A friend has had a lot of fun testing his 10g/10g. Drivers, and what kind of bandwidth you have available on the PCI-E make all the difference, no TCP tuning required beyond being on Windows 10.
The last card he had was PCI-E 2.0 x 4 lanes - this card had trouble around 2-3Gbit/s. The new card he moved to is PCI-E 3.0 x 4 - this he gets full speed out of. It could be the lack of good driver support, but there isn't anything conclusive he ever found. 10g is a finicky beast on Windows. On Thu, Sep 21, 2017 at 1:03 AM, Forrest Christian (List Account) < [email protected]> wrote: > There are TCP/IP tuning things you may need to do to get around the > bandwidth*delay product. > > When a TCP/IP session starts up, one of the things which is exchanged is > the maximum 'receive window' and 'transmit window'. This is effectively > the maximum number of bits which can be in-flight at a given moment. > > The relevant formula here is windowsize=bandwidth * delay. > > For whatever reason my windows 10 machine seems to like to default to a > window size of 65536, or 524288 bits. Dividing this by for instance a > delay (latency) of 5ms, you get just over 104Mb/s maximum. Reducing the > delay by half doubles the maximum. Doubling the window size also doubles > the maximum. > > Windows is supposed to auto-scale the window size. In theory it is > supposed to be able to adjust this all the way up to 16MB based on the > default scaling value it uses. (it probably can even be tweaked further). > This would result in around 3.4GB of throughput at 5ms delay. I don't > know how well this actually works though.... > > Of course there are two parts to this - *both* ends need to be configured > to use the larger window sizes. I have found that often the server-end > of these tests end up with a horribly small transmit window which tends to > be the bottleneck. > > > > > > > > On Wed, Sep 20, 2017 at 2:31 PM, Chuck McCown <[email protected]> wrote: > >> Yeah, that is kinda what I am running in to. There has to be a way >> around it. I posted this problem to ookla too. Hopefully they will >> respond. >> >> *From:* Al Rachide >> *Sent:* Wednesday, September 20, 2017 2:28 PM >> *To:* [email protected] >> *Subject:* Re: [AFMUG] Speed Tests >> >> We did a speed test on a full GIG DIA fiber line. It took THREE computers >> simultaneously, two with I5s and one with an I7 processor. All 3 had >> gigabit internet cards. We did indeed get a totwl throughput of 997mbit >> down and 1003mbit up. But that was the combined total of all 3 computers >> running the speed test at the same time. We think the reality is that any >> one computer, no matter how powerful, can actually send or receive 1000mbit >> speed. Al >> >> On Wed, Sep 20, 2017 at 3:38 PM, <[email protected]> wrote: >> >>> Send Af mailing list submissions to >>> [email protected] >>> >>> To subscribe or unsubscribe via the World Wide Web, visit >>> http://afmug.com/mailman/listinfo/af >>> or, via email, send a message with subject or body 'help' to >>> [email protected] >>> >>> You can reach the person managing the list at >>> [email protected] >>> >>> When replying, please edit your Subject line so it is more specific >>> than "Re: Contents of Af digest..." >>> >>> >>> Today's Topics: >>> >>> 1. Re: Hold on Gino.... (Steve Jones) >>> 2. Speedtests (Chuck McCown) >>> 3. Re: Speedtests (Jesse DuPont) >>> >>> >>> ---------------------------------------------------------------------- >>> >>> Message: 1 >>> Date: Wed, 20 Sep 2017 14:22:53 -0500 >>> From: Steve Jones <[email protected]> >>> To: "[email protected]" <[email protected]> >>> Subject: Re: [AFMUG] Hold on Gino.... >>> Message-ID: >>> <[email protected] >>> ail.com> >>> Content-Type: text/plain; charset="utf-8" >>> >>> their power is like 80 or more percent out >>> >>> On Wed, Sep 20, 2017 at 1:43 PM, Zach Underwood <[email protected]> >>> wrote: >>> >>> > Also looks like his backbone is still online >>> > >>> > https://stat.ripe.net/14979#tabId=routing >>> > >>> > On Wed, Sep 