Re: Throughput riddle
On Fri, 18 Mar 2016 12:44:30 -0600 (MDT)
"John L. Ries" <jries@salford-systems.com> wrote:
> I don't know if it will help, but I hook up my Iomega NAS directly to my
> desktop machine with a regular cat 5/6 cable (each has two gigabit
> Ethernet ports, so each can connect to the rest of my network, as well
> as to each other) and that seems to help the throughput by a lot (but I
> don't have any numbers for you). So if your NAS has an extra Ethernet
> port, you might want to hook it up to your laptop when you're in the
> same room with it and use your wifi interface to connect to your
> network. Certainly, you should avoid connecting to your NAS over wifi
> if you're using it heavily, as that will definitely slow things down (it
> seems that a lot more handshaking is required to connect through the air
> than through a physical cable).
Thanks. Currently, the NAS is used only as a backup target, so it's not
a big deal - I'm mostly just frustrated and curious ...
> On Friday 2016-03-18 10:48, Celejar wrote:
>
> >Date: Fri, 18 Mar 2016 10:48:24
> >From: Celejar <celejar@gmail.com>
> >To: debian-user <debian-user@lists.debian.org>
> >Subject: Throughput riddle
> >
> > Hi,
> >
> > I'm trying to understand the throughput across the different links of
> > my little home network, and am perplexed by the measured wireless
> > throughput.
> >
> > The three main devices I'm interested in:
> >
> > Router: Buffalo WZR-HP-G300NH running OpenWrt (Chaos Calmer 15.05).
> > Gigabit WAN and LAN, 802.11bgn wireless.
> >
> > https://wiki.openwrt.org/toh/buffalo/wzr-hp-g300h
> >
> > Laptop: Thinkpad T61 running Jessie 8.3. Gigabit ethernet, 802.11abgn
> > wireless.
> >
> > NAS: Seagate GoFlex Net [STAK100] runninng Debian Jessie 8.3.
> >
> > https://archlinuxarm.org/platforms/armv5/seagate-goflex-net
> >
> > All throughput measurements taken with iperf (run three times and using
> > the median result), unless specified otherwise. These first results are
> > with the laptop connected to the router via cat5:
> >
> > Laptop - NAS: ~874 Mbps.
> >
> > I suppose this is close enough to the gigabit theoretical max, and there isn't
> > any significant bottleneck.
> >
> > Router - NAS: ~217 Mbps
> > Router - laptop: ~198 Mbps
> >
> > Here the router CPU is apparently the bottleneck (top shows close to
> > 100% CPU utilization by iperf for at least part of the 10 second iperf
> > runs). I suppose that this is due to the bits needing to be copied out
> > of the kernel networking stack into iperf's userspace memory, or
> > something like that. I don't understanding why the NAS seems to be
> > doing better, but I suppose it could be an artifact of the data.
> >
> > Here's the part that baffles me - these are with the laptop connected
> > to the router wirelessly:
> >
> > Laptop - router: ~11.8 Mbps
> >
> > These numbers actually exhibit significant variance, but they're
> > generally at least this much, and at most about 15-20 Mbps.
> >
> > Laptop - NAS: ~14.7 Mbps
> >
> > Once again, these numbers vary widely, but are in line with the laptop
> > - router numbers.
> >
> > But here's the kicker: Ookla's speedtest (run on the laptop with
> > speedtest-cli) shows 29.01/5.89 (d/u), and this is fairly consistent.
> > I'm paying Comcast for 25/5, and they apparently provision at
> > 31.25/6.25, so I'm getting quite close to the theoretical max, even
> > when the laptop is connected to the router wirelessly. Additionally,
> > various Android phones also get close to the Comcast provisioned max
> > when connecting wirelessly to the router.
> >
> > So the wireless link can apparently push at least 30 Mbps or so, so why
> > are my local wireless throughput numbers so much lower?
> >
> > I was originally using one of the common 1/6/11 channels, and I switched
> > to 3 since I saw a lot of other stations on those channels. This may
> > have resulted in some improvement, but I'm still stuck locally as
> > above. What's the explanation for this - how can I possibly be getting
> > much better throughput to servers tens of miles away than to my local
> > stations? Does iperf somehow work fundamentally differently from
> > speedtest? If so, which is a better representation of actual throughput?
> >
> > Celejar
> >
> >
Celejar
Reply to: