How to maximize LAN speeds

[nargalzius]

I know that no matter what ISPs say, that general throughput is lousy in the Philippines.

But my question is for a LAN environment, wherein your routers, nics, and settings determine the speed:

All PCs here are on 10/100 and are set on forced 100Tx full duplex. MTUs are set to 1492 or 1500 (the cap on the router) - which supports 10/100 wired and 54megabit (G) for WLAN

Wireless protable is this albook. The router wlan is set to "G only." And it's obviously working since I'm well past the 11mbit (B) throughput on filetransfers on this sucker

But when i fired up an app like Granet, it measured filtransfers only up to 20-25mbps... what's up with that!? G is supposed to support 54mbps.

I first noticed this when I tried watching video over LAN (via VLC) - which, on a pc setup, worked fine. On the powerbook however, i get the dreaded dropped frame error, suggesting either the cpu's too slow (or the network isn't catching up). I know the former isn't the case, since the files play fine when I copy them locally to the powerbook.

Am I missing any other setting here? (i've tried numerous tweaks on both pc and mac, and throughput still caps out at about 20-25mbps)

Any help would be appreciated.

P.S.
Anybody know an app similar to Granet - but DOESN'T look like $#1t?

[Edited on 10-31-2004 by nargalzius]

[gaol]

Actual throughput speeds for wireless networking are much lower than their theoretical maximum. For standard 802.11g, what you're getting is actually pretty good, as good as it gets.

Linksys, DLink etc have proprietary tehnologies that boost speeds beyond the the standard 802.11g, but these only work if you stick with their own brand for access point/router and the wific cards.

Network standards at a glance (from CNET.com)
Standards > Actual speed > Range
802.11b > 5Mbps (vs. 11Mbps claimed) > 150 feet
802.11a > 22Mbps > 100 feet
802.11g > 20Mbps (vs. 54Mbps claimed) > 150 feet
Dual band > 22Mbps > 150 feet
Bluetooth > 500Kbps > 30 feet

[Edited on 10-31-2004 by gaol]

[nargalzius]

IC, what about if you try changing the firmware to the unofficial "tweaked" ones?

I saw some on the net, but if it doesn't really affect the throughput, then I don't think I even want to waste my time trying them out and having to redo securing the wlan.

Has anyone been using firmware for linksys that isn't officially supported by it? What are your experiences with this?

[Buddhaboy]

I've used Sveasoft Satori as well as Alchemy on my Linksys WRT54G. Satori worked quite well, Alchemy however seems to drop connections quite a bit. I got Satori to work with my Airport Express. I'm sure Alchemy would work too, but I haven't tried it. The feature to increase the transmitter power works very well too.

I tried using HyperWRT as well but I couldn't get a connection on myDSL.

[nargalzius]

Cool, i'll check them out.

I've tried HyperWRT though, and it worked for me (for the wired lan). but at that time the wifi didn't work so I reverted to the official firmware.

Only later did I find out the reason i couldn't wlan was because my wireless digital headphones were operating on the same 2.4Ghz frequency (and was overpowering the internet signal)

i'm curious though to what thoughput you got when using the hacked firmware? If it's still the same (22mbps), then I probably don't have to fix what ain't broke

[Edited on 11-1-2004 by nargalzius]

[nicomendoza]

I've notoriously experienced dropped packets a lot (or lagtime) on wireless routers. I think this is generally the case with WiFi routers particularly with interference and signal degradation. Throughput isn't consistent (and that's why we don't reach full speed) because of these re-transmits, and for data it only causes slower data, but real-time streaming is a different story (even if you average the speed, streaming has another factor - being real time). This is analogous to skips in CD players.

Quick resolution: increase buffer time.

[ginoledesma]

There are several factors that affect overall network performance, but the bottomline is that you shouldn't expect wonders from cheap networking devices. The reason expensive network devices exist is that they're guaranteed to deliver whatever claims they have.

