Re: Sybase ASE & server distribution

["Dana M. Diederich" <dana@realms.org>]

> Interesting article on Sybase's free SQL server for Linux.  This might be
> worth looking into as MySQL does have a major flaw currently: a nasty
> memory leak.  I know for a fact that Slashdot runs MySQL, and they've got
> a script called 'fuck' that restarts it when memory starts filling up.  We
> may not want to deal with that.
That's odd, I've been running MySQL at Wal-Mart for many many months and
the memory utilization has been quite stable.  It's still just using a few
megs.  Maybe only certain versions have this problem?

In MySQL's defense, so far at Wal-Mart, discounting local system problems,
MySQL has been the picture of stability and performance.  My biggest
MySQL installation has almost 100 million records in it, and has sustained
13 queries a second on two months of uptime.

Sybase, however, is going to have a lot of functions that MySQL just won't
have. 

(On a side note, some of the big commercial RDBMS' are missing things that
MySQL has.  Informix, for instance, is missing the concept of 
unsigned integers and in version 7, it has no limit verb.  Bah.)
> 
> 
> On another note, I thought of a interesting idea that I think is worth
> doing.  If you guys do swing a system from IBM, what about co-locating the
> systems at the same site?  The obvious advantage of this is that we can do
> load distribition/failover should one server crash or if we want to take
> it down for maintenance.  I have quite a bit of real world experiance
> trying to use DNS Round Robin for load balancing and I can tell you it is
> horrible.  It also provides no failover capability.
Concur.

> 
> If we can get the IBM box here in Calif., NaviSite would provide us usage
> of their shared Cisco Local Director farm.  This way both boxes would have
> the same IP in DNS, but the LD would distribute the load according to the
> percentages we define.  Also it automattically detects when a server goes
> down for any reason, and redirects all requests to the remaining
> server(s). Data replication issues become easier since both boxes are on
> the same LAN too.
Local Director can have problems with servers that are up, but that are
malfunctioning.  This behaviour can be more common than first inspection
reveals.

This is an inherent limitation of LD: it only knows about the network layer.
As long as the network is moving, LD thinks all is ok.

Don't get me wrong: LD is a life saver.  But it's an incomplete failover
solution.
> 
> I think this is something we should seriously consider.  I've been running
> the network for Vicinity's corporate site which has done in excess of 70
> million page views a month and this is *the* way to do it IMHO.
I concur.

On a more personal note, my time is starting to free up now.  I'll try to be
a little more active here.

Cheers.
-Dana
-*~'`^`'~*-,._.,-*~'`^`'~*-,._.,-*~'`^`'~*-,._.,-*~'`^`'~*-,._.,-*~'`^`'~*-,
Dana M. Diederich                       Phone: 501.855.7175
E-Mail: dana@realms.org                 HTTP:  http://realms.org/
Snail Mail: 19 Leicester Drive, Bella Vista, Ar 72714, USA
`Berkeley invented LSD and Unix, and I don't think that's a coincidence.'