![]() |
![]() |
#1 |
P90 years forever!
Aug 2002
Yeehaw, FL
2·32·7·59 Posts |
![]()
I have cable Internet with 200Mbps down, 12Mbps up.
Is this full duplex? I ask because when I upload proof files, which takes a couple of minutes, downloading seems to practically stop (which does not make the wife watching videos very happy). I've tried pausing 5 seconds between 2MB chunks and while better is still not good. I can try the Curl Library bandwidth rate limiting option and/or increasing the pause or decreasing the chunk size. My goal is to come up with good default settings in prime95 (and gpuowl) so that users rarely complain about impact on their Internet experience. Thus, the question also applies to DSL and whatever else is commonly used these days. |
![]() |
![]() |
![]() |
#2 |
"Rashid Naimi"
Oct 2015
Remote to Here/There
111110111002 Posts |
![]()
A number of factors including wifi half duplex:
https://www.quora.com/When-I-upload-...-and-be-slower |
![]() |
![]() |
![]() |
#3 |
P90 years forever!
Aug 2002
Yeehaw, FL
2·32·7·59 Posts |
![]()
Just uploaded two proofs throttled to 1Mbps. No stuttering videos. May try 2Mbps next time.
|
![]() |
![]() |
![]() |
#4 |
P90 years forever!
Aug 2002
Yeehaw, FL
2·32·7·59 Posts |
![]()
Which leaves the question of what should prime95's default upload speed be? 1Mbps takes 15 minutes to upload a proof. Users probably don't care - it happens in background.
AWS users and the like can set some option to use unlimited bandwidth. |
![]() |
![]() |
![]() |
#5 |
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest
712 Posts |
![]()
How much trouble would it be to give the user a field to enter that? Or to do a brief raw upload rate test and then limit proof uploads to about 10-20% of that? TCP or UDP traffic in one direction always implies a little traffic in the opposite direction. (at least TCP Syn, Syn-Ack, Ack pingpong; dns, icmp, tracert, UDP request and response, control traffic for RTSP, etc. https://tools.ietf.org/html/rfc2326) So thoroughly saturating one direction can bring use of the other direction to a near standstill. I've seen that occur multiple ways in decades past in shared departmental uplinks even with full duplex symmetric data rates of T1 or 10Mhz or 100Mhz full duplex Ethernet. (Rural uplinks are choice-limited so the near monopoly means expensive.) Heavy loads of links were scheduled for off hours to preserve interactive response during the work day. Unexpected prime-time saturating loads were tracked down and dealt with. (Once found a spam mail server on a coworker's workstation from emailed malware, and things got a LOT better after that workstation was isolated. Another time a failing NIC was jabbering and killing a whole LAN segment. Another time a printer on another continent was flooding traffic into our slow link, and the upstream NOC blackholed that IP#. A neighboring department flooded their 100Mhz ethernet uplink with backup traffic to a different zip code, and we negotiated a rescheduling because it was essentially shutting down use of our 10Mhz slice also.)
Last fiddled with by kriesel on 2020-07-25 at 22:05 |
![]() |
![]() |
![]() |
#7 |
If I May
"Chris Halsall"
Sep 2002
Barbados
3×3,181 Posts |
![]()
No.
