little wonder popping up all over the place... offices, retail point-of-sale, on sitcom sets, etc.
In a related note, I'm curious about the "optional parallel port"... is it nothing more than a USB adapter? I would love a real parallel port as I perfer parallel for printers.
I was fooling around with a fresh download of MSIE 5.2.1 on Mac OS X earlier today. The download was 7.2 MB compressed and I belive the final install took up a little over 11 MB. I didn't have time to see what the breakdown was. As I recall, version 5.1.5 for Mac OS 9 is a bit smaller, about a 5.5 MB download and 9 MB installed. Though 5.1 has fewer features and less help/readme fluff.
MSIE + Outlook Express for Solaris and HP/UX is well over 20 MB compressed (www.microsoft.com/unix).
It's hard to tell how much space MSIE takes up on Win32 given that it relies on libraries and bindings built into the OS and windowing system.
That said, I think mozilla's current size isn't too bad. I'd much rather see performance improvements, especially for older machines and the latest crop of tiny all-in-one machines. (Such as those VIA Mini-ITX boxes... WinChip-like performance, but really small/quiet/cool). RAM usage could use some trimming as well.
While I'd imagine The Tech will be open late for the mature gamers (mature games start at around 5 PM) I find it odd that The Tech normally closes at 6 PM.
I don't know how many times I've been in San Jose for a convention, usually less than a few blocks from The Tech, but have been unable to go because they close at 6.
San Jose, the city that's alive way into the wee hours of the morning...
I'm not sure of the whens or whys... but for the past few years, most OmniMax installations are now called "IMAX Dome". The IMAX.com website has information on both formats.
Perhaps the IMAX folks have their fingers in the OmniMax market now, too.
or you have the write cache on your hard drives battery backed
I've seen such an option on big external RAID arrays. Makes sense, lets the write cache be written to disk before the power goes out.
I'm curious, though, do any hard drives have this feature? Maybe not a full battery, but perhaps a capacitor to store enough juice to write that 8 MB of cache data down to disk before it's gone for good? Or perhaps some sort of bolt-on option for existing internal drives?
I ask this as I'm an average joe with home-brew and cheap-label servers (I built most, a few are PII Dells and Gateways). My machines are pretty stable, but I only have about 70 minutes of battery backup from my UPS... and there's no way I could justify buying a generator.
Maybe it's different where you are, but I can't buy PBR in my area anymore. Folks in the local bottleshop claim they can't order it anymore. (No longer made? Out of business?)
Slashdot Mirror
Most beowulfs use a gigE interconnect... perhaps have two or three NICs per node, one for the system interconnect, and the other two for connection to the CCD. The CCD module could easily be broken down into a grid of virtual segments, each with its own cache and a gigE interconnect.
Another option is to wait for 10gigE (along with the rest of the supercomputing world) or go with Myrinet, which has recently broken the 1 gigabit barrier.
But what would constitute a slashdot?
Watch for the browser distribution to shoot from 0.01% Mozilla to about 95% Mozilla!
But seriously... a Slashdotting usually takes the server from light load to an extreme load in a matter of seconds... remaining at max load for about 4 hours (~2 million unique visitors an hour) and then dropping back to normal over the next 12 hours.
We have two RIP servers in our office, both made and packaged by Colorbus. One is NT based and runs on what looks to be a rebadged Intergraph/SGI Zx10. The other is unix based and is a bit faster, it runs on an SGI O2 (although most of the work is being done by a PCI card). I have no idea what these cost, but I'd imagine they wern't cheap...
Way to go IBM! Now let's see you put it to a good use... say...
.edu would build this excellent... um, tool!
CAVE Quake!!
: )
Now if only my
SGI did the graphics cluster thing over two years ago and even released it as a product. Very, very similar to the IBM. They even ported several of their programming APIs and SDKs from InfiniteReality to the Linux Graphics Cluster. Not many were sold, however. Heck, Slashdot didn't even cover it. It was still pretty neat to see multiple spanned monitors and even composited high res projectors driven by a half rack of Linux PCs. Many of the demos were actually ported from SGI's big iron Onyx machines and worked just as well on the cluster. The basic setup was a stack of rackmount Linux boxes using nVidia AGP cards and custom PCI cards daisy chained together to provide sync for glSwapBuffers among other things. Also availble were gigE and Myrinet for networking the machines with something better than 100BT. A compositor (similar to what's used in InfinitePerformance) was also available.
m l
.edu and .org have also rolled their own graphic clusters... though I don't know who supplied the compositors.
More information (white paper and data sheet) can be found on SGI's legacy systems page:
http://www.sgi.com/products/legacy/vis_systems.ht
I belive a few
(Prepare to lose all karma...)
I too wanted to do all of that with my primary workstation... and without sacraficing everything that I was already used to.
So, I bought a G4 with Mac OS X.
Granted, I had to add ram and buy a wireless 3 button mouse, but it's been working great and both my camcorder and Kodak digital camera worked perfect the first time I plugged them in. The next purchases on my list are a larger hard drive, Final Cut Pro (at my university computer store) and a USB jog/shuttle controller.
And I still handle my email with pine and fetchmail.
I don't mean to flame/troll, but I don't get it either...
Not to mention that Postgres has been more open much longer than MySQL has been.
I personally use Interbase, but have used both Postgres and MySQL.
Now this is what I consider to be a proper computer workstation...
http://www.sara.nl/hpc.www/images/unite/unite.jpg
Ok... so it's 1997 technology, but you get the idea.
