Router throughput.

[JackBurton]

I've got money saying the D-Link will work like a charm.

 

[Nothinman]

My guess is these things aren't memory bound, but processor. Who knows, I've never checked the memory on these guys.

I was going to say the opposite because I can't imagine it takes much work to do a hash lookup and then rewrite a few fields of a packet.

 

[spidey07]

Originally posted by: Nothinman

My guess is these things aren't memory bound, but processor. Who knows, I've never checked the memory on these guys.

I was going to say the opposite because I can't imagine it takes much work to do a hash lookup and then rewrite a few fields of a packet.

For normal "routers" they are always processor bound. It does take a fair amout of work to route. Hence why the big dogs do all this stuff in hardware, because there simply aren't processors out there fast enough to do what is necessary of them.

I don't know about the SOHO stuff as they have such little resources as it is.

good discussion though.

 

[Madwand1]

I just did a test with a WRT54G model 1.0 with the latest firmware.

I basically connected it to my regular router via the WRT's WAN port, and connected a single computer to the WRT's LAN-side switch; set the WRT up as a gateway picking up an IP off the regular router, and forwarded a single port, 5001 to its attached computer.

Then I used a version of TTCP (PCAUSA version 2.01.01.08) to measure the bandwidth between a computer on the regular network and the computer now attached to the WRT (via port 5001).

On two separate runs, the utility reported bandwidth around 16.5 Mb/s. This technically meets the FiOS 15 Mb/s download capacity, but just barely -- it'd say it's pushing it.

The upload bandwidth, however, came to around 33 Mb/s, which seems to indicate that the download processing is taking its toll on the link speed, whereas the upload processing is lighter.

Doing simultaneous uploads and download was even more interesting -- here, I get around 14 Mb/s download with simultaneous 5-6 Mb/s upload. Considering that FiOS 15 gives 15 Mb/s download and 5 Mb/s upload, this seems almost precient on Linksys' part. It's also very interesting that the resources were given to process download bandwidth instead of upload, although faster uploads would likely have been achievable in my case at least.

I got around the same results on a couple of different test runs, so this doesn't seem to be just random luck.

Considering that this is the original model 1.0 router, which has gone through several hardware revisions and certainly some hardware upgrades, I think it's a good indication for FiOS 15 Mb/s. However, it's borderline 15 Mb/s, and clearly wouldn't meet FiOS 30 Mb/s.

Moreover, whatever a Linksys 1.0 does is not going to be conclusive for any other router -- they should be measured individually. Especially a $20 router... There is a concept of "ghetto chic", and some computer hardware / applications might fit that term, but a $20 router on a 15 Mb/s internet service? I think we all agree that'd probably be sad at best.

Edit / addendum: I think that this sort of performance actually has little impact / relevance to commercial applications. Any time that a technician spends fiddling around with SOHO gear to get it to work, or even worse, the users spend in downtime, will quickly dissolve the price differences between that gear and commercial gear that actually stays up and works out of the box under stress all the time.

 

[nweaver]

do the v1-4 routers support SNMP? Perhaps a quick test to see CPU/memory load while testing these. Solar winds has an easy to use windows demo that will show near realtime CPU load, or MRTG would work as well.

 

[JackBurton]

Originally posted by: Madwand1
I just did a test with a WRT54G model 1.0 with the latest firmware.

I basically connected it to my regular router via the WRT's WAN port, and connected a single computer to the WRT's LAN-side switch; set the WRT up as a gateway picking up an IP off the regular router, and forwarded a single port, 5001 to its attached computer.

Then I used a version of TTCP (PCAUSA version 2.01.01.08) to measure the bandwidth between a computer on the regular network and the computer now attached to the WRT (via port 5001).

On two separate runs, the utility reported bandwidth around 16.5 Mb/s. This technically meets the FiOS 15 Mb/s download capacity, but just barely -- it'd say it's pushing it.

The upload bandwidth, however, came to around 33 Mb/s, which seems to indicate that the download processing is taking its toll on the link speed, whereas the upload processing is lighter.

