Dual Core Routers

[Shikami]

With dual core routers, there needs to be some confirmation, if possible, that both cores are enabled. Some of the routers will not support typical Linux features to check the kernel for this; such as Linksys EA6700.

Some routers have the second core disable, and some will have them enabled. If at all possible on your reviews if there can be some form of confirmation to know. You do have the ability, better than us, to be in contact with the company of the router. I think this can be beneficial information for the consumer, and help with the purchases (caveat emptor).

 

[thiggins]

What is the benefit of that information? The bottom line is performance, no?

 

[Shikami]

Even RMerlin has a mention about Netgear's R6250 not using the second core (http://forums.smallnetbuilder.com/showthread.php?t=12057). It makes me wonder, especially with the new Broadcom processor being AMP and SMP (and even single core configurable), it might get a little confusing as to how "dual core" the routers are. Some routers may have one core set to only be for particular tasks (AMP). Maybe perhaps this is why the R6250 does well in file copy test, and is a bit lower in over throughput compared to routers of the same architecture?

The bottom line of performance may be disconcerting too, again the R6250. Since most routers now have many forms of offloading and hardware acceleration, a simple benchmark may only be testing the offloading engine and acceleration. Not really asking to change the system you have. But inform the consumer more, if possible. You have the connections, and they are more likely to answer you properly than us.

 

[sfx2000]

Single Core or Multiple Core - doesn't really matter these days - it's all about threads.

The MIPS cores in many of the SOC's do very well with threads - ARM cores, a little bit less so, that's why you're seeing more multiple core SOC's, as ARM is the current favorite...

Not just HW, but the host OS and the tasks assigned...

sfx

 

[sfx2000]

I can think of one task, again this would be at the OS level with a SMP router CPU/SoC implementation where there would be some benefit...

VPN - because of the encryption, they would benefit from both SMP and dedicated HW support...

sfx

 

[Shikami]

Well, that is actually contradictory. It is, indeed, all about the thread. But if you only have one core then the OS will have to time the placement of the execution of the threads to one core. When another core is available for execution, then depending on the method of a multi-core implementation, the threads are executed between cores, or available for particular cores. This being Symmetrical Multiprocessing (http://en.wikipedia.org/wiki/Symmetric_multiprocessing), or Asymmetrical Multiprocessing (http://en.wikipedia.org/wiki/Asymmetric_multiprocessing).

And, of course, the OS has responsibility for it is what feeds the processing, but we are only going to see Linux in these cases; and unfortunate not many use a modern kernel (2.6.x). As for MIPS vs ARM and the better architecture, this will always wax and wane. As for now ARM is better for many reasons, but both architectures are not close to the power of a modern x86-64.

It just seems that the companies are "saying" that a router is multicore, but there is no assurance of this. There has to be a way to know that it is.

 

[RMerlin]

Even RMerlin has a mention about Netgear's R6250 not using the second core (http://forums.smallnetbuilder.com/showthread.php?t=12057). It makes me wonder, especially with the new Broadcom processor being AMP and SMP (and even single core configurable), it might get a little confusing as to how "dual core" the routers are.

The issue with Netgear is (most likely) they rushed the product to market, and need more time to properly implement SMP support to their firmware (which was probably never designed to properly support it). I think they plan to release a firmware update later this year that will enable the second core (in addition to the missing 128 MB of RAM).

Kinda reminds me when, a few years ago, HP released their first ePrint-enabled printer, except that ePrint wasn't implemented yet - it was going to be provided in a future driver/firmware update.

 

[sfx2000]

@shikami - quaint that you bring up AMP - old school stuff for big-iron - still see this from time to time, but modern kernel schedulers and compilers are more focused on SMT/SMP

@RMerlin - you're deeper into the kernel with the builds you've been working on - there's a number of schedulers in linux - what are you seeing the OEM's use most frequently, and have you had a change to tinker about with different schedulers?

sfx

 

[Shikami]

The issue with Netgear is (most likely) they rushed the product to market, and need more time to properly implement SMP support to their firmware (which was probably never designed to properly support it).

True as this may be there is an issue here; and I think this is exactly my point. As a consumer we need to know if this is going to be the case if we are going to be spending the money on such a product, and the cost is equal to a high performing and feature rich router. But yet it is isn't, and how can we [the consumer] know if it is not for you and other kernel hackers and this site?

@shikami - quaint that you bring up AMP - old school stuff for big-iron - still see this from time to time, but modern kernel schedulers and compilers are more focused on SMT/SMP

I understand what you may think and know, but the chip is configurable for SMP, AMP and single core operation. Noting again the complexity and a disconcerting issue. (http://www.broadcom.com/products/Wireless-LAN/802.11-Wireless-LAN-Solutions/BCM4707-4708-4709)

 

[Terry Kennedy]

The issue with Netgear is (most likely) they rushed the product to market, and need more time to properly implement SMP support to their firmware (which was probably never designed to properly support it). I think they plan to release a firmware update later this year that will enable the second core (in addition to the missing 128 MB of RAM).

I think the key word in that sentence is "plan". There have been a number of products where features promised in the literature were either unimplemented or implemented but not working reliable, and then the manufacturer did a "stealth" release of a newer variant with the same packaging, down to the same UPC and users who had the older variant never got what they were promised.

I'm not saying that will happen here, just that a number of potential customers would think twice about buying a product with an unavailable feature planned for future release.