henry@utzoo.UUCP (Henry Spencer) (01/11/84)
Chuck Kennedy observes:
Digital has announced the Digital Storage Architecture (DSA) which
utilizes their nifty new HSC50. The connection to the HSC50 is via the
Computer Interconnect (CI) bus which has a 70Mbit/sec bandwidth.
The HSC50 has some really neat features ...
One of the not-so-neat features is that the box on the other end which
actually runs the disks contains a cpu. A small cpu. I am told that for
some configurations it's a 730. This may limit throughput.
I only have two problems with the HSC50:
1) If you don't run VMS (i.e. UNIX), you will have to write DSA
code to run the HSC50. This is probably more complicated
than your average driver. Besides, you have to yank teeth
to find out all the information from DEC.
Bear in mind that the network architecture involved is approximately
that displayed by the UDA50 controller; I am told that Armando's driver
for that creature is about 30KB of code! "More complicated than your
average driver" is probably the understatement of the century.
--
Henry Spencer @ U of Toronto Zoology
{allegra,ihnp4,linus,decvax}!utzoo!henrygeoff@utcsstat.UUCP (Geoffrey Collyer) (01/11/84)
A DEC salesperson told me that the HSC50 contains an 11/23 or an 11/24 (I can't remember which). Geoff Collyer, U. of Toronto
shannon@sun.uucp (Bill Shannon) (01/11/84)
No, there is no 730 in the HSC50. I believe there are several PDP-11's and some special purpose processors just to move data. One of the big differences between the UDA and the HSC is that the UDA is attached directly to a local bus (the Unibus), whereas the HSC is really attached to a network (the CI). Talking to the HSC is more like writing a network protocol driver. You'd like to use that nice, fast CI bus to do other things besides just talk to your disk subsystem, like use it as a very fast, very local network (which it is). When I wrote the UDA driver I thought about trying to share some code with a (future) HSC driver and decided that the two beasts were different enough that it was not worth making the UDA suffer to fit into the HSC scheme of things. DEC actually has an "architecture" for the interface between "class drivers" (e.g. a general DSA disk driver) and "port drivers" (that actually handle getting the MSCP commands to and from the device). At the time it seemed like a lot of excess baggage (besides not fitting well into the UNIX kernel structure) but as more MSCP devices come along it may be worthwhile. Also, unlike something I read on the net, the shadowing feature of the HSC is to allow you to maintain two disks with the same data invisibly to the user, NOT write to tape. I don't remember but there may be a journally facility to log all transactions to tape. All this is based on information I had 1.5 years ago. Armando, correct me if I remember wrong or things have changed.
steveg@hammer.UUCP (Steve Glaser) (01/11/84)
<non blank line> I'd understood that the CI-780 and CI-750 interfaces to the VAXcluster assumed that system software uses the page tables like VMS does (the HSC50 is just part of the system, it hangs on the CI bus and talks to disks and tapes). The VAX architecture manual used to describe a number of "reserved for software" bits in the page table entries and Unix uses these bits differently than VMS does. Apparently, the CI series interfaces "know" about page tables, but assume those bits are used the VMS way. This would require changes to the 4.1 (or 4.2) paging system to be usable. I also think the DR-780 has the same problem. The new VM architecture of 4.3BSD may fix this (if it ever makes it). DEC's product version of Vax/Unix may also fix this. My understanding is based on a conversation with Bill Shannon at the Boston USENIX - DEC may have changed the hardware since then (It wasn't announced then anyway) [Armando, are you listening?]. In any case, it's going to take more work than just writing another driver, if only because it's another SBI (CMI) device so it doesn't fit into the model of an ordinary Unibus or Massbus device. Steve Glaser tektronix!steveg UUCP steveg.tektronix@rand-relay ARPANET/CSNET
leichter@yale-com.UUCP (01/16/84)
A HSC50 contains multiple F-11's; these are the chips in both the 11/23 and 11/24, which differ in the backplane: 11/23 is Q-bus, 11/24 is Unibus. (Actually, to be complete: 11/23 is 18-bit Q-bus, 11/23+ is 22-bit Q-bus.) It isn't particularly meaningful to ask whether the F-11's inside an HSC-50 are "11/23's" or "11/24's". While the DSA protocol may be complex, it has two advantages over usual disk-driver interfaces: It looks like a fairly high-level network protocol, not a series of random read-only, write-only, and read-write bits plus interrupts and timing constraints; and it is done, once and for all: New disks can connect to the HSC-50 without the host talking to it knowing that they are new...the HSC-50 protocol stays unchanged. -- Jerry