[Spread-users] New to spread, some questions

Kashif Shaikh kshaikh at consensys.com
Mon Feb 7 11:49:39 EST 2005



> A simple protocol would have membership representatives exchange
> configurations when negotiating network merges, and then, provided
> that there were no conflicts, install new configurations as part of
> the membership protocol.
>
> One potential performance enhancement to the simple protocol is an
> exchange of hashes of configuration data, to detect the common case
> when a merge only includes daemons with the same configuration.

Yes I have seen this implemented on one cluster system where the membership
daemons exchange md5sums on their configuration data, and reject all
heartbeat packets that don't match the md5sums. Once all the md5sums
stabilize(i.e. all members have the same configuration) then the new
configuration installed atomically. This helps the membership daemon avoid
inconsistency across all members and makes it easy to minimize race
conditions when a member is being added/deleted.

The other method for adding/deleting a 'heavy-weight' node member to spread
is to shutdown all clients, shutdown spread, make config changes, and
restart everything. Not really a good idea, especially when you want to add
a spread-node so you can scale up for your application.

Kashif


>
> An audit would need to be done, however, to determine the feasibility
> of changes to Spread to support this.  The main concern is that Spread
> relies on the immutability of configurations for sanity checks, and
> for efficient, deterministic decision making.
>
> Cheers,
> Ryan
>
> On Sat, 5 Feb 2005 21:21:41 -0500, Scott Barvick
> <sbarvick at revasystems.com> wrote:
> > Ryan,
> >
> > I've seen the issue of resetting all of the daemons when you
> want to add a new one come up before.  We are also interested in
> having this be more dynamic.  Is it on anyone's list to do or any
> idea of the amout of work it would take to do it?
> >
> > Thanks,
> > Scott
> >
> >
> > -----Original Message-----
> > From: spread-users-bounces at lists.spread.org on behalf of Ryan Caudy
> > Sent: Thu 2/3/2005 9:59 PM
> > To: pedro smith
> > Cc: spread-users at lists.spread.org
> > Subject: Re: [Spread-users] New to spread, some questions
> >
> > You seem to understand this correctly.  The set of spread daemons
> > (identified by hostname/ip address) must be known when the daemons are
> > started.  The limit of 128 spread daemons (#define'd as
> > MAX_PROCS_RING) is fixed at compile-time -- you might be able to make
> > some changes to this, but I'm not sure if anyone has done so, and
> > there may be complications that I can't think of off the top of my
> > head.  As far as I know, there aren't currently plans to change
> > Spread's functionality in these respects.
> >
> > One thing to consider is that the 128 limit is a limit of the number
> > of daemons -- the number of clients is a large multiple of this.
> > Also, you can run more than one Spread network on an overlapping set
> > of machines -- they simply must use different ports, or different
> > multicast/broadcast addresses.
> >
> > Cheers,
> > Ryan
> >
> > On Thu, 3 Feb 2005 18:46:54 -0800 (PST), pedro smith
> <und_pep at yahoo.com> wrote:
> > > Hi there,
> > >
> > > I just found Spread and I am very interested in learning more
> about its
> > > capabilities. I am considering using something like spread in
> a financial
> > > system. However I am worried about two issues. First, it
> seems that the list
> > > of machines participating in a spread segment must be know at
> system boot
> > > time. Is this right? Would I have to reinitialize every
> spread daemon in the
> > > system to add a new machine? Second, do I understand
> correctly that there is
> > > a hard limit of 128 machines in a spread network? If so, are
> there any ways
> > > around this limit, perhaps by bridging separate networks?
> > >
> > > Thanks in advance for any help,
> > >
> > > -pp
> > >
> > >
> > >  ________________________________
> > > Do you Yahoo!?
> > >  Yahoo! Search presents - Jib Jab's 'Second Term'
> > >
> > >
> > > _______________________________________________
> > > Spread-users mailing list
> > > Spread-users at lists.spread.org
> > > http://lists.spread.org/mailman/listinfo/spread-users
> > >
> > >
> > >
> >
> > _______________________________________________
> > Spread-users mailing list
> > Spread-users at lists.spread.org
> > http://lists.spread.org/mailman/listinfo/spread-users
> >
> >
> >
>
> _______________________________________________
> Spread-users mailing list
> Spread-users at lists.spread.org
> http://lists.spread.org/mailman/listinfo/spread-users
> --
> No virus found in this incoming message.
> Checked by AVG Anti-Virus.
> Version: 7.0.300 / Virus Database: 265.8.5 - Release Date: 2/3/2005
>





More information about the Spread-users mailing list