Date: Tue, 24 Sep 1996 18:22:44 -0400 (EDT) From: Brian Tao <taob@io.org> To: Joe Greco <jgreco@brasil.moneng.mei.com> Cc: freebsd-isp@FreeBSD.ORG Subject: Re: Thoughts on a news server cluster Message-ID: <Pine.NEB.3.92.960924161729.415E-100000@zap.io.org> In-Reply-To: <199609241655.LAA06476@brasil.moneng.mei.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, 24 Sep 1996, Joe Greco wrote:
>
> J****!
>
> Let me build this in my mind quickly...
>
> Pentium 133 with ASUS P/E-XXXX????? MB $ 800
> 3 x NCR 810, 1 x SMC EtherPower 10/100 $ 320
> 192MB RAM $1200
> 6 x ST32550N $4100
> 4 x ST31055N $1200
> 2 x ST15150N $1850
> Ext enclosures (3) $ 660
> -----
> $10130
Yep, this is the cost breakdown (in Cdn $... divide by 1.4 or so
to get US bucks):
19" rackmount drive chassis $1275
RAIDION Gandiva controller $4000
8 x 2.1GB drive modules $10200
--------------------------------------------
1 RAID system $15475
Add $400 for an Adaptec 2940UW controller, then multiply by two.
Compare that to $3605 for a PPro200, 256K L1 and 512K L2, ASUS
P/I-P6NP5 motherboard, 128MB of EDO, AHA-2940UW, Seagate ST31230W 1GB
wide drive and SMC EtherPower 10/100Mbps.
> That gives you 24GB usable _local_ disk capacity and additional
> I/O bandwidth on top of it... and you can build three with your
> $25000 plus some change, considering that you can get quantity
> pricing on a purchase of so many drives.
>
> And you get _complete_ redundancy rather than only disk subsystem
> redundancy. That is the part that gets me excited.
True enough. I would like to have the rackmount chassis and
slide-in drive sleds to minimize downtime if I do have to replace a
drive. An 8-bay rack chassis from CRU is around $1400 I think. It
includes the sleds, a second power supply and an external 68-pin wide
SCSI connector. That would knock the price of a 14x2GB subsystem down
to around $16000. More points to ponder... :)
> Yes. I deal with it by not launching nnrp's out of innd. I have
> something (local hackery) called "connectd"
I saw some early source code that you (at least I think it was
you... I might be confusing you with T. W. Wells) had posted to
news.software.nntp. I haven't actually done anything with it since I
didn't have another machine on which to experiment.
> You also have windows of unavailability when the server is running
> news.daily and doing a renumber, etc. etc... spawning out of innd is
> not ideal.
No, the monolithic server approach was a good idea back when news
volumes were small, but now there's too much going on for a non-
threaded process to handle. All innd should be is a glorified inetd
that does some coordination between feeds, but otherwise fire off new
connections as quickly as possible.
> Why not compute a few prime numbers too. ;-)
It's high time I ran Crack on the passwd file too. ;-)
> Grab the active off of a slave - if you are really anal, grab the
> active off all the slaves and write a little perl script to find the
> max for each group (just in case the slaves were a tad out of sync).
> That is a bit of work, I agree, but not hard.
Hrm, I'm still not so sure... XREPLIC seems like a big kludge to
me, and I need to depend on the master server anyway if I want to be
able to post. If I run NFS between the readers and have them all post
back to the primary reader, it at least insulates them from badness
happening on the feeder server. There is still a point of failure on
the primary reader, but that's somewhat better than having everything
depend on a single server being available 100% of the time.
> A qualified "Yes." You have the same problem no matter what you do,
> since INN has a synchronous posting paradigm that in my opinion bites
> the big one.
Something that could quickly batch incoming posts and then inject
them into the news stream a la rnews would be sufficient unless you
were concerned with sub-minute propagation of articles from your site.
> I got exasperated and did something different. I developed a smart
> spooling system to deal with it. Now people can "post" even if the
> master and all the other slaves are dead. At the same time I took the
> opportunity to add a comprehensive posting accounting system that
> records whatever the hell gets posted. It's been useful several times
> already...
I think your work would make a *fine* contribution to the INN
contrib/ directory... ;-)
> XREPLIC is a form of mildly tying your hands. On the other hand, it
> keeps your machines in sync! Which is what an ISP needs.
Keeping multiple reader servers in sync is definitely a must, but
I've heard about an equal number of opinions favouring NFS or
XREPLIC. NFS is cheaper, your articles by definition are "in sync",
you don't have to run an innd process for each server, but it sucks up
I/O and CPU resources. XREPLIC requires local disks, can keep
articles in sync, hogs more RAM because of innd, but offers better
redundancy (at least for reading).
> Must be nice to have a small active file. ;-)
Back when I kept detailed readership stats, I found that over a
period of one week, only ~4500 groups had at least one access. Most
of the readership was in 50 to 60 of the alt.binaries and alt.sex
groups. Over the past year, I've added a couple hundred groups on
user request, and people seem pretty happy with that arrangement.
--
Brian Tao (BT300, taob@io.org, taob@ican.net)
Senior Systems and Network Administrator, Internet Canada Corp.
"Though this be madness, yet there is method in't"
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.NEB.3.92.960924161729.415E-100000>