home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.apollo
- Path: sparky!uunet!caen!sdd.hp.com!apollo.hp.com!netnews
- From: ced@APOLLO.HP.COM (Carl Davidson)
- Subject: RE: Need Some Advice...
- Sender: usenet@apollo.hp.com (Usenet News)
- Message-ID: <BxxuH7.GKu@apollo.hp.com>
- Date: Thu, 19 Nov 1992 00:53:31 GMT
- References: <BxxJo6.3KH@apollo.hp.com>
- Nntp-Posting-Host: watson.ch.apollo.hp.com
- Organization: Hewlett-Packard Company, Chelmsford, MA
- Keywords: atr
- Lines: 91
-
- In article <BxxJo6.3KH@apollo.hp.com>, giza@apollo.HP.COM (Peter E. Giza) writes:
- |> Herb Peyerl Writes:
- |> |>If you want to make the assumption that IP packets are encapsulated within
- |> |>DDS packets, then in order to make HP-UX reside on ATR, HP must then have
- |> |>ported DDS to HP-UX on 400's and 700's... That must mean 700's have
- |> |>node-ids. Somehow I doubt that.
- |>
- |>
- |> Just for the record. The ATR cards are same with the exception of
- |> few changes for the HP. The packets are encapsulated in a 73 byte
- |> DDS header just like good ole Domain, this is all done at the driver
- |> level for the card. All of the DDS-like behaviour is from the driver
- |> there is not a complete DDS port, just enough to make you pregnant.
- |>
- |> -peg
- |>
- |> Peter E. Giza HP DCE T&D ~:@)
-
- Actually, the ATR card used in Snakes is *exactly* (I know, I spec'd it)
- the same one used in the old DPCI-Ring product. That means it's the same
- board as used in the DN4XXX-series, but with the Aegis boot PROM removed
- and a node ID inserted in the socket that was always there (way back in
- the REAL old days, you couldn't even boot an Apollo node without a
- network card in it because then it wouldn't have a node ID to generate
- UIDs from, but somewhere along the line we got smart and started putting
- them on the system board, but that's another story...).
-
- As to IP encapsulation on ATR, Pete is right, the entire basic DDS header,
- all *70* bytes of it, is there. It has to be or I'd have broken real
- Domain nodes. Right after the DDS header is a little 12 byte header called
- the DR header that Domain TCP uses for its own purposes. After that comes
- a conventional IP packet.
-
- I *didn't* port DDS to HP-UX in order to get ATR support into it. That would
- have been an extremely time-consuming effort with very little payoff. What I
- did do is put enough support in the driver to be able to answer lcnode
- (ask_who, ask_who_notopo, and ask_rem_who) requests, as well as ask_time,
- ask_bldt, ask_diskless, and ask_node_root. lcnode I *had* to support.
- If I didn't, I'd again break the existing rings these nodes were destined
- to go in to. The rest I put in because of the following scenario, which
- in my opinion is very common:
-
- 1) lcnode
-
- long list returned including some uncataloged nodes
-
- 2) bdlt -n NNNNN
-
- where NNNNN is some uncataloged node's node id
- in this case, it turns out to be an HP-UX node, so we
- return the same string which would be returned by a
- 'uname -a' command
-
- 3) ctnode foobar NNNNN -root
-
- this puts the node into the Domain NS Helper database
- the next time it is queried with an lcnode
-
- I attempted to violate the "Principle of Least Astonishment" as little
- as possible, but as Pete put it, we are "a little bit pregnant." There
- are places where this scheme breaks down, such as the 'lcnode -from'
- command when run from the other side of a Domain router or when
- the Domain Automount Daemon (damd) tries to make an NFS mount point in
- the network root for a node that's already been cataloged this way, but
- I figure it's still better than a poke in the eye with a sharp stick.
- ATR on HP-UX is only intended to help customers transition from Domain
- on ATR to HP-UX on a better network, either FDDI or Ethernet (let's please
- not have a religious argument about how Ethernet isn't a better network than
- ATR; the market has already decided that one...), not to be used as the
- core network of new installations.
-
- Well, this got kinda long, but I've been watching this argument fester on
- the net for a while and I just couldn't stay out of it any longer (fools
- rush in...). I hope that my explanation helps clear up any questions that
- may have been lingering. Even if you don't agree with my implementation
- choices, at least now you have some idea why I did it the way I did.
-
- BTW: I'm not the person who maintains this driver nowadays. That's done
- in another division by someone who still does lan drivers for a
- living, so yelling at me to "FIX IT" won't have much effect. It's
- not that I'm not sympathetic, but I don't even have access to the
- code anymore.
-
- Best regards to all of you who still love Domain. With any luck we'll have
- something as good again someday.
-
- --
- Carl Davidson (508) 436-4361 |
- Chelmsford System Software Lab | Microkernels: Where less is more.
- The Hewlett-Packard Company |
- DOMAIN: ced@apollo.hp.com |
-