Doing simultaneous uploads and download was even more interesting -- here, I get around 14 Mb/s download with simultaneous 5-6 Mb/s upload. Considering that FiOS 15 gives 15 Mb/s download and 5 Mb/s upload, this seems almost precient on Linksys' part. It's also very interesting that the resources were given to process download bandwidth instead of upload, although faster uploads would likely have been achievable in my case at least.

I got around the same results on a couple of different test runs, so this doesn't seem to be just random luck.

Considering that this is the original model 1.0 router, which has gone through several hardware revisions and certainly some hardware upgrades, I think it's a good indication for FiOS 15 Mb/s. However, it's borderline 15 Mb/s, and clearly wouldn't meet FiOS 30 Mb/s.

Moreover, whatever a Linksys 1.0 does is not going to be conclusive for any other router -- they should be measured individually. Especially a $20 router... There is a concept of "ghetto chic", and some computer hardware / applications might fit that term, but a $20 router on a 15 Mb/s internet service? I think we all agree that'd probably be sad at best.

Edit / addendum: I think that this sort of performance actually has little impact / relevance to commercial applications. Any time that a technician spends fiddling around with SOHO gear to get it to work, or even worse, the users spend in downtime, will quickly dissolve the price differences between that gear and commercial gear that actually stays up and works out of the box under stress all the time.

A small correction. The most common plan for FIOS is 15 down 3 up. The more expensive plan ($200/m) is 30 down 5 up. You tested an older WRT54G and it looks like it was able to handle a heavy FIOS load. Now, the D-Link I linked is a faster router (processor wise) than even the newest WRT54G. So like I said, for $100 (Best Buy has it on sale), you can have one of the best consumer routers (DLink DGL-4300) and it should handle the FIOS connection just fine. Get it!

 

[Goosemaster]

For reference, my soekris 4801 with a 266mhz geode (amd) processor and 128MB ram has a raw eference thoroughput speed of about 50Mb/s without many rules. With IPsec VPN traffic the thoroughput goes down to about 3-5Mbps, and will hit about 8 with harware encryption. I currently do not use hardware encryption and only turn the VPN on when needed (not much since I use other technologies). In addition, my connection is limtied to 6Mbps.

With large NAT, accesslist, and traffic shaping rulesets, which I have, about 50 simultaneous bitorrent downloads with a global connection ceiling set set to 800 maximum conenctions, VoIP and additional http browsing, my CPU might blip at 99% for second or two, but stays right around 60-70%.

It maxes out my connection and can defintiely do a LOT more. While gaming + voIP + downloading a large file + surfing simultaneously, I have never seen the CPU past 20-30% and it always maxes out my connection.

As far as consumer routers go, the linksys wrtseries @ 16Mbps are some of the fastest.

 

[JackMDS]

http://en.wikipedia.org/wiki/Wrt-54g

:sun:

 

[Nothinman]

For normal "routers" they are always processor bound. It does take a fair amout of work to route. Hence why the big dogs do all this stuff in hardware, because there simply aren't processors out there fast enough to do what is necessary of them.

Even so, I wouldn't exepect them to lockup or kill the connection as many people have said happens when they start a torrent. Drop packets, yes I can understand that, but not dropping the whole interface.

 

[spidey07]

Nothingman,

moving a packet through the router involves in interrupt. It can't do anything if it is overloaded - effectively locking up. Nor would I expect it to have sophisticated queueing. Once it gets behind it may simply never catch up. Now if it doesn't come back after the load dies down then that's a problem with the code.

Let's just say I've killed even the most expensive routers with too much traffic.

 

[Madwand1]

Originally posted by: JackBurton
A small correction. The most common plan for FIOS is 15 down 3 up. The more expensive plan ($200/m) is 30 down 5 up. You tested an older WRT54G and it looks like it was able to handle a heavy FIOS load. Now, the D-Link I linked is a faster router (processor wise) than even the newest WRT54G. So like I said, for $100 (Best Buy has it on sale), you can have one of the best consumer routers (DLink DGL-4300) and it should handle the FIOS connection just fine. Get it!

