Subject: Re: email (was Re: Recoverable Network File System?)
To: None <Netbsd-Users@NetBSD.org>
From: Chuck Yerkes <chuck+nbsd@2003.snew.com>
List: netbsd-users
Date: 12/12/2003 19:38:36
Quoting Thor Lancelot Simon (tls@rek.tjls.com):
> On Fri, Dec 12, 2003 at 06:47:05PM -0500, Chuck Yerkes wrote:
> > 
> > NETWORKS
> > At 100Mb/s, 100baseTX can give us files at about the speed of a 7200RPM
> > baracuda drive.  Maybe 8MB/s.  GigE is faster, but still not as fast as
> > local 15kRPM drives.  Not close.
> 
> That's nonsense.  Show me a single 15kRPM drive that sustains 90KByte/sec.

My unclarity:
GigE is faster than 100baseT, but not nearly as fast as my Baydel
RAID boxes with 5 or 10  15K RPM drives.

I've debugged NFS.  I've seen NFS servers very rarely implemented
well for truly high speed bashing (2 different clients getting an
exclusive create on the same file - that was sweet).  NFS has gotten
less predictable in the last couple years across platforms.  (it
was udp only, slow and unsecure but pretty damn reliable in 93 when
I could mount NFS on PCs, Macs, 4 Unixes and VMS for home dirs and
the like).

I like AFS, I wish it had been "open" 10 years ago when we really needed
it to take off.  But I'm not sure I'd be using it for mid performance
email systems.

Too often I've seen HA sold as a "magic bullet" when, in fact, the effort
to filter who NOT to sell it to would have reduced our (vendor) tech support
costs.

Too often, I find "redundant systems" that in fact, aren't redundant
when it comes to it - as they get out of sync, as complex software is
forgotten because you don't actually have a failure for 9 months or
more and it just slides out of your mind.

Support costs are lower and systems easier to debug when they are
simpler.  EMail is an enterprise critical service, and yet, in
Fortune 500, at ISPs, at smaller businesses, if email is down
unexpectedly for an hour once every 6 months, nobody's going to
get fired for just that.

My systems have dual power supplies, quality ECC RAM and the whine
and complain when something is going bad (like a PS).

If a non-redundant component, like a CPU or motherboard dies, then
I *can* often swap systems quickly with limited effect to the enterprise.
Mail still goes out (mail relays are mainly dataless).  Incoming mail
spools.  The loss is to people seeing their mail during this time.