0

If a copy a file from my desktop to my laptop, the speed is usually 3 or 4 times faster than if I go in the other direction.

Desktop is connected to the router by a cable, laptop is wifi.

If I connect the laptop with a cable, speeds are fast both directions.

Router is a wrt54gl with dd-wrt. Both machines running Win7

Why would this happen? How can I identify the problem and fix it?

foosion
  • 461
  • 7
  • 16
  • How do you copy it? SMB? SCP? Does one side have a much larger/faster random pool? Do the speeds change when you use wired connections for both (if not then you know it has nothing to do with Wireless). Etc etc. – Hennes Jul 21 '13 at 15:37
  • Windows machines at both ends, so using various windows utilities. How should I test speed? What's a random pool? Wired on both sides is MUCH faster. – foosion Jul 21 '13 at 15:45
  • 1
    You might want to edit your question... for as it stands, you say " from my desktop to my desktop" but further on in the question, you say "Desktop is connected to the router by a cable, laptop is wifi." Which is it? Are you talking about two desktops, or one desktop and one laptop? Also, are both computers running the same Operating System? If it is indeed a laptop, have you tried connecting it to your router with a network cable, and checking the transfer speed that way? – Bon Gart Jul 21 '13 at 15:50
  • SCP and other encrypted links need random information for their encryption. Both on Windows and other operating systems this can exhaust the available supply of random numbers and thus slow down. (No idea if SMB/samba uses random, but I know sFTP and SCP do). -- As to wired is much faster. Aye. Generally I expect 100mbit wired to outperform 300mbit wireless. – Hennes Jul 21 '13 at 15:50
  • @Hennes: **Neither** protocol requires a lot of cryptographically secure randomness. At most, SSH and SSL need **only a few hundred bits** – for things like creating a *static* session key that's used for the entire communication. (This applies to SFTP and SCP which use SSH as a transport.) SMB doesn't even need that; it lacks encryption. – u1686_grawity Jul 21 '13 at 15:53
  • Also... and this might seem trivial... but when you transfer from the laptop to the desktop... are you starting the transfer from the laptop, or are you starting it from the desktop? You should try it both ways (Sit at the laptop, browse to the desktop shared, select the file, copy it, and paste it somewhere on the laptop.... THEN after it is finished sit at the desktop, browse to a file you want to transfer, select and copy it, browse your network to the shared laptop directory, and paste it). See if those procedures both result in slow transfers – Bon Gart Jul 21 '13 at 15:57
  • @Bon Gart: Oops! Corrected – foosion Jul 21 '13 at 16:09
  • @grawity I know smb can be encrypted or not. No idea how much encryption it uses. I never even thought about this stuff until I noticed pocketPutty starting much faster on my smart phone when I performed random input on the screen. (Windows CE5, a phone from 2005, so probably a lot more limiting than any modern desktop). – Hennes Jul 21 '13 at 16:14
  • This is just a guess, but perhaps your laptop cannot transmit as fast as it can receive? The router has a higher power antenna and so can transmit at a faster speed than the laptop can transmit. (The relation between the speed and power being that higher power = less errors = less retransmissions = higher speed) – Akash Jul 21 '13 at 16:30
  • @Akash when the laptop is next to the router, read and write speed are approximately equal. At a normal distance, read speed is unchanged, but write from laptop to desktop speeds plummets. – foosion Jul 21 '13 at 20:23
  • @foosion Then my guess makes even more sense. From further away, the transmission requires more power to maintain the same speed. The laptop cannot provide the extra power, the router can. Similar to how a phone goes out of range before a laptop does usually – Akash Jul 22 '13 at 04:06
  • @Akash, any suggestions for testing this (that don't require special testing equipment)? Also, any idea why we don't see more reports of this issue, given the number of people who likely are in the same situation? – foosion Jul 22 '13 at 16:16

3 Answers3

1

I can say this. Your problem isn't isolated to your situation. Apparently people have been posting about this very same issue for a few years now.... IE slow network transfers in one direction only across wireless.

What I'm finding, is that there appear to be different solutions, and that most appear to be specific to the hardware. For example, from smallbuilder.com...

So, after trying a few things, I set "Enable HW Accelerator" to Off under LAN - Switch Control and now I get 1.5MB/s in both directions!

This previous SU question resulted in a working solution for someone OTHER than the person who posted the question...

I went to the configuration screen of my network adapter and changed the following configuration settings:

  • Large Send Offload V2 (IPv4) = Disabled (was Enabled)
  • Large Send Offload V2 (IPv6) = Disabled (was Enabled)

A solution from a Microsoft Technet post, but no news on whether it was successful...

Try disabling Receive Window Auto-Tuning:

  • 1) Go to Start and type cmd.