Thanks for the corection. Re the DGL-4300, -- I got one for the gigabit switch a long time ago, and have posted about owning one a few times already. I don't have FiOS, nor expect to get it hereabouts in the near future. I thought that the Linksys' performance was more intersting for the majority of readers, so haven't tested the DGL-4300 yet (not to mention that taking that out would upset my network / WAN far more than fooling around with the unused Linksys). Wouldn't it be "funny" if it wasn't as fast? Hehe...

I expect it to be fine; note that Verizon supplies some D-Link routers with their service (modified for their own access/diagnostics at least), and have a support note up about some other D-Link routers. It would be strange for the "higher end" DGL-4300 to not work in this case. To be honest though, although unlikely, I consider it a theoretical possibility that the DGL's more complex processing / feature set might make the WAN speed slower beyond the expected max consumer internet speed (which would be less than 10 Mb/s).

I also find that IPSec kills my bandwidth -- bringing gigabit on decent computers down to fast ethernet speed or so. I did these tests some time ago, and posted them, but don't recall in detail. It was too horrific. Really bad at high speed, not so bad when the speeds are lower.

 

[Nothinman]

moving a packet through the router involves in interrupt. It can't do anything if it is overloaded - effectively locking up. Nor would I expect it to have sophisticated queueing. Once it gets behind it may simply never catch up. Now if it doesn't come back after the load dies down then that's a problem with the code.

Receiving a packet requires an interrupt, I don't think sending one does. And even so, the interrupt handler should be smart enough to not stall in the middle if it gets called too frequently. Usually all they do is copy the packet to kernel memory and reenable the appropriate interrupt line. And in Linux there are ways to work around that, NAPI which polls for new packets periodically instead of handling each packet individually at each interrupt. The NAPI docs in the kernel say they've achieved 890K packets/sec with only 17 interrupts. For the custom embedded systems I can understand how they might have problems with the devices locking up simply because their code is stupid, but the Linux based devices shouldn't have any such problems unless the hardware itself locks up or there is some unknown kernel memory leak.

 

[spidey07]

Nothingman,

You're assuming the manufacturers actually care and want to spend time, effort, money, QA to good/efficient code.


Not being a smart alleck, but it's true. They're more concerned with tricking the customer into the latest, next big thing! And getting product to market ASAP.

 

[Nothinman]

You're assuming the manufacturers actually care and want to spend time, effort, money, QA to good/efficient code.

That's part of the point, they wouldn't have to if they would use half-decent hardware (i.e. nothing from Broadcom) and use Linux as a base. =)

Not being a smart alleck, but it's true. They're more concerned with tricking the customer into the latest, next big thing! And getting product to market ASAP.

I know, it's the same thing in the 'enterprise market'. Most of the enterprise tools I see for sale are just a dozen or so other tools cobbled together with some buzzwords slapped on the packaging. We just had CA come in to setup a demo of some software and their product was no less than 4 seperate products that they acquired tied together with XML (yay buzzwords) and duct tape. None of the apps speak the same languages or even have similar UIs, they just acquired them, changed the name and started reselling them as one big package.

 

[JackBurton]

Originally posted by: Madwand1

Originally posted by: JackBurton
A small correction. The most common plan for FIOS is 15 down 3 up. The more expensive plan ($200/m) is 30 down 5 up. You tested an older WRT54G and it looks like it was able to handle a heavy FIOS load. Now, the D-Link I linked is a faster router (processor wise) than even the newest WRT54G. So like I said, for $100 (Best Buy has it on sale), you can have one of the best consumer routers (DLink DGL-4300) and it should handle the FIOS connection just fine. Get it!

Thanks for the corection. Re the DGL-4300, -- I got one for the gigabit switch a long time ago, and have posted about owning one a few times already. I don't have FiOS, nor expect to get it hereabouts in the near future. I thought that the Linksys' performance was more intersting for the majority of readers, so haven't tested the DGL-4300 yet (not to mention that taking that out would upset my network / WAN far more than fooling around with the unused Linksys). Wouldn't it be "funny" if it wasn't as fast? Hehe...

