Date: Tue, 25 Sep 2007 14:41:43 +0400 From: "Artem Kuchin" <matrix@itlegion.ru> To: "Nikos Vassiliadis" <nvass@teledomenet.gr> Cc: freebsd-stable@freebsd.org Subject: Re: device polling and weird timer interrupt count from vmstat Message-ID: <00f901c7ff60$aa84ed50$0c00a8c0@Artem> References: <01ae01c7fd3e$e6ff28f0$0c00a8c0@Artem> <200709251048.20808.nvass@teledomenet.gr> <00b801c7ff54$6796f2b0$0c00a8c0@Artem> <200709251249.11287.nvass@teledomenet.gr>
next in thread | previous in thread | raw e-mail | index | archive | help
Nikos Vassiliadis wrote: > On Tuesday 25 September 2007 12:13, Artem Kuchin wrote: >> I have dual CPU with HT. If i turn on HT (and it does help in my >> case) it shoud 2000 int x4 = 8000 ints per second. SO, i have saved >> 200 int/second from NIC and got myself 8000 ints/second from timer. > > This kind of load(200intrs/s) earns nothing from polling. > >> Did i >> really win anything? I wish there were some good explanation on this. > > It would make a difference for let's say 8000 - 10000 interrupts/sec. > You should read polling(4). It explains a lot:) Yes, i have read the man for polling. However, it does not contain info on when it worth enabling. From this point of view it seems like if i have 8CPU server an a NIC with 8000 ints/sec when i enable polling with HZ=1000 i get about 30 ints/sec from NIC about 16000 ints from NIC So, we were at 8000 ints a second and now we are at 16000 ints/second. Is it worth in this case? It seems like enabling polling on really multi cpu computers never worth it. I think there is a mistake in my logic here. It just canot be like this. any thoughts? -- Regards, Artem
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?00f901c7ff60$aa84ed50$0c00a8c0>