Subj : Re: kermit protocol in syncterm To : fusion From : Digital Man Date : Sat Aug 19 2023 00:05:35 Re: Re: kermit protocol in syncterm By: fusion to Digital Man on Sat Aug 19 2023 02:28 am > On 18 Aug 2023, Digital Man said the following... > > DM> Correct, I'm talking about the protocol itself. Unless you're using a > DM> compression feature with Kermit (are you?), it's really impossible for > DM> the Kermit *protocol* to outperfrom Ymodem-G protocol over the same > > no compression. i poked around on another machine for more testing.. built > sexyz from http://synchro.net/Synchronet/sbbs_src.tgz and used it on my > linux 'router' (i5-3550) to transfer a 700meg movie to my main machine with > syncterm and then with ZOC via kermit and it turned out maybe closer to what > you'd expect: > > sexyz zmodem: ~6+MB/s (~95% cpu) > sexyz ymodem-g: <4.5MB/s (~95% cpu) > gkermit kermit: ~3.7MB/s ( 11% cpu) > > so that's a wash. zmodem still won though, so that's two of my machines that > behaved that way. You only listed one side of the transfer in that table. There are at least 3 variables here: - the sending program - the receiving program - the protocol used (and many of the protocols have options that different protocol drivers/implementations may make use of or allow the user/sysop to tweak) Changing anything about those 3 variables (versions, options, etc.) will likely give you different numbers. I'm guessing from the paragraph before the table that the sexyz transfers were to SyncTERM and the gkermit transfer was to ZOC. Did you try measuring what through-put you might get via ZMODEM or YMODEM-G when the receiving program is ZOC? Or any of another assortment of terminal programs that support X/Y/ZMODEM? SexyZ and SyncTERM actually use the same x/y/zmodem code, so I'd expect reversing the sender or receiver role between those 2 programs wouldn't likely have much effect. Now, it's possible that the BBS program (Mystic?) is passing a socket descriptor to sexyz that it then has to read and retransmit to the connected-client socket (that's how Synchronet does it anyway), and the efficiency of that passthrough socket implementation can have a big impact on the through-put and CPU utilization on the BBS side. > this though: > > lrzsz zmodem: ~55MB/s to ZOC and ~36MB/s to syncterm. (50% cpu) > > is crazy.. syncterm did manage to receive from lrzsz at ~66MB/s with > ymodem-g, but i had no successful transfers with it. Which version of SyncTERM were you using? If it was built from a recent download of sbbs_src.tgz, then I certainly would expect YMODEM-G transfers to work, but maybe the sending version of lrzsz doesn't support the 'G' variant of YMODEM? > so yeah, ymodem-g is better, but which machines are you getting this > performance from using sexyz? Me? I don't really make any peformance claims. > does anyone else? and what's it doing with all those cpu cycles? It's trying to keep the TCP socket transmit buffer full but likely not quite as efficiently as lszrz does (which is a bit baffling since lszrz uses stdio and not socket I/O). You could try using sexyz in stdio mode (if Mystic supports that) and see what impact that has. > DM> Your "real life" test is over a localhost link or a local network? Why > DM> on earth would you be using a serial/modem file transfer protocol over > DM> a local network (Ethernet, WiFi?) and call that "real life"? > > there isn't a significant difference between over the local network and the > internet anymore. I suppose that depends on both factor. While Gigabit Ethernet and WiFi are pretty prevalant (and multi-gig Ethernet making in-roads), Gigabit Internet access to the home is still only available to the vast minority of users. > why shouldn't i be able to fetch a movie off a bbs via > zmodem @ 55MB/s? many people with VPSes or that have fiber at home have that > capability. You should. But I don't think Kermit is the magic bullet to providing the optimal through-put. If anything, that protocol would be FTP or (to a bit lesser degree) HTTP, but I do agree: sexyz and SyncTERM could probably do for some optimizations and possibly more robustness with X/Y/ZMODEM file transfers still. And there are a lot of options for ZMODEM operation too (at least in SexyZ), you might want to play with those and see what you find. -- digital man (rob) This Is Spinal Tap quote #45: I don't really think the end can be assessed as of itself as being the end Norco, CA WX: 68.0øF, 75.0% humidity, 0 mph WNW wind, 0.00 inches rain/24hrs --- þ Synchronet þ Vertrauen þ Home of Synchronet þ [vert/cvs/bbs].synchro.net .