I expect it to be fine; note that Verizon supplies some D-Link routers with their service (modified for their own access/diagnostics at least), and have a support note up about some other D-Link routers. It would be strange for the "higher end" DGL-4300 to not work in this case. To be honest though, although unlikely, I consider it a theoretical possibility that the DGL's more complex processing / feature set might make the WAN speed slower beyond the expected max consumer internet speed (which would be less than 10 Mb/s).

I also find that IPSec kills my bandwidth -- bringing gigabit on decent computers down to fast ethernet speed or so. I did these tests some time ago, and posted them, but don't recall in detail. It was too horrific. Really bad at high speed, not so bad when the speeds are lower.

Maybe stelleg151 can try the DGL-4300 out for us on FIOS.

 

[Goosemaster]

Originally posted by: nweaver
do the v1-4 routers support SNMP? Perhaps a quick test to see CPU/memory load while testing these. Solar winds has an easy to use windows demo that will show near realtime CPU load, or MRTG would work as well.

try looking up the various hacked firmwares....one ofthem probably supports it

 

[skyking]

Originally posted by: Goosemaster

Originally posted by: nweaver
do the v1-4 routers support SNMP? Perhaps a quick test to see CPU/memory load while testing these. Solar winds has an easy to use windows demo that will show near realtime CPU load, or MRTG would work as well.

try looking up the various hacked firmwares....one ofthem probably supports it

DD-WRT v22 and v23 support SNMP, I don't know about the other firmwares.

 

[Goosemaster]

Originally posted by: skyking

Originally posted by: Goosemaster

Originally posted by: nweaver
do the v1-4 routers support SNMP? Perhaps a quick test to see CPU/memory load while testing these. Solar winds has an easy to use windows demo that will show near realtime CPU load, or MRTG would work as well.

try looking up the various hacked firmwares....one ofthem probably supports it

DD-WRT v22 and v23 support SNMP, I don't know about the other firmwares.

<---using dd-wrt v23.something:thumbsup:

 

[stelleg151]

So I guess I will try doing a test like Madwand1, although it seems possible that the bandwidth measuring program does not stress the router as much as BT and a Gameserver could.

If the D-Link cannot handle it, I plan on using a P3 1Ghz as the router/fileserver.

As for IPSec, Im not sure I see the need for it. I was under the impression that NAT makes everything behind the router basically completely secure from a networking standpoint.

Also, is it possible to tell once the network is setup if slowdowns are occuring due to the router or due to bandwidth, because I have always noticed slowdowns when using BT, and I just assumed it was the ISP bandwidth.

Thanks for all your help guys, this is an informative thread.

Originally posted by: JackBurton
Maybe stelleg151 can try the DGL-4300 out for us on FIOS.

Heh, yeah maybe, once FIOS is in Seattle...

One note about the DGL-4300, it looks identical to my current router, except for the color.

I am suspicious that it is the same router with a new software to give gaming ports priority.

 

[Goosemaster]

Originally posted by: stelleg151
So I guess I will try doing a test like Madwand1, although it seems possible that the bandwidth measuring program does not stress the router as much as BT and a Gameserver could.

If the D-Link cannot handle it, I plan on using a P3 1Ghz as the router/fileserver.

As for IPSec, Im not sure I see the need for it. I was under the impression that NAT makes everything behind the router basically completely secure from a networking standpoint.

Also, is it possible to tell once the network is setup if slowdowns are occuring due to the router or due to bandwidth, because I have always noticed slowdowns when using BT, and I just assumed it was the ISP bandwidth.

Thanks for all your help guys, this is an informative thread.

Originally posted by: JackBurton
Maybe stelleg151 can try the DGL-4300 out for us on FIOS.

Heh, yeah maybe, once FIOS is in Seattle...

One note about the DGL-4300, it looks identical do my current router, except for the color.

I am suspicious that it is the same router with a new software to give gaming ports priority.

1. a p3qghz will be FINE. I would ordinarily recommend astaro security linux to run on it. It is a full-featured suite that is free to use and has a very inexpensive $49/year home license available for updates which are crucial. Since it has Intrusion prevention and detection services, not to mention spam services running as well, I would defintiely buy a license after getting it up and running and familiarizing yourself with it.

