Load Balancing Questions/Concerns

Re: Load Balancing Questions/Concerns

From:

Date:

Time:

>I am currently using QuickDNS under the 15 day trial license, and am
>attempting to setup a load balancing system, so that I can see how it
>works, and choose whether or not I want to go this route. During my
>tests, I have noticed something, and am wondering if I might be
>performing the test wrong, or whether there is another solution for it.
>Here's what happens:
>
>I setup QuickDNS with two WebStar Servers on a pair of brand new G3's (I
>am actually running the QuickDNS on a PowerMac 7200). I have the
>WebStar QuickDNS responder plugin installed in each copy's plug-in
>folder. If I start up my own (development) computer and run Netscape as
>my browser, and have QuickDNS logging set to "debug" so I can see the
>queries from my browser. If both servers are up and running initially,
>it's a "6-of-one, half-dozen-of-the-other" type scenario. My browser
>queries my DNS server, and the DNS server routes to one server or the
>other. If I then disconnect the server that holds the browser's
>connection, and attempt to reload the same site, the browser never
>connects to the alternate server, and it appears as though the server is
>completely down. My browser never sends another query. Even after
>quitting the browser and relaunching, it still does not send an
>additional query. Even if I quit Netscape and launch Explorer, I never
>see another query come across, and I have to restart my computer in
>order to then get my browser to connect to the site via the alternate
>server.

What value did you put in the TTL field for the load balancing record? This
is how long (in seconds) the record *should* be held in memory by Open
Transport's resolver. Of course, it's possible that the resolver is holding
the record longer than it should.

How long did you wait between disconnecting the web server and retrying?

>Now, if I perform a different query to another site after my failed
>attempt to access the disconnected server, then attempt to connect again
>to the original site, I notice in QuickDNS's log file that another query
>is sent for the original site, but, the connection to the alternate
>server is still never made.

That's a little odd and suspicious.

>Is this normal? I guess that I expected an "on-the-fly" server
>switching.

There is no true "on-the-fly" fault tolerance. There are several methods to
come close, though:

1) QuickDNS Pro's intelligent lb/ft solution, when used with a very low
TTL, comes very close. There are similar solutions for other platforms.
However, it is possible for a misbehaving resolver to cache a down server's
address for too long.

2) A relay server can perform "on-the-fly" load balancing and fault
tolerance across several web servers, but this puts all your traffic
through the relay server. If it has a problem, your whole site disappears,
even though your web servers are fine.

>After all, if someone accesses one of my hosted sites early
>in the day, then leaves their browser running as they (for example)
>download a file from somewhere else on the internet, and a few hours
>later come back to the site on my server... if I have that server down
>for maintenance, it will appear to end user as though the entire site
>is down and inaccessible. They will not know instinctively to restart
>their computer to access the site. Now, if this happens to be one of my
>clients (for whom the site was created), who is just accessing their
>site as they regularly do, I can expect a phone call wanting to know why
>they cannot access their site. If I tell them to restart their computer
>to access it, I am sure that I will get a tongue-lashing.
>
>Please help alleviate my fears about this.

By design, your TTL for the lb record should be set low (300 or lower) to
minimize this problem. But the QDNS fault tolerance system is susceptible
to misbehaving resolvers, as described above.

>Another question... Just though of this while I was writing this
>letter, so I have not had a chance to try this out... If I select
>"Refuse Connections" on one server, will that stop the WebStar Responder
>from sending its report to QuickDNS, and cause a re-route to the other
>server?

Sorry, this is a StarNine product, so I'd have to test it to find out, same
as you. I would hope it would stop communicating with QuickDNS Pro when set
to refuse connections.
____________________________________________________________________
Chris Buxton Men & Mice
cbuxton@menandmice.com http://www.menandmice.com

Sales:
888/4-DPOINT

Support:
858/452-3696