[1315] in Hotline Meeting
Re: pc rvd on seurat has constant 'sector not found errors' when written to
daemon@ATHENA.MIT.EDU (Jerome H Saltzer)
Fri Aug 3 14:28:51 1990
From: saltzer@src.dec.com (Jerome H Saltzer)
Date: 3 Aug 1990 1127-PDT (Friday)
To: smyser@ATHENA.MIT.EDU
Cc: hotline@ATHENA.MIT.EDU, pete@ATHENA.MIT.EDU, crlstaff@ATHENA.MIT.EDU
Cc: jhs-out@ALLSPICE.LCS.MIT.EDU
Reply-To: Saltzer@mit.edu
In-Reply-To: Message of Fri, 03 Aug 90 11:11:06 EDT
Rob,
> (I copied jerry in because he might still be around and might
> remember what-all was involved in this stage of the process....)
Happy to help, though I'm at a sufficient distance that I can't come
over to look at it right now.
> The error was always 'sector not found writing drive X:'
The DOS repertoire of possible error codes is a kind of strait jacket,
so PC/RVD had to make some arbitrary mappings from problems to error
codes. The "sector not found" error really means "RVD timeout",
which in turn means that the client timed out before the RVD server
acknowledged the write. The timeout is measured in seconds, and
the client retries a few times. This error usually shows up when
a server or gateway goes down after spinning up a pack. The next
time the client tries to read or write on that pack, it gets no read
response or no write acknowledgements, and the resulting timeout
produces that error code. But the case you have is clearly a different
scenario.
Is the error message immediate, or does there seem to be a several-
second pause before it appears?
> What is actually involved in preparing the rvd packs for use as a
> DOS filesystem? I mounted each drive in exclusive mode and ran
> 'mkrfs', which said it was doing it (though it didn't spend any time
> on it that I could tell) and spun the drive down.
Yes, that is the correct procedure. Making a DOS file system is
very quick; only the boot sector and two copies of a File Allocation
Table need to be written.
Does the problem appear with both "small" and "large" packs? Thanks
to a DOS design blunder, there are two pack formats, "large" and
"small". If all the packs you made are large, you might try making
a small one. (I think that the boundary between "large" and "small"
may be somewhere around 12 Mbytes. In any case a one Mbyte pack
is for sure "small".)
> I experimented with copying dos to rvd. Files up to 10000 bytes copied fine.
> Files larger than 10000 bytes failed consistently.
10240 bytes is a magic number for PC/RVD; it writes bursts of twenty
512-byte packets, pauses for complete acknowledgement of the burst,
resends anything that doesn't get acknowledged, and then goes on
to the next burst. (It is possible that it isn't as clever as it
should be, in which case it may stupidly resend the entire burst.)
> pc rvd on seurat
Could someone please tell me what exactly is "seurat"? What kind
of machine is it? Is it a brand of RVD server that has never served
PC's before? Or a just-installed server? Or has it been for some
time successfully serving PC's with the packs that once were on Gaia?
If it has been working OK with PC clients, is this the first time
anyone tried to create a new PC file system on it, or has that
operation been successful on this server in the past?
One possibility, if this is the first pack creation attempt on a
new class of server: seurat may be a much faster processor that
is sending write acknowledgements back in a burst that is so rapid-fire
that the PC can't catch them all. If so, it might work to perform
the write through a gateway or two. (Are there still some RVD-equipped
PC's in the Sloan School that can be tried for the purpose?) Reads
may be similarly affected, or perhaps not; since a read response
is normally a larger packet than a write acknowledgement,the packets
arrive at a more leisurely pace.
Good luck; please let me know how things go.
Jerry