2. IPsec is a very robust yet complicated VPN (virtual private network) protocol. You don't even seem to be remotely familiar with it so stay away.

3. NAT is a term used to describe IP masquerading. Using 1 public IP for example, many private IP address can masquarade as that one public one, and appear to be transmitting and recieving at that one address. As far as security goes, NAT provides you with a veritable wall that doesn't allow unsolicited traffic onto your network. Basically, inbound traffic is always blocked and many times, on cheaper units, all outbound traffic is allowed. Using a attributes called 'ports' you can explicitly allow traffic to enter using these ports, and send that traffic to a specific IP address within your network. This allows for regulated inbound and outbound traffic. As for the security aspect of it, yes it provides inbound annonymity provided that the computer isn't making solicitation requests which results in a vulnerable computer. With NAT, one can createa ruleset that creates inbound and outbound paths that can allow traffic to share that public ip address rather seemlessly. In addition to NAT one would use a firewall to create rulesets that govern the type and content of the actual data going in and out, and the capabilities of the firewall governs how critical firewalls can be in their scrutiny.

It is defintiely a GREAT step up from directly connecting yoru PC to the net, but by itself, NAT isn't much more than the outer gate of a fortress.

As for slowdowns, you will need to check to see what the CPU usage is on the router. Honenstly that really depends on the router lodel, firmware, and its native capabilities.

Finally, you might want to hold off on astaro right now. It is fantastic, just as a Cisco router is, but like a cisco router, it has a steep learning curve that requires great familiarity with networking terminology and topological planning.

For now, I recommend Clarkconnect or smoothwall. Both are GREAT and simple products that will msot, if not everything that you need for now

Bsst of all, all of these products always let you check traffic/bandwidthusage/cpu usage/firewall logs in realtime.

 

[stelleg151]

Originally posted by: Goosemaster

Originally posted by: stelleg151
So I guess I will try doing a test like Madwand1, although it seems possible that the bandwidth measuring program does not stress the router as much as BT and a Gameserver could.

If the D-Link cannot handle it, I plan on using a P3 1Ghz as the router/fileserver.

As for IPSec, Im not sure I see the need for it. I was under the impression that NAT makes everything behind the router basically completely secure from a networking standpoint.

Also, is it possible to tell once the network is setup if slowdowns are occuring due to the router or due to bandwidth, because I have always noticed slowdowns when using BT, and I just assumed it was the ISP bandwidth.

Thanks for all your help guys, this is an informative thread.

Originally posted by: JackBurton
Maybe stelleg151 can try the DGL-4300 out for us on FIOS.

Heh, yeah maybe, once FIOS is in Seattle...

One note about the DGL-4300, it looks identical do my current router, except for the color.

I am suspicious that it is the same router with a new software to give gaming ports priority.

1. a p3qghz will be FINE. I would ordinarily recommend astaro security linux to run on it. It is a full-featured suite that is free to use and has a very inexpensive $49/year home license available for updates which are crucial. Since it has Intrusion prevention and detection services, not to mention spam services running as well, I would defintiely buy a license after getting it up and running and familiarizing yourself with it.

2. IPsec is a very robust yet complicated VPN (virtual private network) protocol. You don't even seem to be remotely familiar with it so stay away.

3. NAT is a term used to describe IP masquerading. Using 1 public IP for example, many private IP address can masquarade as that one public one, and appear to be transmitting and recieving at that one address. As far as security goes, NAT provides you with a veritable wall that doesn't allow unsolicited traffic onto your network. Basically, inbound traffic is always blocked and many times, on cheaper units, all outbound traffic is allowed. Using a attributes called 'ports' you can explicitly allow traffic to enter using these ports, and send that traffic to a specific IP address within your network. This allows for regulated inbound and outbound traffic. As for the security aspect of it, yes it provides inbound annonymity provided that the computer isn't making solicitation requests which results in a vulnerable computer. With NAT, one can createa ruleset that creates inbound and outbound paths that can allow traffic to share that public ip address rather seemlessly. In addition to NAT one would use a firewall to create rulesets that govern the type and content of the actual data going in and out, and the capabilities of the firewall governs how critical firewalls can be in their scrutiny.

