1. Computer problem? Tech Support Guy is completely free -- paid for by advertisers and donations. Click here to join today! If you're new to Tech Support Guy, we highly recommend that you visit our Guide for New Members.

What's Better: 80 mbps Download Speed Or "Buffer Bloat"

Discussion in 'Networking' started by LettuceChat, Jun 15, 2016.

Thread Status:
Not open for further replies.
Advertisement
  1. LettuceChat

    LettuceChat Thread Starter

    Joined:
    Aug 27, 2002
    Messages:
    67
  2. zx10guy

    zx10guy Trusted Advisor Spam Fighter

    Joined:
    Mar 30, 2008
    Messages:
    5,956
    I'm not sure how an online tool can measure a network device's buffer size. The whole discussion about bufferboat is over various networking gear having large cache/buffet spaces to guard against dropped packets in the event of congestion on the network. SOHO and even most if not all SMB networking devices have such small buffers it's not even a worry.

    Enterprise level switches can have up to 2 Gigs of buffer space but this is shared across all ports on the device. Where as SOHO devices will typically have a few KB of buffer. Certainly huge orders of magnitude difference. In addition, enterprise deployments will utilize WAN optimization devices which make bufferbloat a non issue.
     
  3. richb-hanover

    richb-hanover

    Joined:
    Jun 16, 2016
    Messages:
    4
    This is a misperception... RAM is cheap, and SOHO devices routinely have megabytes of buffering (which translates to seconds (not msec) of latency unless the router is controlling queue lengths.)

    It doesn't: www.dslreports.com/speedtest measures the actual latency when the link is heavily loaded in each direction. It basically does http pings during the download/upload speed tests to see the change in latency. (Try it out on your home router.)

    Bufferbloat is undesirable latency caused by a router buffering too much data. It's the major reason that people complain, "The internet is slow today."

    The team at www.bufferbloat.net has been working on this buffering/latency/lag problem. Their solution - fq_codel - controls queue lengths for all the flows of data, reducing latency for all. It has been available for four years in the Linux kernel, in OpenWrt/DD-WRT/other home routers. For a detailed description of bufferbloat, see my blog at http://richb-hanover.com/bufferbloat-and-the-ski-shop/

    Response to OP:
    1. Which is better - 280 mbps or good latency? You have to decide what you need from your current network setup. If you game, or use VoIP, then a grade of F from dslreports.com means you'll get lagged out, your call quality will be terrible, and the network will just feel draggy whenever someone's uploading/downloading.
    2. The fq_codel algorithm/queue discipline of OpenWrt will top out around 60-80 mbps on older routers such as your WNDR4000.
    3. A new optimized qdisc - CAKE - is available for new builds of OpenWrt/LEDE (www.lede-project.org) that handles traffic considerably faster. CAKE is new enough that we don't have a good characterization of its top speeds, but come help us!
     
    Last edited: Jun 16, 2016
  4. zx10guy

    zx10guy Trusted Advisor Spam Fighter

    Joined:
    Mar 30, 2008
    Messages:
    5,956
    RAM is cheap but the logic behind how the network device allocates this space is not. There's also tiering of RAM as not all RAM is created equal. Some RAM can operate at a faster speed/latency and is usually reserved for cache/buffer type applications.

    Here's a Cisco article on how complex this division of buffer space is on both the physical port level and when system memory is utilized to allocate additional space needed to prevent packet drops:

    http://www.cisco.com/c/en/us/produc...quality-of-service-qos/sol_ov_c22-708224.html

    My statement above was more directed to the OP's use of a SOHO level switch. That switch per Netgear's specs shows it having 1MB of buffer memory for all 8 ports. Divided up that's about 128KB of buffer space. So tell me where I'm wrong?

    And how many home users are going to get into tweaking the actual OS/firmware of a router? Also, per your explanation, you would think the bufferbloat would be in reverse in terms of the rating. Explain to me why going through a switch which has no real sizeable buffer space would rate an F compared to the router the OP used with 64 MB of RAM?
     
  5. richb-hanover

    richb-hanover

    Joined:
    Jun 16, 2016
    Messages:
    4
    I agree that it's hard to understand the OP's observations without a more complete description of the network setup (a SOHO switch isn't really an equivalent replacement for a WNDR3800 router...) What else is in the picture: a cable modem? FTTH? Enterprise network? Could any of these be buffering data?

    A good hands-on measurement of latency/lag is to ping a remote host while running a speed test. This is described more completely at A Quick Test for Bufferbloat - use the second one on that page (after DSLReports.com and ThinkBroadband.com)

    If the OP is still interested in pursuing this, I would be interested to hear:
    1. the results of the bufferbloat test (upload/download speeds and average ping times when idle, during download, and during upload)
    2. more about the full network infrastructure.
    Thanks!
     
  6. Sponsor

As Seen On
As Seen On...

Welcome to Tech Support Guy!

Are you looking for the solution to your computer problem? Join our site today to ask your question. This site is completely free -- paid for by advertisers and donations.

If you're not already familiar with forums, watch our Welcome Guide to get started.

Join over 733,556 other people just like you!

Loading...
Similar Threads - What's Better mbps
  1. OM2
    Replies:
    4
    Views:
    290
  2. koyezx
    Replies:
    5
    Views:
    319
Tags:
Thread Status:
Not open for further replies.

Short URL to this thread: https://techguy.org/1172963

  1. This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
    By continuing to use this site, you are consenting to our use of cookies.
    Dismiss Notice