Software devs still not doing their jobs right!

[StormPC]

Oh I see. Well I can't really argue the point that software people are keeping up with hardware. They aren't. In fact, it seems that the faster computers get the more inept programmers become, simply because they can. They are dealing with virtually unlimited storage and memory, and incredible processor and graphics card performance. They simply do not need to be as good as programmers used to be.

Fanboy aidanjt? That's really unnecessary.

And just about everyone knows why servers use Intel more often than AMD. It's pretty much the same reason more people use Celeron than A64. Nothing complicated, just marketing.

The new X2 will stomp most Xeon dualies at almost everything. The dual-core Opterons are so much more powerful than anything Intel has to offer. Does it matter? No, because most IT Networking types know squat about hardware.

[aidanjt]

StormPC.. thats a true and valid point, hardware is performing extreamly well, however all that is driven by monies invested by companies and individuals splashing out every two years on new desktops and workstations because software is bogging down their old machines with bloat, poor code, and not taking full advantage of the processors. Which is the thing that is annoying me.

[Nein]

Woh... First of all, your network annalogy is completely and utterly incorrect.. Hubs replicate the raw binary electrical signal and throw it out to all the ports at the same time.

How about one data stream broadcasting to all ports for one intended destination? How did it perform 2 data streams to 2 destinations -- at once or sequential one at a time?

quSwitches makes a decission as to which port it should forward a packet to based on the packets destination based on a MAC table the switch builds as it receives and sends traffic. This is why Switches are far better at passing packets around a network.

Switches are simultaneous n-ways devices, one data stream to 8 ports or 8 data streams to 8 ports, hubs are one data stream at a time devices.

Secondly, I'm no brainwashed nieve idiot.. DMP has been used for a longggg time with mainframes and supercomputers, AMD isn't the first to use it, and the scale they do it on is largely irrelivent since their design is almost identical to that of SMP without Northbridge intervention between their processors.. All AMD has really done is slap a memory controller onto their CPU's, and renamed the FSB to Hypertransport Bus.. If Opterons were so amazing then why do high capacity servers use SPARC and Xeons?.. seriously get real fanboy.

Actually, we graphic hardware guys were the majority and the first users of DMP. Our graphic problems had always been with insufficient never enough bandwidth, bandwidth based processing is our normal solution. Notice GPUs tended to be very low core frequency? They aren't Time-Share Processing designs, had not been for a long long time.

Real DMP extends it scope way out to fibre links and clustering. Distrubting processing on a different level to software applications.

No wonder when nVIDIA mentioned "Scalable Link Interface" and no one seemed to have a clue.

[aidanjt]

A hub is nothing more than a multi-port repeater.
Switches DO NOT send the same packet out to different ports the recipent is already listed on the MAC table

Who the hell taught you networking?

Right.. so nVidia GPU's existed long before the days of mainframes? Do some research.

And you're still going off topic. How rude.

[Nein]

A hub is nothing more than a multi-port repeater.
Switches DO NOT send the same packet out to different ports the recipent is already listed on the MAC table

A switch can function as a hub by sending one data stream to all ports, a hub can not function as a switch as it can only send one data stream at a time.

Who the hell taught you networking?

I worked on Scalable Link Interface hardware even before T1 (mid '80) came into existence.

Right.. so nVidia GPU's existed long before the days of mainframes? Do some research.

nVIDIA didn't invent GPUs.

And you're still going off topic. How rude.

off topic? Proper Processing is proper programming.

[aidanjt]

Listen to me, NIC1 sends out a packet to NIC2 on Tx, the hub receives it and fires the same packet off to NIC1,2,3,4,5,6,7,8 on Rx all at the same time, thus the term 100BaseTx ethernet, it is a dumb simple electronic device.

A switch on the otherhand is different, with NIC1 adding NIC2's MAC address to the Ethernet header, it allows the switch to determine what port NIC2 is connected to via its internal MAC table.

They do two completely different things and that is why switches are a hell of a lot more efficent at dealing with heavily congested networks.

"I worked on Scalible Link Interface hardware even before T1 came into existance"
I'm sure you did considering its an nVidia patiented technology. Graphics technology before T1 consisted of a VGA controller and the CPU doing Bitblt processing. You're not kidding anyone.

"nVidia didn't invent GPUs."
Wow really?

And yes, you're completely off topic in every way possible.. I started this thread discussing how software developers are failing end users by bogging down their systems with crap software which doesn't tap the potential of their hardware. You've went on an SMP vs. DMP rampage and complete distorted the flow of discussion.

[Nein]

Listen to me, NIC1 sends out a packet to NIC2 on Tx, the hub receives it and fires the same packet off to NIC1,2,3,4,5,6,7,8 on Rx all at the same time, thus the term 100BaseTx ethernet, it is a dumb simple electronic device.

Yes, one data stream at a time broadcasting everywhere (one data stream to all points).

A switch on the otherhand is different, with NIC1 adding NIC2's MAC address to the Ethernet header, it allows the switch to determine what port NIC2 is connected to via its internal MAC table.

Yes, one data stream to all MAC addresses (one data stream to all points), or multiple different data streams to and from multiple different MAC addresses (point-2-point).

They do two completely different things and that is why switches are a hell of a lot more efficent at dealing with heavily congested networks.

More efficient because it can do more than one data stream at a time, multiple connections simultaneously (point-2-point). Or it can function as a hub by sending one data stream at a time everywhere (one data stream to all points).

I'm sure you did considering its an nVidia patiented technology. Graphics technology before T1 consisted of a VGA controller and the CPU doing Bitblt processing. You're not kidding anyone.

nVIDIA trademark is "SLI", they can't patent "Scalable Link Interface", it's decades old technology already. Examples of "Scalable Link Interface" hardware - T1, T2, T3, HyperTransport, PCI-Express, 10/100 Ethernet, Fibre-Channel, USB, FireWire, DVI, etc....

And yes, you're completely off topic in every way possible.. I started this thread discussing how software developers are failing end users by bogging down their systems with crap software which doesn't tap the potential of their hardware. You've went on an SMP vs. DMP rampage and complete distorted the flow of discussion.

Did I not mention the nearly complete ignorance of proper processing? Hence the lack of proper programming, for example...

"Its out of the scope of this discussion as DMP isn't an architecture that people use at home/office... you don't see DMP systems sitting under peoples desks fragging in quake.."

"Current AMDs are Distributed Processing designs, point-2-point, direct connect architecture, AMD multi-processor versions are DMPs not SMPs as most ignorant people assumed. Especially Intel trained and raised experts, most of them didn't even know what "Symmetric Processing" actually meant, they had no clue it was Intel trademark name for Time-Share/Time-Division Processing."

[Dougal]

Shouldn't this be moved to the Software dev thread?

[aidanjt]

Dougal, I don't think so, because im focusing on the failure to exploit CPU technology. That our pockets belovenly empty out for new kit. Though if a forum moderator wishes to move the thread thats fair enough.

Nein, you're off topic because your trying to claim one certain arch type is better than another, both have their merits.. Application software doesn't concern itself with one multi-processing layout to the next. It only needs to have a parent thread and worker threads, the kernel and hardware architecture takes care of the rest.

And no, multipule data streams can be sent out from a hub at the same time. Ethernet uses two pairs to reduce collisions, the nature of ethernet allows for multipule streams because of its packet based nature.. Switches only behave like a hub when the Ethernet header doesn't have a recipient MAC address.. its not the same as processing architecture.

Can we keep to the topic please?

[Nein]

Dougal, I don't think so, because im focusing on the failure to exploit CPU technology. That our pockets belovenly empty out for new kit. Though if a forum moderator wishes to move the thread thats fair enough.

Yep, failure to comprehend and use CPU properly is about right.

Nein, you're off topic because your trying to claim one certain arch type is better than another, both have their merits.. Application software doesn't concern itself with one multi-processing layout to the next. It only needs to have a parent thread and worker threads, the kernel and hardware architecture takes care of the rest.

You are off topic because you are too ignorant to be on topic.

And no, multipule data streams can be sent out from a hub at the same time. Ethernet uses two pairs to reduce collisions, the nature of ethernet allows for multipule streams because of its packet based nature.. Switches only behave like a hub when the Ethernet header doesn't have a recipient MAC address.. its not the same as processing architecture.

Ah... PLEASE NOT AGAIN. I am not ignorant enough to claim a switch can not broadcast to all points just as a hub could, and broadcasting to all points is the only thing a hub could do.

Can we keep to the topic please?

I am on topic, ignorance of processing is common and no programmer can do their job right if they didn't have much of a clue.

[aidanjt]

You're calling me ignorant, and you have produced a flawed network annalogy, continuing this DMP vs SMP debate.. SMP distributes threads, if a kernel supports CPU affinity (like Windows NT, linux, bsd, and other kernels do) a task can be asigned to one particular CPU that isn't loaded, AMD's cpus still operate on a threading model, a multi-threaded application works on both. Reread what I said, kernel and hardware arch takes care of thread distribution. This is fact, application level software doesn't need to interface with the arch. End of discussion, now can we get back to the problem at hand, please!

[Nein]

You're calling me ignorant,

I didn't call you ignorant, you talked like an ignoramus, acted like an ignoramus, insisted on being an ignoramus.

and you have produced a flawed network annalogy

I am not the one claiming a switch can not broadcast like a hub.

, continuing this DMP vs SMP debate.. SMP distributes threads, if a kernel supports CPU affinity (like Windows NT, linux, bsd, and other kernels do) a task can be asigned to one particular CPU that isn't loaded, AMD's cpus still operate on a threading model, a multi-threaded application works on both. Reread what I said, kernel and hardware arch takes care of thread distribution. This is fact, application level software doesn't need to interface with the arch. End of discussion, now can we get back to the problem at hand, please!

Distributed Processing and Symmetric Processing needed not be "Multi", only when multiple processors are in use then "Multi" applied.

A single processor can perform Symmetric Processing (Time-Share/Time-Division Processing), it is how Intel's processors normally fuction... just as a single processor can perform Distributed Processing.

[Moby-Dick]

please avoid the insults folks.

FWIW I think software does seem to lag a few years behind hardware , but is that because the development platforms themselves are out of date ?

[aidanjt]

Muti-processing is defined when there are two or more cores doing processing, regardless of packaging.. Weither the work is done in a distributed (parallel) or symmetric (serial) mannor is completely irrelivent in the scope of mutli-threaded software applications. Ethernet is a serial signaling communication medium btw, not parallel.

post edited - I'm sure you didn't want to keep insulting each other after my warning

[aidanjt]

Moby-Dick, yes, there is a lot of truth in what you say, however the tools and methods to leverage current hardware exist. I think developers these days think libraries should do everything for them and not give a toss weither or not their application bogs down people's systems.

[Moby-Dick]

It depends on your defenintion of a developer - If you are contructing an application from "building blocks" then you have to have faith in those blocks and routines. Knowledge of the next layer down isn't nessesary for you to do the job in hand.

Its likes going down from code to chip level , from chip level down to individual logic gate level , and from there down to transistor level and from there down to fabrication process and dopant levels in the substrate.