An example: someone asked me why some years ago I purchased a P38,000 16-port switch when I could have bought a 24-port or 32-port one for less than half the price. The switch in question is an Intel 410T. The answer: huge backplane (memory fabric). That's what makes core switches handle extremely high loads of traffic.

The cheap devices we buy for home use have puny backplanes, memory buffers, and use standard optimization/caching/look-ahead routines. You might argue that a 2-PC on a cheap switch shouldn't bog it down, and here comes the other factors.

Cheap network cards are also like cheap switches: they work well for the most part but will not give you guaranteed, high performance. I've been a fan of the 3Com 3C90X0-series of cards, but was even more impressed with Intel's EtherExpress Pro or Pro/100+ series of cards. Those could really push packets. Again, the cheap ones from Linksys, D-Link, Surecom, and others usually rely on the common chipsets -- RTL81x9, SMC, Broadcom, SiS, etc. None of these have the extra boost offered by 3Com's ParallelTasking stuff or Intel's version of it.

Assuming all things equal, we should still be getting higher throughput than what we are currently getting, and here's yet another factor: the Operating System.

On the Intel 410T switch I used above, I did some benchmarks using a PowerMac G4/400 running Mac OS X (built-in 10/100 Ethernet) with a PC equipped with a high-performance NIC. The G4 could push about 8.1MB/s using FTP, SMB, or NFS, which is quite good since the maximum theoretical limit is 12.5MB/s. On Linux, I could get 8.8MB/s, almost 9MB/s during peaks. On Windows, I could get 8.5MB/s.

Yet still there are other factors involved:
- network cabling (Cat5 vs Cat5e vs Cat6) -- can affect greatly or none at all
- application used
- environment

As for wireless environments, I've not had a chance of using "high performance" Access Points (if there are any), but using my Netgear WGR614v4, I've hit the 2.8MB/s barrier (which translates to about 22.4Mbps). A Linksys WTR54G gives me the same. A D-Link DI-624+ gives me 2.7MB/s.

I'm guessing this has to do with the Airport Express card, which uses the Broadcom chipset. It's no secret that Apple's network cards aren't the best and fastest in the world.

[Maccess]

For the wired ethernet connections, it can't hurt to double check the crimping and whether or not you've followed good cabling practice. The following link has a complete guide to ethernet cabling.


http://www.bluemax.net/techtips/networking/Wiring_Tips/Wiring100TX/100TXContents.htm


I've seen a lot of networks with poor performance caused by improperly crimped and routed cables.

Many PC shops that crimp their own cable do so incorrectly, mixing up the pairs.

Ethernet Cable should be paired as follows:

Pair Pin
Orange/White 1
Orange 2

Green/White 3
Green 6 <=== NOTE THIS!!

The jump is because the blue pair uses pins 4 & 5! Only the Orange and Green Pairs are used by Ethernet!

The blue pair is where it is so that is can also be wired to carry a telephone signal (if you plug in a 2-conductor RJ-11 phone plug it will contact the blue pair for telephone use -- do not use a four conductor RJ-11 phone plug in an RJ-45 socket)

Blue 4
Blue/White 5

Note that the Blue/Stripe pair order is reversed compared to the other pairs. This is so the cable has a consistent stripe/color/stripe/color sequence.

The brown pair is connected as follows:

Brown/White 7
Brown 8

Nothing specific is assigned to the brown pair. You can pretty much use it for whatever you want. Some uses I've seen:

1. A second phone line
2. Apple's Localtalk network system
3. Power over ethernet (note that you should make sure that the power is separated from the line that connects to the ntwork device).

A crossover cable interchanges the Orange and Green pairs only, at ONE end of the cable.

Incorrectly wired cables will work over short distances, and may work up to 10 - 20 feet, but they will work slowly and impose a heavy reactive load on the network card and hub/switch. This may lead to premature failure of the port.