The problem is that each hop has its own physical connection, and it is likely that they will not all be the same. The slowest or most congested physical link is the one that will determine the overall speed of the end to end connection. For an end to end link, you can't just say that you have a WiFi connection. Your WiFi hotspot is connected somehow to a router (possibly through a switch or 2). The router is connected to your ISP through a third type of connection to the next router in the series. There will be 0 or more other routers and switches between your ISP and the ISP of the project. Past the router at the ISP of the project there will be one or more switches and routers and possibly a load balancer before you get to the server you are contacting. You cannot know in advance which of these will be the bottleneck. You can't even tell in advance which routers will be used by your connection other than the one at each end.
In my job, we have to worry about this a little. We have one customer that has to use packet radio as the backbone between its end clients and headquarters. This is the fastest and most reliable connection that they can get. It runs sub 300 bps, and goes out frequently. The networks at the end points are fast though. I do not believe they are running BOINC, but it is an example of the sorts of network problems you will have to contend with when thinking about internet connections worldwide. -----Original Message----- From: boinc_dev [mailto:boinc_dev-boun...@ssl.berkeley.edu] On Behalf Of Robert Miles Sent: Sunday, April 2, 2017 9:21 PM To: elliott...@comcast.net; boinc_dev@ssl.berkeley.edu Subject: Re: [boinc_dev] An additional preference to prevent, downloading when on WiFi, to enable downloading only on when connected, to cable Charles Elliott, Looks like you're using a rather different definition of "connection" than most of the other people in this discussion. I, and probably most of the others, use it to mean a physical method of connection, such as ethernet, wifi, analog modem, and so on. The ports are a way to effectively divide up one IP connection 65536 ways. This is important for servers, which otherwise could not handle exchanging information with more than one computer at a time for each physical connection. On 4/2/2017 9:04 AM, Charles Elliott wrote: > A computer can have as many connections to an IP address as it has free > ports (numbered up to 65535, although many of the numbers less than 32768 > are reserved) and memory for the software implementation of an abstract > concept called a "socket." > > A TCP/IP connection is defined by four numbers: client IP address, client > port; server IP address, server port. For example, right now I have, inter > alia, the following connections displayed in the connections table of my > router: > > 192.168.0.100:61266 151.101.22.2:80 > 192.168.0.100:61368 151.101.22.2:80 > 192.168.0.100:61367 151.101.22.2:80 > 192.168.0.100:61265 151.101.22.2:80 > 192.168.0.100:61268 151.101.22.2:80 > 192.168.0.100:61269 151.101.22.2:80 > > > 151.101.22.2 is an IP address owned by fastly.com, a CDN, which I believe > stands for Content Delivery Network. When a browser accesses a webpage, it > is told to download the advertisements from fastly.com. Advertisers upload > their content to fastly.com, and it is delivered when and where requested. > Fastly may also help with billing issues and generating statistics on > impressions delivered and viewed and possibly follow thru, but I am not > certain about that. > > I don't understand what this discussion is about, though I suspect it is > fundamentally about net neutrality. At one point the Oxford project person, > David, wrote: > >>> It would be much easier if the clients did this. My Mac for example >>> is able to tell me the latest network bandwidth it has for any of its >>> interfaces. > If by client he means his application, then all he has to do is put the > download timing code in his application. Then if he wants to refuse to > deliver very large files to computers with very narrow bandwidth > connections, it is all on him and the school. I believe that if you just > explain to people with clear, jargon-less statements why it is infeasible to > deliver very large files in kilobit dribbles, they will not object. I did > not like it when GPU Grid told me I could not participate when I signed up > because my video cards were inadequate, but I got over it (and eventually > bought new video cards, which I used elsewhere). > > Charles Elliott > > > -----Original Message----- > From: boinc_dev [mailto:boinc_dev-boun...@ssl.berkeley.edu] On Behalf Of > Robert Miles > Sent: Sunday, April 2, 2017 1:26 AM > To: boinc_dev@ssl.berkeley.edu > Subject: Re: [boinc_dev] An additional preference to prevent, downloading > when on WiFi, to enable downloading only on when connected, to cable > > From what I've seen, Windows operating systems do not allow multiple > connections to the internet to be active at once, which means that only one > such connection can be reachable from BOINC at any one time. > > On 3/30/2017 8:34 AM, boinc_dev-requ...@ssl.berkeley.edu wrote: >> Date: Thu, 30 Mar 2017 16:15:55 +0300 >> From: Vitalii Koshura <lestat.de.lion...