Date: Wed, 2 Oct 1996 14:18:58 -0500 (CDT) From: Joe Greco <jgreco@brasil.moneng.mei.com> To: cassy@loop.com (Cassandra Perkins) Cc: freebsd-isp@FreeBSD.org Subject: Re: RAID Controller Product Message-ID: <199610021918.OAA05825@brasil.moneng.mei.com> In-Reply-To: <Pine.BSF.3.91.961002113008.10691A-100000@patty.loop.net> from "Cassandra Perkins" at Oct 2, 96 12:03:39 pm
next in thread | previous in thread | raw e-mail | index | archive | help
> On Wed, 2 Oct 1996, Joe Greco wrote: > > > > > > > > I am curious as to why you would choose to do this. > > > > If you currently have your I/O spread amongst several machines, you gain > > from having the data replicated. If you have N machines, you have > > approximately N times the I/O bandwidth available as compared to a > > solution where you only have the main system's I/O bandwidth available. > > > [snip] > > Considering the constant growth of Usenet, it seemed better to use the > extra drives, required to duplicate the number of articles, to increase > storage and time articles are kept on the server. We would still have > multiple machines handling news request, however, the articles would be on > the raid-controlled fileserver. > > As for the fault tolerance issue, must problems I've seen with servers > going down are due to failed drives. So using RAID level 5 I hoped would > reduce the down time considerably. Although, this is less of a benefit if > the RAID controller had poor or non-comparable I/O performance than > concatenated drives (ccd). How do you intend to deal with the transaction per second demands imposed by anything more than a few hundred readers? I've seen as little as 250 simultaneous readers max out a spool built out of a few Barracuda drives. It is truly cool to see the drive lights on solid and hear the heads chattering like its the end of the world. I expect DG's noatime patch helps some. I also expect that the fact that I started striping spools at that point helped considerably. Even so, just how hard do you want to push? You will start to fall behind in your news processing. Your INN daemon will not be able to get the disk timeslices necessary to process a few articles per second. You will become a Slow Site(tm). I have seen it happen... a customer who thought they knew better... they were very sorry in the end :-( ... JG
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199610021918.OAA05825>