You are right on #2. However, the largest contribution to ping times is not the Tx buffer at the ISP, but the Tx buffer on your modem! As you all know, most broadband connections have much less uplink than downlink data rates. Most of the bb connections have like 128 kbps of uplink. If you try to send data faster than that, it is buffered inside the modem. It doesn't matter how you prioritize things at that point, because it all gets backed up in the modem. That is why QoS routers have some form of uplink rate matching. That means that you prioritize packets, and only send them at the rate the modem can forward them on. If you choose just the right uplink rate at the router, then you are prioritizing right before the packet gets sent to the ISP. Only if you do all of this right will the ISP's buffering play a major role. Of course, this situation becomes a little more complicated if you have a shared pipe leading to the cable head-end unit.
It may come as a surprise to some people, but it is more accurate to think of Linksys, D-Link, Netgear, etc as marketing and customer support companies than as engineering firms. Go to the Careers page of the Linksys web site. They have lots of openings for Industrial Designers, Product Managers, and Buyers, but none for Programmers. Even the job of 'Product Engineer' focuses on working with partners and fluency in Mandarin.
The truth is that these OEMs all shop for their technology in Taiwan. They buy the boards and software from a handful of companies, and then customize the UI and industrial design to their specifications. The quality of a Linksys or D-Link product comes from who they bought the software and hardware from, and the testing they did before release. Different OEMs may be willing to pay for higher end parts, or more stable software. When you buy a particular product based on the brand name, you are placing your faith not in that OEM's engineers, but in the marketers and project managers that wrote the spec for the product.
I have worked in marketing for wireless products for several years. I have seen firsthand the companies that write the code for these products. The engineers sit in cubicles with 20 square feet of space and ride Chinese-made scooters to work. They work damned hard, but don't necessarily use the best quality controls.
I used a 120 MHz PC as my router for years, and learned a lot by configuring the drivers and adding extra features to it. This box never needed rebooting, but the fan noise was driving me crazy. Finally I broke down and bought a cheap Linksys unit. It needed to be power-cycled several times a week. I gave it away, and bought the DGL from D-Link a couple months ago to move my network to gigE. It has run rock-solid ever since. It does piss me off that I need to reboot it to change most parameters, though.
Out of curiousity, I cracked the case, and saw that it uses a different processor than the Linksys. I believe this is why it cannot load Linux. IMHO Linux is great if you need lots of flexibility. However, if you need something to 'just work', you may be better off with a high quality commercial product.
A custom build of Linux for a router can do many things that a standard product cannot. However, these features are still based on standard HW-SW interactions between the OS and the underlying hardware. To really do special things, you need some type of hardware acceleration that has hooks to software control. A commodity MIPS or ARM core does not provide this.
I play Quake 3 and UT 2004 both wired and wirelessly. Even with my Linux router, I had to turn off file sharing on my other PC in order to play online. With the GameFuel feature turned on, I can play with no problems.
The surprising thing is that the QoS makes such a significant difference, even though the bandwidth of the games are so low compared to my uplink. I measured only about 10 kbps being sent by both games. Unfortunately most FTP transfers will try to use all available bandwidth. This kills the game any time the transfers start. With QoS on, I get 100ms pings on wireless no matter how much I load the network.
I did try the QoS rules on the router. However, I really didn't see much of an improvement above the default automatic provisioning. Maybe someone has some tips on how to optimize it.
The takeaway is that it is best to evaluate networking products on an individual basis, rather than basing it on brands. If you need tons of flexibility, you can't beat a custom Linux router, but for something that works really well out of the box, I am happy with a good quality off -the-shelf product.
Slashdot Mirror
You are right on #2. However, the largest contribution to ping times is not the Tx buffer at the ISP, but the Tx buffer on your modem! As you all know, most broadband connections have much less uplink than downlink data rates. Most of the bb connections have like 128 kbps of uplink. If you try to send data faster than that, it is buffered inside the modem. It doesn't matter how you prioritize things at that point, because it all gets backed up in the modem. That is why QoS routers have some form of uplink rate matching. That means that you prioritize packets, and only send them at the rate the modem can forward them on. If you choose just the right uplink rate at the router, then you are prioritizing right before the packet gets sent to the ISP. Only if you do all of this right will the ISP's buffering play a major role. Of course, this situation becomes a little more complicated if you have a shared pipe leading to the cable head-end unit.
It may come as a surprise to some people, but it is more accurate to think of Linksys, D-Link, Netgear, etc as marketing and customer support companies than as engineering firms. Go to the Careers page of the Linksys web site. They have lots of openings for Industrial Designers, Product Managers, and Buyers, but none for Programmers. Even the job of 'Product Engineer' focuses on working with partners and fluency in Mandarin. The truth is that these OEMs all shop for their technology in Taiwan. They buy the boards and software from a handful of companies, and then customize the UI and industrial design to their specifications. The quality of a Linksys or D-Link product comes from who they bought the software and hardware from, and the testing they did before release. Different OEMs may be willing to pay for higher end parts, or more stable software. When you buy a particular product based on the brand name, you are placing your faith not in that OEM's engineers, but in the marketers and project managers that wrote the spec for the product. I have worked in marketing for wireless products for several years. I have seen firsthand the companies that write the code for these products. The engineers sit in cubicles with 20 square feet of space and ride Chinese-made scooters to work. They work damned hard, but don't necessarily use the best quality controls. I used a 120 MHz PC as my router for years, and learned a lot by configuring the drivers and adding extra features to it. This box never needed rebooting, but the fan noise was driving me crazy. Finally I broke down and bought a cheap Linksys unit. It needed to be power-cycled several times a week. I gave it away, and bought the DGL from D-Link a couple months ago to move my network to gigE. It has run rock-solid ever since. It does piss me off that I need to reboot it to change most parameters, though. Out of curiousity, I cracked the case, and saw that it uses a different processor than the Linksys. I believe this is why it cannot load Linux. IMHO Linux is great if you need lots of flexibility. However, if you need something to 'just work', you may be better off with a high quality commercial product. A custom build of Linux for a router can do many things that a standard product cannot. However, these features are still based on standard HW-SW interactions between the OS and the underlying hardware. To really do special things, you need some type of hardware acceleration that has hooks to software control. A commodity MIPS or ARM core does not provide this. I play Quake 3 and UT 2004 both wired and wirelessly. Even with my Linux router, I had to turn off file sharing on my other PC in order to play online. With the GameFuel feature turned on, I can play with no problems. The surprising thing is that the QoS makes such a significant difference, even though the bandwidth of the games are so low compared to my uplink. I measured only about 10 kbps being sent by both games. Unfortunately most FTP transfers will try to use all available bandwidth. This kills the game any time the transfers start. With QoS on, I get 100ms pings on wireless no matter how much I load the network. I did try the QoS rules on the router. However, I really didn't see much of an improvement above the default automatic provisioning. Maybe someone has some tips on how to optimize it. The takeaway is that it is best to evaluate networking products on an individual basis, rather than basing it on brands. If you need tons of flexibility, you can't beat a custom Linux router, but for something that works really well out of the box, I am happy with a good quality off -the-shelf product.