shekita@provolone.cs.wisc.edu (Eugene Shekita) (10/09/89)
I've found this discussion on *big iron* I/O very interesting, but I don't
have a clear picture in my mind of what a typical workstation SCSI
configuration looks like. Surely, there is someone out there who could fill
in the gaps, and I'd *really* appreciate it if they did. I think a lot of
other people would too. (Unfortunately, this is the kind of stuff that
generally gets ignored in Architecture 101 so please forgive my ignorance.)
From what I've read over the net, the way I view a typical SCSI configuration
is as follows:
------------- ----------
| Processor | | Memory |
------------- ----------
| |
================================== Main Bus
|
------------------- -------
| SCSI Controller |--- | Disk|
------------------- -------
Is this right, or what else is there? Other basic questions I have are:
1) How does DMA logic get added to the picture -- is it typically
a separate unit with its own buffer, thereby allowing full-word
transfers over the main bus instead of 8-bit SCSI words (ala Amiga)?
Or do DMA units typically have no buffer?
2) Dual-ported memory systems let DMA transfers go straight into memory
without tieing up the main bus, right? Roughly, how much cost does
a dual-ported memory system add, and is that why they're usually only
found on *big iron*? And while we're at it, what level of multi-porting
do some of these *big iron* machines have (a Cray, for example)?
3) Why does a VME bus help here? Do they allow bock-mode
transfers and that's why?
4) Is it becoming common for SCSI controllers to have a track-size
buffer of their own and always do track-size reads? (I am told
that Conner contollers have this feature.)
5) What does the corresponding picture look like for a *big iron*
machine?
6) Finally, any good reading references I could take a look at to
find out more?
thanks in advance -- Genedsleffel@ensub.Wichita.NCR.COM (Don Sleffel) (10/09/89)
In article <8774@spool.cs.wisc.edu> shekita@provolone.cs.wisc.edu (Eugene Shekita) writes: > >From what I've read over the net, the way I view a typical SCSI configuration >is as follows: Should be more like: > > ------------- ---------- > | Processor | | Memory | > ------------- ---------- > | | > ================================== Main Bus > | > ------------------- ------- |Host Adapter or | |File Processor | ------------------- | |--------------|-------------|-------etc. --------------- ------------ ---------------- |Disk Controller| |Tape | |Disk Controller | --------------- |Controller | ---------------- | ------------ | |--Disk Drive | |--Disk Drive | |--Tape Drive | |--Disk Drive |--Disk Drive | | etc. etc. > > 1) How does DMA logic get added to the picture -- is it typically > a separate unit with its own buffer, thereby allowing full-word > transfers over the main bus instead of 8-bit SCSI words (ala Amiga)? > Or do DMA units typically have no buffer? DMA logic along with buffers and/or file processing software resides on the file processor. This also allows effecient matching of SCSI transfer rates to the system bus transfer rate. Of course not all systems contain all of the hardware options to configure something this extensive, depending upon the speed and price of the system. > > 2) Dual-ported memory systems let DMA transfers go straight into memory > without tieing up the main bus, right? Roughly, how much cost does > a dual-ported memory system add, and is that why they're usually only > found on *big iron*? And while we're at it, what level of multi-porting > do some of these *big iron* machines have (a Cray, for example)? Not neccessarily. There are a lot of different reasons why you might or might not want to use dual ported memory and why you might or might not want to use the side opposite the processor for IO operations. Keep in mind there are a lot of different classes and application types of machines. "Big Iron" does not neccessarily imply a Cray. > 4) Is it becoming common for SCSI controllers to have a track-size > buffer of their own and always do track-size reads? (I am told > that Conner contollers have this feature.) Yes, as well as various forms of caching. SCSI used on "Big Iron" will probably have from 2 to 200 Megabyte of buffer space, depending upon the type of usage. ---------------------------------------------------------------------- Don Sleffel 3718 N. Rock Rd. Wichita, KS 67226 <d.sleffel@Wichita.NCR.COM> <{ece-csc,hubcap,gould,rtech}!ncrcae!ncrwic!ensub!dsleffel> <{sdcsvax,cbatt,dcdwest,nosc.ARPA,ihnp4}!ncr-sd!ncrwic!ensub!dsleffel> -------------- Contempt Breeds Contempt
jesup@cbmvax.UUCP (Randell Jesup) (10/11/89)
In article <196@ensub.Wichita.NCR.COM> dsleffel@ensub.Wichita.NCR.COM (Don Sleffel) writes: >In article <8774@spool.cs.wisc.edu> shekita@provolone.cs.wisc.edu (Eugene Shekita) writes: > --------------- ------------ ---------------- > |Disk Controller| |Tape | |Disk Controller | > --------------- |Controller | ---------------- > | ------------ | > |--Disk Drive | |--Disk Drive > | |--Tape Drive | > |--Disk Drive |--Disk Drive > | | > etc. etc. This can be done, but is rare (use of Logical Unit Numbers). Most are single drive/disk controller. >Not neccessarily. There are a lot of different reasons why you might or >might not want to use dual ported memory and why you might or might not >want to use the side opposite the processor for IO operations. Keep in mind >there are a lot of different classes and application types of machines. Dual ported memory is _expensive_ (and usually not dense), as well. >> 4) Is it becoming common for SCSI controllers to have a track-size >> buffer of their own and always do track-size reads? (I am told >> that Conner contollers have this feature.) > >Yes, as well as various forms of caching. SCSI used on "Big Iron" will >probably have from 2 to 200 Megabyte of buffer space, depending upon the >type of usage. There are two different type of buffers: buffers in the drive/controller and buffer in the host adaptor. Many drives (such as Quantum) now have cache on the controller (64K on the Q80s). Most DMA host adapters have some form cache or FIFO to cover main memory not being available for DMA. However, I've never heard of 2Meg of cache, let alone 200Meg. If you can afford that much memory, why buy disks? (hyperbole, no flames please) -- Randell Jesup, Keeper of AmigaDos, Commodore Engineering. {uunet|rutgers}!cbmvax!jesup, jesup@cbmvax.cbm.commodore.com BIX: rjesup Common phrase heard at Amiga Devcon '89: "It's in there!"