• Open Menu Close Menu
  • Apple
  • Shopping Bag
  • Apple
  • Mac
  • iPad
  • iPhone
  • Watch
  • TV
  • Music
  • Support
  • Search apple.com
  • Shopping Bag

Lists

Open Menu Close Menu
  • Terms and Conditions
  • Lists hosted on this site
  • Email the Postmaster
  • Tips for posting to public mailing lists
Re: What exactly is net.inet.tcp.sockthreshold?
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: What exactly is net.inet.tcp.sockthreshold?


  • Subject: Re: What exactly is net.inet.tcp.sockthreshold?
  • From: "Peter Sichel" <email@hidden>
  • Date: Mon, 3 Oct 2005 09:21:54 -0400

On 10/2/05, Chris McFarling wrote:

>In my case, two clients send to the server simultaneously over our LAN (all
>are on gigabit). The transfer rate of each client drops to ~6-8MB/s.
>Whereas, each individual client is capable of sending to the same server at
>~30-40MB/s. While there is obviously going to be a decrease in performance
>when two Gb devices are sending to one Gb device, I would expect a better
>transfer rate than 6MB/s.

Yes.  Since you previously observed that reducing the window size helped,
I inferred that perhaps ACKs are being delayed by other traffic.

>So anyway, based on the fact that the AFP server overrides the system
>default TCP window size by using SO_RCVBUF, is IPNetSentryX able to override
>that behavior of AFP?

Yes.  IPNetSentryX uses an NKE to examine and modify the packets
at the data link layer.  It does traffic shaping by modifying
the advertised window size of TCP ACK segments on the fly,
so it does indeed override the default behavior of AFP.
I've run a number of tests with AFP transfers and the results
are very consistent.  AFP opens a single TCP connection when
a remote volume is mounted and uses that connection for all
transfers.  One detail is to start IPNetSentryX before you
mount the remote volume so it can see the connection being
established.  Otherwise it has to guess and adapt itself to
determine the window scale factor (which is sent as a TCP
option on the SYN segment).

FYI: You can try IPNetSentryX free for 21-days.  The latest
candidate 1.3c8 includes the traffic shaping code.  I haven't
tested it at gigabit speeds yet, but as an NKE it should be
quite fast.  [I suspect an ipfw divert socket used by some
other implementations wouldn't help much at this speed.]

Kind Regards,

- Peter


 _______________________________________________
Do not post admin requests to the list. They will be ignored.
Macnetworkprog mailing list      (email@hidden)
Help/Unsubscribe/Update your Subscription:

This email sent to email@hidden

  • Follow-Ups:
    • Re: What exactly is net.inet.tcp.sockthreshold?
      • From: "Chris McFarling" <email@hidden>
References: 
 >Re: What exactly is net.inet.tcp.sockthreshold? (From: "Peter Sichel" <email@hidden>)
 >Re: What exactly is net.inet.tcp.sockthreshold? (From: "Chris McFarling" <email@hidden>)

  • Prev by Date: Re: What exactly is net.inet.tcp.sockthreshold?
  • Next by Date: Re: SPNEGO bug in CFNetwork
  • Previous by thread: Re: What exactly is net.inet.tcp.sockthreshold?
  • Next by thread: Re: What exactly is net.inet.tcp.sockthreshold?
  • Index(es):
    • Date
    • Thread