DB design document

[Al B. Snell]

On Fri, 22 Dec 2000 linas@linas.org wrote:

> > No way - RPC is far more lightweight than CORBA.
> 
> I am an rpc novice ... so excuse me, but what about the overhead of
> having to contact 3 or 4 different servers, e.g. portmapper, statd,
> rpciod, and all that?  My impression was that this wasn't
> a particularly speedy operation.

All but one of those are NFS daemons!

rpciod is a seperate process to handle waiting for block RPCs so the NFS
server doesn't have to block.
statd pings statds on other machines to see if they're up and keeps a
database of host availability.
lockd keeps track of NFS file locks.

...only portmapper (now known as rpcbind) is an RPC service; the others
just *use* RPC. rpcbind

> By comparison, in 1997, I measured 2-8 milliseconds per corba call
> (latency), over a 10base2 (coax cable) ethernet, betwen a 486 and
> a pentium-133, using crappy ne2000 ethernet cards.   
> See http://www.linas.org/linux/corba.html for details.

RPC over UDP can issue a call and get the response in the time it takes to
send a ping packet!

> This aint great performance, but it aint terribly slow; acceptible,
> I'd say, for lightweight duty.  And I beleive that rougly similar
> figures would accrue to http/apache.  I find it hard to beleive that
> rpc would do a whole lot better. 

I don't have a benchmarking tool handy, the best I can do is:

alaric@lust$ time rpcinfo -n 111 -u www.demon.net 100000 4
program 100000 version 4 ready and waiting

real    0m0.024s
user    0m0.001s
sys     0m0.015s

alaric@lust$ ping -nqc 1 www.demon.net
PING www.demon.net (194.159.254.213): 56 data bytes

----www.demon.net PING Statistics----
1 packets transmitted, 1 packets received, 0.0% packet loss
round-trip min/avg/max/stddev = 7.465/7.465/7.465/0.000 ms

...rpcinfo -u does a ping of a specified RPC service on a specified
host. The -n is a port number, skipping the rpcbind / portmap stuff.

Don't ask me why Demon run an unfirewalled RPC server...

> 
> --linas
> 

ABS

-- 
                               Alaric B. Snell
 http://www.alaric-snell.com/  http://RFC.net/  http://www.warhead.org.uk/
   Any sufficiently advanced technology can be emulated in software