[ZODB-Dev] ZEO and replication of BTree based objects

[Roché Compaan]

* Toby Dickenson <tdickenson at geminidataloggers.com> [2003-06-05 09:36]:
> On Thursday 05 June 2003 06:49, Roché Compaan wrote:
> > * Jeremy Hylton <jeremy at zope.com> [2003-06-04 22:37]:
> 
> > Yes I am aware of this. I was wondering if you would consider to force a
> > load on a client as soon as it receives an invalidation message based on
> > a configuration variable.
> 
> This will waste your precious bandwidth if an object changes multiple times 
> between it being read on one client. 

It is unlikely in this application. In this case our customer generates
renewal certificates for software licenses and the license type with the
shortest lifetime is 45 days.

> 
> > Although local reads are very
> > fast a client sometimes waits a long time for a bunch of objects that
> > was part of a big transaction.
> 
> Is ZEO utilizing your bandwidth efficiently?
> 
> The current ZEO implementation loads each object on demand, and waits for one 
> object to be fully retrieved before requesting the next. This could leave 
> your network underutilized if it has high *latency*. I wonder whether 
> pipelined reads would improve this.....

Most probably. We currently run OpenVPN between ZEO client and server
with LZO compression. I guess pipelined reads will cause more data being
compressed at a time before being transmitted.  

-- 
Roché Compaan
Upfront Systems                 http://www.upfrontsystems.co.za