[SSI-devel] Re: openSSI webView 0.1 released!
Brought to you by:
brucewalker,
rogertsang
From: Kilian C. <kil...@st...> - 2004-11-02 10:30:08
|
On Friday 29 October 2004 23:09, Brian J. Watson wrote: > Friggin' beautiful!! Your utility looks like a nice enhancement to > OpenSSI clustering! Thanks a lot. > One suggestion I have is to not separate the nodes into master and > slaves. A master/slave architecture implies that the master is a single > point of failure, which is not true in OpenSSI. It's better to call it > the "init node". I see you use this term in one of your detail views. I understand and agree with this.=20 However, from my user point of view, the initnode still appears as a single= =20 point of failure, since if it crashes, all the cluster collapses. Even with= =20 redundant root filesystem (DRBD or shared SCSI), loosing init nodes will=20 bring the whole cluster down. So, in my understanding, and from a user=20 point of view, init nodes have a special status in the cluster, which seems= =20 to be incompatible with a full peer-to-peer architecture. Contrary to openMosix, eg., or to an ideal peer-to-peer cluster, an openSSI= =20 cluster could not survive without its init nodes. > Another problem with the master/slave connotation is that OpenSSI is a > very peer-to-peer architecture. > One last problem is that master/slave could imply that only the master > node has a terminal, as in a Beowulf cluster.=20 I agree with this, that's why I'll change the (master nodes / slave nodes)= =20 denomination to something like (init nodes / nodes). Or would you prefer no= =20 distinction at all between nodes? Best regards, =2D-=20 Kilian CAVALOTTI Ing=E9nieur Syst=E8mes & R= =E9seaux Laboratoire STIX =C9cole Polytechniq= ue =4691128 Palaiseau Tel : +33 1 69 33 41 = 13 |