Date: Wed, 15 Nov 1995 10:30:04 +0500 (GMT+0500) From: "Serge A. Babkin" <babkin@hq.icb.chel.su> To: terry@lambert.org (Terry Lambert) Cc: terry@lambert.org, karl@mcs.com, current@FreeBSD.ORG Subject: Re: ISP state their FreeBSD concerns Message-ID: <199511150530.KAA03475@hq.icb.chel.su> In-Reply-To: <199511150434.VAA00712@phaeton.artisoft.com> from "Terry Lambert" at Nov 14, 95 09:34:53 pm
next in thread | previous in thread | raw e-mail | index | archive | help
> > > I was thinking more in terms of an "email database". It wants the write
> > > guarantees to be honored so the locking protocol functions correctly.
> >
> > The closing of file may be used as an implicit "sync" request for this file,
> > IMHO it should help for UUCP-like locking.
> > Or if they use fsync() it is an explicit request to "sync" the file. Are
> > there some other locking methods ?
>
> Well, NFS lockd, for one.
I'm sorry, I said unclear. I meant the file-based "implicit" locking methods.
I think lockd must make synchronization anyway and it must flush the caches.
> > > The other being lack of long file name support and DOS's inability to
> > > operate multiple I/O instances and/or drives as multiple session
> > > instances and issue async requests.
> >
> > Async requests need more buffers and the main DOS idea is "save this 1K
> > or die".
>
> Tell that to Novell; they do client caching. 8-).
>
> > > DOS has the same problem with NetWare or SMB, both of which are
> > > request/response in nature. Another request can't be issued until the
> > > previous one is satisfied. NetWare has fixed windows for binary downloads
> > > ("packet burst") to partly alleviate the problem in trade for congesting
> > > the bejesus out of your network in certain circumstances.
> >
> > But the syncronous writes in NFS makes wery big overhead. The write throughput
> > for DOS and different networking systems is like (on empty network):
> >
> > Netware - about 700 K/s (may be Netware client has more buffers ? but it takes
> > less of memory than NFS client)
>
> Nope. Packet-burst. Effectively, async writes for 32 or 64k. Fixed
> window, needs an ack every burst run.
OK, why it is good for Netware and bad for NFS ?
> > BSDI NFS - about 300K/s
> > HP/UX 9.04 NFS - about 200K/s
> > SCO NFS - about 80K/s
> > FreeBSD NFS - about 12K/s
>
> One wonders what BSDI does... maybe implements pcnfsd in kernel code?
Nope, I have experimented with pcnfsd from BSDI compiled in FreeBSD. I know
no details of PCNFS protocol, but if every request neeeds additional request
to pcnfsd it's IMHO strange. BTW, I have experimented without pcnfsd at all,
with -n key of mountd and results were the same.
May be BSDI uses asyncronous mode by default.
> > FreeBSD immediately writes every block it gets from DOS and reports about it
> > only after write is completed.
>
> Yep. File-based IPC won't work any other way.
>
> Consider a print job that you cause the file to be deleted after it's put
> in the queue -- a lot of email systems, like CC:Mail, do this.
Explain me please where is the problem ? I see none. You run printing,
it gets vnode of file, opens it and starts copying into its spool. After
this you gat the same vnode and issue an "unlink" request. The vnode gets
unlinked but it is still in-core with nonzero number of "opens" and thus
it will not be removed until the print system closes it. I see no place here
where immediate reporting or windowed writes can crash anything.
> > > The easy answer is: don't use DOS, it sucks.
> >
> > Yes :-( But the numbers above is one of reasons why people are using Netware,
> > not NFS.
>
> Well, these figures were current as of last year...
>
> You can license NUC (NetWare UNIX Client) source code for $100,000.
> You can license NWU (NetWare 4.x for UNIX Server) source code for $150,000.
>
> Or you can license both + UnixWare source code for $250,000 (hmmmm... I
> wonder what value UNIX source code has... $250,000 - $150,000 - $100,000.
> 8-) 8-)).
>From the point of view of end-user the cost of binary licence is much
different from these numbers. The licence per 1 user for NW 3.12 costs
about $50. For example SCO is not cheaper.
> With IPX stack latency corrected (mostly an artifact of the Streams
> implementation on UnixWare), NWU will outperform Native NetWare on
> identical hardware. By as much as 30%.
BTW will UnixWare+NWU binary license be comparable with native NetWare
by cost ?
> > > I think the issue was multple news reader processes and I/O being
> > > bottlenecked by requestst being queued for too few service engines.
> >
> > OK, sorry... The low DOS+NFS performance is one of my problems and I ask
> > about it when I see something about NFS performance. :-)
>
> Change the PCNFS server, and if possible, the PCNFS client.
I know that different PCNFS clients make difference in performance but I
never throught that pcnfsd may change the performance. Its manual page
says that it is used for logging in and out only.
Serge Babkin
! (babkin@hq.icb.chel.su)
! Headquarter of Joint Stock Commercial Bank "Chelindbank"
! Chelyabinsk, Russia
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199511150530.KAA03475>