View Single Post
Old 29-01-2013, 17:22   #3324
qasdfdsaq
cf.mega poster
 
Join Date: Aug 2004
Posts: 11,207
qasdfdsaq is cast in bronzeqasdfdsaq is cast in bronzeqasdfdsaq is cast in bronzeqasdfdsaq is cast in bronze
qasdfdsaq is cast in bronzeqasdfdsaq is cast in bronzeqasdfdsaq is cast in bronzeqasdfdsaq is cast in bronzeqasdfdsaq is cast in bronzeqasdfdsaq is cast in bronzeqasdfdsaq is cast in bronzeqasdfdsaq is cast in bronzeqasdfdsaq is cast in bronzeqasdfdsaq is cast in bronzeqasdfdsaq is cast in bronzeqasdfdsaq is cast in bronzeqasdfdsaq is cast in bronzeqasdfdsaq is cast in bronze
Re: Think Broadband Ping Monitor Results (POST YOURS)

Quote:
Originally Posted by Sephiroth View Post
The takeaway in the context of Qasi's post is that the frequency of ACKs sent back (through a busy) upstream on data received will slow the download (decrease throughput).

Whether or not that is a 5-10% overhead I'll leave to Qasi to justify, if he cares to. But it is an important point that he makes.
Whoops, missed this earlier (haven't been on much lately)

TCP overhead is typically 1.5-3% of the downstream usage. Then add on top DOCSIS overheads, and the fact an upstream channel has really about 1/3rd the capacity of a downstream channel, that means 4.5-9% of the upstream channel I just rounded that to 5-10%.

---------- Post added at 17:22 ---------- Previous post was at 17:16 ----------

Quote:
Originally Posted by Kushan View Post
Believe it or not, this situation is nothing new to VM. Remember when the 50Meg first came out and had an upload speed of barely over 1.5Mbit? (If I recall correctly). The ACKs alone would nearly congest the upload stream in itself, yet it actually ran fairly smoothly (For most, there's always an exception).
Heh, well... On 50Meg they had 2 upstreams for every area (theoretically), but at half the bandwidth, so one upstream per 4 channels. What they have now, I don't know, but if it is 1-between-8, then that is not a good thing given general utilization in both directions will be higher now than it was when 50 came out.

50 didn't run particularly smoothly in areas with high utilization - engineers in my area quoted 90% capacity in use 90% of the time (!) - and getting more than 300Kbps upload was a chore some days.

Quote:
Anyway, there is a point in what you're saying but he's having this issue without actually utilising his downstream, which is the point I'm making - just HAVING 8 downstreams versus 1 upstream isn't the issue, there's obviously something else going on there.
The problem isn't him using his connection, both upstream and downstream channels are shared between hundreds of users. Other people using it to capacity will cause problems for him whether he uses it himself or not.

Quote:
I'd like to know where you're getting your figures, though, 5-10% of upstream per downstream for example. Are you referring to the available bandwidth for the connection (What profile the modem is set to), or the available bandwidth for the channels (8x50Mbit on the downstream, 1x~30Mbit on the upstream?) or what?
Bandwidth over the whole channel (8x55Mbit on the downstream, 1x20Mbit on the upstream).

Remember upstream is inherently inefficient on cable, because of the requirement for CSMA and guard intervals, etc. and you will rarely be able to use entirety of a channel's upstream capacity. The practical maximum is closer to 16-18Mbit available for actual use.

55Mbps dowstream with 2% used for ACKs is around 1.1Mbps if everything behaves efficiently, depending on RWIN, MTU, OS, etc. this can vary both up and down. Don't forget also a single 100/10 user could eat up over half the upstream capacity on such a channel.

(An ACK takes around 54 bytes 'on the wire', which with a pessimistic minimum MTU of 576 bytes, and no ack suppresion, etc., means 8 downstream channels could actually require ~40Mbps of upstream capacity in an almost-worst-case-scenario, without anybody doing any actual data uploads)
qasdfdsaq is offline   Reply With Quote