Plus it's cooler looking than SARA's current monster.
BMW installs fine seats in their automobiles... for firm support needed in performance driving. If I were to build a computer desk around an automobile seat, I would probably use something a bit more plush from a Lincoln, Cadillac, or Jaguar. Or better yet, get an overstuffed leather chair from a furniture store.
Don't be too depressed if the first home versions are somewhat crude... remember, you have to start somewhere...
The Telstar Ranger wasn't exactly Quake.
Why isn't the equipment wireless, using bluetooth or something similar for everything to communicate. Its not going to feel very realistic to me if I have a strand of wires attached to me.
SGI was showing off some examples of what you are describing. Basicly, the big iron (clusters, or large machines such as Onyxes) sit in the machine room, while the users have wireless webpads and such elsewhere. It's the only way we can currently tap the power of thousands of processors and dozens of 3D accelerators in a handheld using current technology.
http://www.sgi.com/visualization/van/
SIGGRAPH exhibits closed on Thursday evening, but here are some of the biggest highlights:
SGI annouced their Infinite Reality 4 option for the Onyx series... comes standard with 1gbyte of texture ram and 2.5gbyte of buffer, expandable to 10gbyte of buffer for a total of 11gbyte of onboard gfx ram. Up to 16 IR4 subsystems can be installed in a single machine. Each subsystem can drive up to 8 monitors... or all subsystems can run in parallel for greater performance. The Virtual LA Urban Simulation project demoed part of their 3D LA using IR4 and the older IR3. They currently have over 1TB of texture and geometry data from Los Angeles, mostly in downtown areas... though they have 20,000 square miles mapped out, 4,000 of which are quite detailed.
Sun was showing off their XVR-4000 gfx option, a cardset that uses the IPA slot found in most Ultra-series machines. It has about 8x the geometry performance of IR3 and about 50% of the fill performance of IR4... for a fraction of the cost. 1gbyte of texture and 144mbyte of buffer. Different market targets, but interesting none the less.
At least we know Jobs has a sense of humor!
Maybe it's my gamma setting, but that dude almost looks as though he was computer generated!
All of my coworkers are excited about the SS51, I have a feeling it's going to be a huge seller... perhaps even one of the best selling pieces of PC kit this year. I wouldn't be too surprised if, over the next six months or so, we start seeing this little wonder popping up all over the place... offices, retail point-of-sale, on sitcom sets, etc.
In a related note, I'm curious about the "optional parallel port"... is it nothing more than a USB adapter? I would love a real parallel port as I perfer parallel for printers.
I was fooling around with a fresh download of MSIE 5.2.1 on Mac OS X earlier today. The download was 7.2 MB compressed and I belive the final install took up a little over 11 MB. I didn't have time to see what the breakdown was. As I recall, version 5.1.5 for Mac OS 9 is a bit smaller, about a 5.5 MB download and 9 MB installed. Though 5.1 has fewer features and less help/readme fluff.
MSIE + Outlook Express for Solaris and HP/UX is well over 20 MB compressed (www.microsoft.com/unix).
It's hard to tell how much space MSIE takes up on Win32 given that it relies on libraries and bindings built into the OS and windowing system.
That said, I think mozilla's current size isn't too bad. I'd much rather see performance improvements, especially for older machines and the latest crop of tiny all-in-one machines. (Such as those VIA Mini-ITX boxes... WinChip-like performance, but really small/quiet/cool). RAM usage could use some trimming as well.
While I'd imagine The Tech will be open late for the mature gamers (mature games start at around 5 PM) I find it odd that The Tech normally closes at 6 PM.
I don't know how many times I've been in San Jose for a convention, usually less than a few blocks from The Tech, but have been unable to go because they close at 6.
San Jose, the city that's alive way into the wee hours of the morning...
*sigh*
I'm not sure of the whens or whys... but for the past few years, most OmniMax installations are now called "IMAX Dome". The IMAX.com website has information on both formats.
Perhaps the IMAX folks have their fingers in the OmniMax market now, too.
I'm curious, how can a script (software) reboot a a server that has already halted?
I'm a diabetic and have a blood glucose testing kit. The bottle that holds the actual strips has a sticker that reads "the accuracy of palladium!".
Found that kinda funny.
Must be the heatstroke talking...
or you have the write cache on your hard drives battery backed
I've seen such an option on big external RAID arrays. Makes sense, lets the write cache be written to disk before the power goes out.
I'm curious, though, do any hard drives have this feature? Maybe not a full battery, but perhaps a capacitor to store enough juice to write that 8 MB of cache data down to disk before it's gone for good? Or perhaps some sort of bolt-on option for existing internal drives?
I ask this as I'm an average joe with home-brew and cheap-label servers (I built most, a few are PII Dells and Gateways). My machines are pretty stable, but I only have about 70 minutes of battery backup from my UPS... and there's no way I could justify buying a generator.
http://www.lucent.com/minds/telstar/telstarsat.jpg
It sickens me that this is hosted by Lucent, but it does the job. Too bad more neat "online" photos wern't at this resolution...
Quoth the NASA site, "Frequencies used were 6,390 MHz uplink and 4,170 MHz downlink".
Is this a typo? How were such frequencies possible in the early 1960s? And using less than 15 watts to boot!?
Maybe it's different where you are, but I can't buy PBR in my area anymore. Folks in the local bottleshop claim they can't order it anymore. (No longer made? Out of business?)
I wonder if this has anything to do with the IP (patents, designs, etc) Microsoft recently bought from Silicon Graphics, Inc (SGI)...