It is defintiely a GREAT step up from directly connecting yoru PC to the net, but by itself, NAT isn't much more than the outer gate of a fortress.

As for slowdowns, you will need to check to see what the CPU usage is on the router. Honenstly that really depends on the router lodel, firmware, and its native capabilities.

Finally, you might want to hold off on astaro right now. It is fantastic, just as a Cisco router is, but like a cisco router, it has a steep learning curve that requires great familiarity with networking terminology and topological planning.

For now, I recommend Clarkconnect or smoothwall. Both are GREAT and simple products that will msot, if not everything that you need for now

Bsst of all, all of these products always let you check traffic/bandwidthusage/cpu usage/firewall logs in realtime.

Awesome, thanks.

I would like to start with something simple for sure, and maybe even stick with it if it works. I actually have never had any problems with the functions/features of my consumer routers, but I would like some more power if its needed.

As far as security goes, I figure a basic firewall with only a few ports open for certain games/ torrents should do me just fine.

 

[Madwand1]

There's a recent article on THG on this topic, where they provide upload and download benchmarks for a number of routers. As hoped for / expected here, the D-Link DGL-4300 comes near the top, and various Linksys WRT's do fairly well. Their numbers don't match mine, but there's likely a difference in methodology / tools / hardware (device under test and ancillary) which affects that (20 Mb/s to my 16 Mb/s download speed with a WRT45G).

One anomaly is the Asus, which is the only GbE capable router here besides the DGL-4300 -- it has much worse download performance, while its upload performance is good.

http://www.tomsnetworking.com/2006/03/31/hardware_router_chart/

 

[luh3417bis]

I have a D-Link DGL-4300 and a 10 megabit (both directions) fiber pipe. The D-Link DGL-4300 maxes out at about 1.8 megabits/sec up the WAN. I had hoped, based on the THG review, and other optimistic posts above, that it would do better than this. Why did they bother putting a 10/100 port on the WAN since it can't even come close to filling the 10.

My test is to run eMule. This machine fills the upstream pipe (a megabyte a second, sweet!) if I put it straight into the fiber box. I tried the 1.6 firmware. I tried putting the machine in the DMZ. I've contacted D-Link tech support to see if they have any ideas. The only good news is so far at least the router doesn't crash or lock up like the SMC Barricade did. Note also that the gigabit LAN does not support jumbo frames.

If anyone has any other ideas I'm game.

 

[Goosemaster]

Originally posted by: luh3417bis
I have a D-Link DGL-4300 and a 10 megabit (both directions) fiber pipe. The D-Link DGL-4300 maxes out at about 1.8 megabits/sec up the WAN. I had hoped, based on the THG review, and other optimistic posts above, that it would do better than this. Why did they bother putting a 10/100 port on the WAN since it can't even come close to filling the 10.

My test is to run eMule. This machine fills the upstream pipe (a megabyte a second, sweet!) if I put it straight into the fiber box. I tried the 1.6 firmware. I tried putting the machine in the DMZ. I've contacted D-Link tech support to see if they have any ideas. The only good news is so far at least the router doesn't crash or lock up like the SMC Barricade did. Note also that the gigabit LAN does not support jumbo frames.

If anyone has any other ideas I'm game.

looks like that device has a ceiling of 24Mbps before packet loss ensues. Have you tried turning off QoS, port filtering and everything else?

Basically try to turn off as much as you can.

In addition, the switch might be causing issues. I am guessing you have a gigabit NIC, right? Basically, if the NIC, in its propterties, is set for autonegotiation, change it to 100 or 100FD or if it is set to that, set it to autonegotiation. I have found that certain card +switch combinations act strangely.

 

[Madwand1]

eMule is a file sharing app, right? A straight-ahead bandwidth measuring tool would probably be better to use for analysis to factor out the different components of performance. E.g. TTCP.