|
"Matthew S. Crocker" wrote:
>
> On Tue, 2 Oct 2001, Julian Anastasov wrote:
>
> I'm seeing this behavior with my LVS cluster. Here is the scenario
I saw similar things with my VS-DR setup although I forget the exact details.
I decided that the problem was that the ActiveConn/InActConn represented
the ipvsadm table's view of the number of connections, which for VS-DR
is inferred from time-outs and watching incoming packets, but is not
neccessarily a good reflection of the number of connections for each
realserver. ipvs will keep this ipvsadm value balanced.
What your realserver is seeing is another matter. I noticed
that when I had a pair of realservers that could run with 2.2.x or
2.4.x kernels (same application), that the number of connections
the realservers were was always higher for the 2.2.x kernel (I think, and
I seem to remember about 30-50% higher). I wondered if FIN_WAIT etc
were different for the two kernels.
I decided that the problem was that the realservers weren't indentical and
it wasn't worth spending a lot of time on. Your imbalance seems a lot worse
than I would have expected if this is the explanation.
Joe
> LVS - DR
> 1 VIP
> 3 Real Servers,
>
> I'm using lc scheduler forwarding :25, :110, :143,
>
> The first qmail server has processed 1500 viruses (qmail-queue-scanner)
> and 22k rblsmtpd blocks in a 25 hour period. the other two servers are
> 400 viruses and4k rblsmtpd blocks. Basically the the primary server is
> getting 3/5th the load and the other two are 1/5th each.
>
> [root@lvsd-1 /root]# /sbin/ipvsadm -v
> ipvsadm v1.19 2001/08/27 (compiled with getopt_long and IPVS v0.9.3)
>
> [root@lvsd-1 /root]# /sbin/ipvsadm -l
> IP Virtual Server version 0.9.3 (size=65536)
> Prot LocalAddress:Port Scheduler Flags
> -> RemoteAddress:Port Forward Weight ActiveConn InActConn
> TCP mail.crocker.com:smtp lc
> -> cw1.crocker.com:smtp Route 1 132 16
> -> cw2.crocker.com:smtp Route 1 131 30
> -> cw3.crocker.com:smtp Route 1 131 21
> TCP mail.crocker.com:pop3 lc
> -> cw1.crocker.com:pop3 Route 1 3 3
> -> cw2.crocker.com:pop3 Route 1 3 2
> -> cw3.crocker.com:pop3 Route 1 3 2
> TCP mail.crocker.com:imap2 lc
> -> cw3.crocker.com:imap2 Route 1 1 0
> -> cw1.crocker.com:imap2 Route 1 0 0
> -> cw2.crocker.com:imap2 Route 1 0 0
>
> Funny thing is it shows the connections are balanced but it seems the
> qmail server #1 is doing a LOT more work. I know that possibily only RS
> #1 is getting the spammer connections and the virus e-mails but it should
> average out over time.
>
> -Matt
>
> >
> > Hello,
> >
> > On Tue, 2 Oct 2001, Josh Marcus wrote:
> >
> > > Hello--
> > >
> > > I'm using LVS to serve a few thousand domains, but I can't quite
> > > figure out how to get the behavior I need. Here are my two
> > > problems:
> > >
> > > 1) No matter what the weights, lvs (under wlc) always seems to
> > > send packets along to the first realserver (as listed in ipvsadm).
> >
> > In some cases this is the right thing to do
> >
> > > This is a serious problem for me, because ...
> > >
> > > 2) I don't see how I can setup LVS to load balance all of the domains
> > > as if they were all a single ip. In my ideal world, I would have a
> > > single entry *:80 that would forward all of our ips at port 80 to
> > > our set of realservers, and load balance all requests coming in. The
> > > way LVS is working for us now, the vast majority of all of our requests
> >
> > May be here is the problem, what is "our ips"? If these boxes
> > send enough requests to keep each connection reused then it is possible
> > only one RS to be used (look in the HOWTO why flood tests from clients
> > near the cluster make strange results, i.e. only one RS used or
> > conn/sec rate equal to 0). But you have to understand some details, how
> > LVS works because the problem can be enough complex.
> >
> > > are going to the server that is for some reason being listed first.
> > > Only sites with heavy traffic get pushed along to the other servers.
> > >
> > > Any advice? We're running version 0.9.12 -- the kernel module might be
> > > 0.9.15.
> >
> > The first advice is to upgrade, of course :) This is a LVS from
> > more than one year. In any case, if you want more help you have to
> > show all details from your setup and settings. WLC is the mainly used
> > method, I can't believe that it could be broken.
> >
> > > --j
> >
> >
> > Regards
> >
> > --
> > Julian Anastasov <ja@xxxxxx>
> >
> >
> > _______________________________________________
> > LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
> > Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
> > or go to http://www.in-addr.de/mailman/listinfo/lvs-users
> >
>
> --
> ----------------------------------------------------------------------
> Matthew S. Crocker
> Vice President / Internet Division Email: matthew@xxxxxxxxxxx
> Crocker Communications Phone: (413) 587-3350
> PO BOX 710 Fax: (413) 587-3352
> Greenfield, MA 01302-0710 http://www.crocker.com
> ----------------------------------------------------------------------
>
> _______________________________________________
> LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
> Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
> or go to http://www.in-addr.de/mailman/listinfo/lvs-users
--
Joseph Mack PhD, Senior Systems Engineer, Lockheed Martin
contractor to the National Environmental Supercomputer Center,
mailto:mack.joseph@xxxxxxx ph# 919-541-0007, RTP, NC, USA
|