20, 2017 at 2:39 PM, Zach Underwood <[email protected]> >>> > wrote: >>> > >>> >> More than half of the prefix from PR are offline >>> >> https://stat.ripe.net/PR#tabId=routing >>> >> >>> >> On Wed, Sep 20, 2017 at 2:28 PM, SmarterBroadband <[email protected]> >>> wrote: >>> >> >>> >>> Maria must have be giving you hell. >>> >>> >>> >>> No power on the island!!! >>> >>> >>> >>> Hold on and Good luck. >>> >>> >>> >> >>> >> >>> >> >>> >> -- >>> >> Zach Underwood (RHCE,RHCSA,RHCT,UACA) >>> >> My website <http://zachunderwood.me> >>> >> advance-networking.com >>> >> >>> > >>> > >>> > >>> > -- >>> > Zach Underwood (RHCE,RHCSA,RHCT,UACA) >>> > My website <http://zachunderwood.me> >>> > advance-networking.com >>> > >>> -------------- next part -------------- >>> An HTML attachment was scrubbed... >>> URL: <http://afmug.com/pipermail/af/attachments/20170920/96558efd >>> /attachment-0001.html> >>> >>> ------------------------------ >>> >>> Message: 2 >>> Date: Wed, 20 Sep 2017 13:34:36 -0600 >>> From: "Chuck McCown" <[email protected]> >>> To: <[email protected]> >>> Subject: [AFMUG] Speedtests >>> Message-ID: <B6646750F342408BAD1B98628F2B839F@ChuckMcCownPC> >>> Content-Type: text/plain; charset="iso-8859-1" >>> >>> Hard to find an app or appliance that will reliably show a customer the >>> speed they are getting when it is above 100 Mbps. We have increasingly >>> more 250, 500 and 1G customers and when they complain that speed test shows >>> a lower number I need something to prove them wrong. An average laptop >>> does not cut it. >>> >>> We have installed our own speedtest server with the ookla recommended >>> hardware etc. But it takes a pretty good computer that actually show a >>> gig. Ditto iperf. Be nice if there was some kind of handheld device that >>> could do this. There are all kinds of hand held computers designed to >>> roll your own piece of test or control gear. Just not sure what is >>> important. >>> >>> CPU speed >>> Memory size >>> PHY circuit >>> Memory type >>> >>> I guess I should ask this question of ookla... >>> -------------- next part -------------- >>> An HTML attachment was scrubbed... >>> URL: <http://afmug.com/pipermail/af/attachments/20170920/0a9b1b90 >>> /attachment-0001.html> >>> >>> ------------------------------ >>> >>> Message: 3 >>> Date: Wed, 20 Sep 2017 13:38:23 -0600 >>> From: Jesse DuPont <[email protected]> >>> To: [email protected] >>> Subject: Re: [AFMUG] Speedtests >>> Message-ID: <[email protected]> >>> Content-Type: text/plain; charset="us-ascii" >>> >>> An HTML attachment was scrubbed... >>> URL: <http://afmug.com/pipermail/af/attachments/20170920/c79c66ee >>> /attachment.html> >>> -------------- next part -------------- >>> A non-text attachment was scrubbed... >>> Name: celeritynetworks-GIF.gif >>> Type: image/gif >>> Size: 1869 bytes >>> Desc: not available >>> URL: <http://afmug.com/pipermail/af/attachments/20170920/c79c66ee >>> /attachment.gif> >>> -------------- next part -------------- >>> A non-text attachment was scrubbed... >>> Name: Screen Shot 2017-09-20 at 9.48.58 AM.png >>> Type: image/png >>> Size: 110022 bytes >>> Desc: not available >>> URL: <http://afmug.com/pipermail/af/attachments/20170920/c79c66ee >>> /attachment.png> >>> >>> ------------------------------ >>> >>> Subject: Digest Footer >>> >>> _______________________________________________ >>> Af mailing list >>> [email protected] >>> http://afmug.com/mailman/listinfo/af >>> >>> >>> ------------------------------ >>> >>> End of Af Digest, Vol 37, Issue 321 >>> *********************************** >>> >> >> >> >> -- >> Al Rachide >> Pink Hill, NC >> > > > > -- > *Forrest Christian* *CEO**, PacketFlux Technologies, Inc.* > Tel: 406-449-3345 | Address: 3577 Countryside Road, Helena, MT 59602 > <https://maps.google.com/?q=3577+Countryside+Road,+Helena,+MT+59602&entry=gmail&source=g> > [email protected] | http://www.packetflux.com > <http://www.linkedin.com/in/fwchristian> <http://facebook.com/packetflux> > <http://twitter.com/@packetflux> > >
