1995-08-29 - Re: SSL trouble

Header Data

From: Mark <mark@lochard.com.au>
To: ab411@detroit.freenet.org
Message Hash: ccc6f7c08aa21d07ecb557e3f05ba6e701eabf31d5578ef87a503f40c4f5f36c
Message ID: <199508282326.AA32757@junkers.lochard.com.au>
Reply To: <199508281852.OAA22478@detroit.freenet.org>
UTC Datetime: 1995-08-29 00:25:13 UTC
Raw Date: Mon, 28 Aug 95 17:25:13 PDT

Raw message

From: Mark <mark@lochard.com.au>
Date: Mon, 28 Aug 95 17:25:13 PDT
To: ab411@detroit.freenet.org
Subject: Re: SSL trouble
In-Reply-To: <199508281852.OAA22478@detroit.freenet.org>
Message-ID: <199508282326.AA32757@junkers.lochard.com.au>
MIME-Version: 1.0
Content-Type: text


>Patrick Horgan <patrick@Verity.COM> writes:
>>I did a distributed scheme for something else that had two levels, a master
>>and a group of slaves.  Only the slaves talked to the master.  For this 
>>effort I think a variation of the idea would be better.  Have all of the
>>brutes contact the master, who will, in the first transaction assign them
>>to the next slave in a round-robin fashion.
>
>Why not just have the brutes pick a slave at random?  Of course, you need
>to give them a complete list of slaves to choose from.  But then the only
>difference between the master and the slaves will be that the master
>doesn't get any keyspace (it's got it all to begin with) and doesn't
>report any results upward.

Better to include in the clients a list of all slaves and have the initial
contact to a slave random, maybe weighted by network proximity, and either
have the clients cycle to each slave with each ACK, or have the slave TELL
the clients what server to ACK to next, based on slave to slave balancing.
Slaves could tune their pointers to faster slaves and transparently handle
crashed slaves

The idea is to have all the slaves working evenly, assuming they are on equal
nets and equal machines. If not then you can have the slaves tell clients to
only point to a weaker slave once in a while.

I dont think advertising the master is a good idea, better to have the slaves
talking to it only. A backup mirror master would be worthwhile too.

Web people would most likely have to communicate with the one central http
server as you have to Keep It Simple for them.

Comments?

Mark




Thread