This is known as asymmetric bandwidth allocation, which is almost universal for consumer-grade connectivity. In addition to this, your provider will almost certainly be "oversubscribing" the bandwidth. If you read your contract closely, you'll find that these speeds will be the maximum you can use, but you'll often see slower speeds (particularly during peak usage periods). |
![]() |
![]() |
![]() |
#8 | |
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest
116618 Posts |
![]() Quote:
Full or half duplex and their details, and the usual case of rate asymmetry, are separate distinctions. I'm pretty sure you know that, given your occupation. This was a good review for me. It's the difference between traffic in opposite directions at the same time but different speeds, or flow in only one direction at a time, as when during road construction there are flagmen at either end of a one lane section, admitting n cars one way, waiting for them to clear, then admitting n cars going in the opposite direction, with buffering (waiting) on each end for the flagmen to give the go signal. http://tcpipguide.com/free/t_Simplex...xOperation.htm See for example https://en.wikipedia.org/wiki/Asymme...line#Operation, which is highly asymmetric (I had nominally 8:1 ADSL) but usually full duplex. (Using different frequency bands for the two directions is common.) Cable internet appears even more asymmetric. https://en.wikipedia.org/wiki/Cable_Internet_access It appears to be true that until recently, coax cable internet was based on standards that were not full duplex. https://en.wikipedia.org/wiki/DOCSIS My current internet service is nominally symmetric 300Mbps duplex fiber, with browser-measured download and upload rates of 33/25, 96/94, 155Mbps/127Mbps or better or worse depending on system speed and LAN topology. (The lowest of those is an old i3 laptop with old nominally 802.11n wireless builtin adapter that's limited in performance to 72Mbps. Plugging in a USB wireless-ac gets 135/124 on the same system.) These are actually pessimistic numbers. A speed test application is recommended for line rates above 100Mbps, but these were obtained with a browser tab. Installing the Ookla speedtest app on the same old i3 laptop gave 226down/241up, vs 135/124 via browser, for the USB wireless-ac. It was a great relief to get fiber here. I got the slowest cheapest fiber rate offered; there was also 600/400 and 1000Mbps/400. The old slow ADSL was nominally 768k/128k, but often in practice 0/0 in the last year before fiber became available. And ADSL cost >50% more than the fiber does now. ADSL is mostly full duplex. https://en.wikipedia.org/wiki/Asymme...line#Operation Even early dialup internet protocols were full duplex. Mobile data such as I sometimes used by tethering my phone to a laptop during ADSL outages can be either full duplex (frequency division) or half duplex (time division). https://en.wikipedia.org/wiki/LTE_(telecommunication). I saw LTE rates ~25Mbps down, 5M up on a cheap prepaid plan. Terrestrial wireless is apparently mostly half duplex https://en.wikipedia.org/wiki/Wireless_broadband Satellite internet is varied. Huge latency, sometimes masked by the modem giving an immediate acknowledge, or reduced by using terrestrial uplinks, TDMA uplink, some using dialup or other separate channels for uplink, monthly quotas on downloads, etc. https://en.wikipedia.org/wiki/Satellite_Internet_access Last fiddled with by kriesel on 2020-07-26 at 20:13 |
|
![]() |
![]() |
![]() |
#9 | |
If I May
"Chris Halsall"
Sep 2002
Barbados
3×3,181 Posts |
![]() Quote:
![]() A networking engineering friend of mine in BC, CA said Telus got lucky because they designed their GPON backhaul to be largely symmetric. Rogers, on the other hand, was seriously asymmetric, and so couldn't deal with the unexpected residential networking demands during the CV crisis. |
|
![]() |
![]() |
![]() |
#10 | |
"/X\(‘-‘)/X\"
Jan 2013
3·977 Posts |
![]() Quote:
But that brings up a point: does the upload process support resuming on interruption? |
|
![]() |
![]() |
![]() |
#11 | |
"Mihai Preda"
Apr 2015
5×271 Posts |
![]() Quote:
Each upload (throttled or not) keeps a TCP connection open with the server. You want the lifetime of the TCP to be short, and this mainly for the benefit of the server. You don't really know how the well the server would behave if it starts having hundreds of open TCP connections at the same time. |
|
![]() |
![]() |
![]() |
Thread Tools | |
![]() |
||||
Thread | Thread Starter | Forum | Replies | Last Post |
Internet down question | ctteg | Information & Answers | 2 | 2018-03-13 17:57 |
PCI-E USB 3.0 Extension Cable | vsuite | GPU Computing | 7 | 2017-07-10 20:45 |
The Expanse has to be the most epic 45 minutes to get a first showing on regular cable. | jasong | jasong | 25 | 2016-01-21 16:03 |
best buy 13.99 for 3ft cat-5e cable | crash893 | Soap Box | 5 | 2012-03-31 02:06 |
Too Much Internet Access. | M0CZY | Software | 3 | 2005-10-17 15:41 |