@gmail.com> >> Subject: Re: [boinc_dev] An additional preference to prevent >> downloading when on WiFi, to enable downloading only on when > connected >> to cable >> >> Hello gyus, >> >> Maybe it would be preferably to let user choose the connection BOINC >> uses for download/upload? >> So in this case BOINC need to detect all available connections and >> make a checkbox list where user can choose which connection should be >> used when available? >> Because doind automated testing of connection speed is not good for >> these purposes. >> >> Thanks >> >> Best regards, >> Vitalii Koshura >> >> 2017-03-30 16:08 GMT+03:00 McLeod, John <john.mcl...@sap.com>: >> >>> We have had this discussion before -- back in the days when dialup >>> was common. Without doing an end to end test of the connection there >>> is no way to tell what the connection speed is. Dialup is still the >>> way things are in some places in the world. Based on dialup with a >>> router and a switch, I can have a 1Gb connection locally but only a >>> 1.3Kb connection to the outside world, and the computer cannot know >>> without doing an end to end test. With ADSL links (much more common >>> at the moment) it is possible to have a 1Gb connection locally and a 5Mb > connection to the outside world. >>> Running a check like this against some BOINC BOINC projects could tip >>> them over into congestion where nothing gets through. >>> >>> So, it would require: >>> A selection for what high speed meant. >>> A checkbox to indicate if enabled. >>> An end to end check to see what the connection speed was. >>> Some idea of when to use it as doing an end to end connection check >>> when there was only a few hundred bytes to transfer does not seem > reasonable. >>> At the time it was decided that this was not something we wanted to >>> pursue. Of course, this can change. >>> >>> Jm7 >>> >>> -----Original Message----- >>> From: boinc_dev [mailto:boinc_dev-boun...@ssl.berkeley.edu] On Behalf >>> Of David Wallom >>> Sent: Thursday, March 30, 2017 8:28 AM >>> To: elliott...@comcast.net; 'Nicol?s Alvarez' >>> <nicolas.alva...@gmail.com>; Andy Bowery <andy.bow...@oerc.ox.ac.uk> >>> Cc: boinc_dev@ssl.berkeley.edu >>> Subject: Re: [boinc_dev] An additional preference to prevent >>> downloading when on WiFi, to enable downloading only on when >>> connected to cable >>> >>> Hi Charles, >>> >>> With the increasing prevalence of mobile computing devices then >>> having the system (scheduler) doing the test is not really scalable >>> as people move their devices. >>> >>> It would be much easier if the clients did this. My Mac for example >>> is able to tell me the latest network bandwidth if has for any of its >>> interfaces. >>> >>> David >>> ________________________________________ >>> From: boinc_dev [boinc_dev-boun...@ssl.berkeley.edu] on behalf of >>> Charles Elliott [elliott...@comcast.net] >>> Sent: 30 March 2017 13:10 >>> To: 'Nicol?s Alvarez'; Andy Bowery >>> Cc: boinc_dev@ssl.berkeley.edu >>> Subject: Re: [boinc_dev] An additional preference to prevent >>> downloading when on WiFi, to enable downloading only on when >>> connected to cable >>> >>> Boinc could just download a test file from the Oxford website 5 times >>> and average the times. If the average was above a limit deemed the >>> minimum acceptable speed, the user would be permitted to proceed. >>> OW, the Oxford website would post a very polite, very detailed, and >>> very well written message to Boinc/the user explaining why a high >>> bandwidth connection is necessary for the user's progress and enjoyment > of Oxford's project. >>> One of the Boinc GPU projects, as I recall in Spain, does this now >>> WRT the capacity of the user's GPU(s). It is no fun for, or use to, >>> anyone if the user processes a work unit on an older GPU, the GPU >>> overheats, and the WU fails 3/4 of the way through. It is annoying > though. >>> Charles Elliott >>> >>> -----Original Message----- >>> From: boinc_dev [mailto:boinc_dev-boun...@ssl.berkeley.edu] On Behalf >>> Of Nicol?s Alvarez >>> Sent: Wednesday, March 29, 2017 3:40 PM >>> To: Andy Bowery >>> Cc: BOINC Developers Mailing List ?[boinc_dev@ssl.berkeley.edu]? >>> Subject: Re: [boinc_dev] An additional preference to prevent >>> downloading when on WiFi, to enable downloading only on when >>> connected to cable >>> >>> 2017-03-29 14:45 GMT-03:00 Andy Bowery <andy.bow...@oerc.ox.ac.uk>: >>>> Hi, >>>> >>>> We would be interested in an additional BOINC preference, a tickbox >>>> on >>> the 'Network' tab, with something like 'Download only when connected >>> to a high bandwidth connection'. Ticking the box of this preference >>> would prevent download of the application and supporting files when >>> the machine (for example: a laptop) was connected only to WiFi and >>> not connected to a higher bandwidth networking cable. Would it be >>> possible for this to be scheduled to be added as an item to be included > in a later release? >>>> With regards, >>>> >>> What does "high bandwidth connection" mean, how could BOINC know if >>> it's connected to one? >>> >>> -- >>> Nicol?s >>> _______________________________________________ boinc_dev mailing list boinc_dev@ssl.berkeley.edu https://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address. _______________________________________________ boinc_dev mailing list boinc_dev@ssl.berkeley.edu https://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address.
