[PLUG] Network Transfer Speed Bottleneck Identification

Nathan W nathan at nathanewilliams.com
Sat Sep 11 08:53:17 UTC 2010


> 54Mb/s for 802.11G is the "theoretical maximum". There's a few things that
> can affect transfer speeds:
> 
> 1. Signal quality / Interference - What is the signal quality on both
> clients?
> When you scan the wireless networks from your clients do you see other Wi-Fi
> APs on the same channel? If there are many, you set your Wi-Fi AP to a
> different channel.
> 
> 2. Distance - How far away are the clients from the Wi-Fi AP? With 11.G,
> after 100' the throughput drops off significantly to the point where it's
> not all that much faster than "B".
> 
> 3. Do the Wi-Fi cards in both computers support "G"? - If the Wi-Fi AP
> supports 11.B & G, and a client connects at "B" it will force the other
> clients to connect at "B" as well.
> * If your seeing speeds of 6.5Mb/s, this almost sounds like what's happening
> as the theoretical max is 11Mb/s and so throughput of 6.5 - 8Mb/s is about
> on par with what you would see.
> 
> HTH!
your suggestions for places to check reminded me that i probably failed
at 'asking good questions', and gave me some thoughts as to possible
causes. thanks mike.

here's answers to your follow-up questions, and some additional
information.

1a). signal quality - 80-100% dependent on where in the apartment i am
1b). interference - being in an apartment, there's a ton of other
networks around, and if i fire up dd-wrt on my spare router to do the
scan (not sure how to see other networks channels in network mgr/cli), i
see that most are on the channel 6, so changing channels is definitely
something i will be looking into.

2). distance - there's probably a max of 30 feet that i stray from the
router, but there's almost always an internal wall between the laptop
and the router, which i credit for getting 80% signal strength 25-30
feet away in the living room.

3). network configuration - the two pc's involved are basically in a
server/client configuration, w/ the "server" being an old HP laptop
running debian testing, which shares a 1TB external HDD over NFS. the
server is connected to the router via ethernet. i've tried changing the
port on the router as i initially thought this may have been tied to a
failing port, and i changed the cable at the same time to no effect.

the laptop is a newer dell inspiron 1545, running Ubuntu 10.10, which
connects over a WPA2 wlan w/ mac filtering (not sure if this would
contribute to overhead). the laptop has the dreaded Broadcom card in it,
so i currently have to run the broadcom-sta driver, though normal
network speeds are fine, so i don't think it's an issue on the "client"
network hardware configuration

add'l info: 
"client" fstab entry - 92.168.1.105:/media/disk       /media/net
nfs     soft,intr,rsize=8192,wsize=8192
"server" exports entry - /media/disk
192.168.1.0/24(rw,async,no_subtree_check)

other thoughts: 
by changing the wlan channel and switching to async export, was able to
squeeze about another 100KB/s out of the connection, though still much
lower than i'd hoped. i set the router to g-only mode to eliminate the
possibility of a b-mode connection causing the slower speeds, so it
doesn't seem that's what's going on.

one possibility might be the USB read speed on this older laptop (circa
2000), though i'm not sure how to test read-speed except...<pause for
googling>

okay, head-desk. hdparm -t /dev/sdb shows me this fabulous output:
hdparm -t /dev/sdb

/dev/sdb:
 Timing buffered disk reads:    4 MB in  5.50 seconds = 745.40 kB/sec

which is right in range with the maximum speeds i'm seeing. seems i've
found the bottleneck. if anyone has pointers from here on ways to
improve this, i'd be stoked, though i have a feeling this may be due to
crappy old USB ports.

ty much for your assistance.

regards,

-- 
Nathan W <nathan at nathanewilliams.com>




More information about the PLUG mailing list