Several correspondants asked me to post a summary of my
escapades with a Wren VIII on our SparcServer 4/470.
Here is my initial posting (18 July):
-----------------------------------------------------------------
I have a Sun 4/470 server. Recently I bought a Seagate Wren 8
drive (synchronous SCSI, 5Mb ext transfer rate) and installed it
in the 4/470's internal SCSI card cage.
The new drive works, but operates v..e..r..y s..l..o..w..l..y.
My other SCSI drives copied a 1.4M test file in 3 seconds, but
the new Wren 8 took 30 seconds! Seems to me something is wrong here.
QUESTION: Was it a mistake to use the 4/470's internal SCSI cage?
I have since found out that the internal SCSI cage does *not* support
synchronous SCSI I/O, but even at asynchronous speeds, the drive
should run much faster than it is now. Any help much appreciated.
-----------------------------------------------------------------
Here are some things I have found out since:
1. Indeed, the Sun 4/470 does *not* support synchronous SCSI. Here is
a quote from my Sun engineer:
-----------------------------------------------------------------
Sun has several SCSI controller types the latest of which is the
SM controller and driver this combination supports synchronous
SCSI transfer under 4.1.1. The 4/470 uses a SCSI-3 VME controller that
was first implemented on the Sun 3 product line in 1986. This is also
known as the SI controller and uses the SI driver. Neither the hardware
nor driver is capable of synchronous SCSI, nor is capable of controlling
a drive as large as the Wren VIII.
-----------------------------------------------------------------
2. My vendor (from whom I bought the Wren 8) has not specifically tested
a Wren 8 on a 4/470, but claims to have several clients running this
configuration without problems (I have not verified this). He feels
that the engineer's quote above is "a Sun cop-out".
3. When I moved the Wren 8 from the Sun 4/470 to a 4/60 (Sparc 1), the
drive functioned at its proper speed. This would suggest that the
drive itself is OK.
4. When I moved another SCSI drive (a Maxtor 8760S) from the 4/60 to
the 4/470, it functioned at its proper speed. The 8760S runs
asynchronous only. This would suggest that an async drive on a
4/470 is OK.
MY OPINION SO FAR: Why did the Wren 8 operate so slowly inside the 4/470?
I don't know for sure yet, but it seems likely that the problem is an
"argument" between the Wren, the OS, and the SCSI controller: the Wren
and the OS wanted to run synchronous, but the SCSI controller wanted to
run asynchronous. This argument resulted in timeouts and retries
(but no weird console messages) resulting in very slow performance.
WHAT DO WE DO NOW? We still need a big disk on our 4/470, so we are
investigating IPI drives. A 2.5Gb (unformatted) disk can be had for
about $8350, about $3.50 per Mb (the Wren 8 cost us $2.30 per Mb).
We'll probably go this route, and leave the Wren 8 on the 4/60, where
it can operate synchronous under 4.1.1.
-----------------------------------------------------------------
David Mostardi Phone: (415) 643-6071
Systems Administrator FAX: (415) 643-5348
Mathematical Sciences Research Institute Email: david@msri.org
1000 Centennial Drive, Berkeley CA 94720
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:06:20 CDT