  • 2) Right-click on cmd and select “Run as administrator”.

  • 3) Type: netsh interface tcp set global autotuninglevel=disabled and press Enter.

If you want to to re-enable it:

Type: netsh interface tcp set global autotuninglevel=normal and press Enter.

Also you can try configure the network card :

  • 1) Click the Start Button, type "devmgmt.msc" (without quotation marks) in the Start Search box and press Enter.

  • 2) Double click to expand "Network Adaptors".

  • 3) Right click your network card and click Properties

  • 4) Click Advance tab. High light Speed & Duplex

  • 5) If you would like to use the full functionality, please set the Value to the highest Full.

  • 6) Click OK.

There were also some smaller posts where people fixed their issues after finding that settings in their routers or network adapters had not been set to Full Duplex... that changing from an AUTO setting to a specific Full Duplex setting resulted in equal (and faster) network transfer speeds.

Sorry I don't have one specific answer to the issue... there just doesn't seem to be a single fix.

Community
  • 1
Bon Gart
  • 12,950
  • 1
  • 25
  • 36
  • Changing the desktop ethernet adapter to full duplex (from auto) helped with write to desktop speed when the notebook was close to the router, but not otherwise. Changing the router ports to full duplex (from auto) improved write to desktop speed a little, but there's still a large spread. Large send offload made no difference. Netsh resulted in an error message. I don't have Enable HW acceleration. – foosion Jul 21 '13 at 20:16
  • What speed does your laptop say it is connected at? What kind of wireless security are you using? Does the problem persist if you disable that security? – Bon Gart Jul 22 '13 at 02:07
  • Connection speed varies. 54Mbs is the most common, but it can go as low as 5.5. It appears to be more variable and to go lower while testing upload/download speeds and shortly afterwards. Even with connection speed bouncing around, read (download) is much faster than write (upload). Download was 11-12Mbs in most recent test, while upload was 2.5-6.0 (tested using a program which writes and reads a 20Mb file). Security is WPA2 personal - AES. Disabling doesn't help. – foosion Jul 22 '13 at 10:04
  • BTW, on wifi, I can read from the internet (per speedtest.net) slightly faster than I can read from the desktop on my LAN. – foosion Jul 22 '13 at 10:30
  • The connection speed between the laptop and the router should be more constant... it shouldn't bounce up and down like *that*. Granted, if you never saw very fast connection speeds to the router at all, it would bounce... but that's usually when a 54g card can't seem to connect faster than 10-15 max... but if you can hit 54, and that's the max for the WNIC, it shouldn't bounce unless there's an issue with the antenna, or material between the laptop and the router, or interference on the channel (change the channel you are using) like a 2.4ghz cordless phone or a nearby microwave oven etc – Bon Gart Jul 22 '13 at 13:39
  • Speed does not seem related to signal strength (as measured by inSSIDer). It seems there is some relation between connection speed (as reported by Windows connection status) and read/write testing, although that could be a coincidence. I'm in an apartment building, and typically can detect 20 APs, so who knows what's interfering. I get about the same speeds at times people are unlikely to microwave or talk on the phone. I'm also getting about the same read/write speed at an RSSI of -25 (next to router) and -45 (farther away on other side of wall). – foosion Jul 22 '13 at 16:06
  • Also, why would interference, signal strength, etc. affect upload speed but not download speed? – foosion Jul 22 '13 at 16:11
  • Not sure exactly why speed would be affected in one direction but not another, but if there are a large number of APs in close proximity, it could indeed be that one or more is interfering with data transmission. – Bon Gart Jul 22 '13 at 23:12
0

Your WiFi connection simply does not have the bandwidth of the wired connection, even at a 10/100 ethernet port. More modern routers offer faster wireless connections than the WRT54G, plus gigabit ethernet. See: http://community.linksys.com/t5/Wireless-Routers/WRT54GL-Is-wireless-faster-than-20Mbps-possible/m-p/517359?comm_cc=HSus&comm_lang=en#M227951

JDLowe
  • 1
  • 1
  • 1
    That does not explain why it's faster in one direction than the other. In both cases it's using wifi for the laptop - router connection. – foosion Jul 21 '13 at 15:43
0

This is part of the nature of the way access points work. Access point to client transmissions will always be more efficient than client to access point transmissions.

A Wifi network wouldn't work unless every node could reach the access point. Thus the access point can transmit very reliably, because it is guaranteed to hear all other nodes. However, client nodes frequently can't hear each other reliably. So it's much more likely for the laptop to have its transmissions stepped on, and thus need to repeat them. This greatly reduces the available bandwidth from the client to the access point.

David Schwartz
  • 61,528
  • 7
  • 100
  • 149