Fastream Technologies wrote: > What I tried to explain was that if your program is TCP/IP-dense, > then in order to take advantage of multi-cores, you MUST code MT.
That's true, but does that neccessarily mean that you MUST run socket communication in multiple threads? I can think of many other bottlenecks that could benefit from MT. -- Arno Garrels [TeamICS] http://www.overbyte.be/eng/overbyte/teamics.html > > Regards, > > SZ > > > On 10/11/07, Arno Garrels <[EMAIL PROTECTED]> wrote: >> >> Fastream Technologies wrote: >>> I test my ICS-based MT proxy with 20k connections on our dual-core >>> system. It performs 2GBps, local-to-local. So that's one CPU >>> performance basically since the tester also uses CPU! I would not >>> imagine such performance with single thread. >> >> Those numbers don't tell me much. I'm missing a serious >> comparison between an optimized MT and an optimzed ST server. >> One comparison was how many clients can be connected until the >> first is being rejected due to a full ListenBackLogQueue. >> Another was how long would it take until all clients finished >> their concurrent download. >> >> -- >> Arno Garrels [TeamICS] >> http://www.overbyte.be/eng/overbyte/teamics.html >> >>> >>> Best Regards, >>> >>> SZ >>> >>> >>> On 10/11/07, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote: >>>> >>>> --- QUOTE: SZ >>>> If you do not want the ability to use multi-cores for >>>> communication threads, >>>> then async is the way to go. But IMO, it is an ill >>>> design since chipmakers >>>> are talking about 64-core CPUs and 10Gbps networks. >>>> --- END. >>>> >>>> Thanks, SZ. At this point I'm not so much concerned >>>> about complexity (although, of course, that is a >>>> concern), but more about performance and efficiency. >>>> I know that the arguments are always against >>>> multiple threads because they are harder to debug and >>>> synchronize, which is a very valid argument, and one >>>> that's biting me in the ass right now, but is that at >>>> the cost of a perceivable performance hit? Or is the >>>> async component not only simpler to use, but just as >>>> fast (or at least not significantly slower)? >>>> >>>> -dZ. >> -- >> To unsubscribe or change your settings for TWSocket mailing list >> please goto http://www.elists.org/mailman/listinfo/twsocket >> Visit our website at http://www.overbyte.be >> > > > > -- > Fastream Technologies > Software IQ: Innovation & Quality > www.fastream.com | Email: [EMAIL PROTECTED] | Tel: +90-312-223-2830 > Join IQWF Server Yahoo group at > http://groups.yahoo.com/group/IQWFServer Join IQ Reverse Proxy Yahoo > group at http://groups.yahoo.com/group/IQReverseProxy -- To unsubscribe or change your settings for TWSocket mailing list please goto http://www.elists.org/mailman/listinfo/twsocket Visit our website at http://www.overbyte.be