Date: Mon, 8 Sep 1997 15:18:23 +0930 From: Greg Lehey <grog@lemis.com> To: Mike Smith <mike@smith.net.au> Cc: Simon Shapiro <Shimon@i-connect.net>, FreeBSD Chat <chat@freebsd.org> Subject: Re: lousy disk perf. under cpu load (was IDE vs SCSI) Message-ID: <19970908151823.35891@lemis.com> In-Reply-To: <199709080505.PAA01452@word.smith.net.au>; from Mike Smith on Mon, Sep 08, 1997 at 03:05:39PM %2B1000 References: <19970908145837.07934@lemis.com> <199709080505.PAA01452@word.smith.net.au>
next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, Sep 08, 1997 at 03:05:39PM +1000, Mike Smith wrote: >>> >>> However, Simon is close; the ESMD spec allows for a data clock of 25MHz >>> (the data separator is on the disk, not the controller, IIRC). >> >> Depends on the drive. > > What "depends" on the drive? The location of the data separator. Thinking about it, though, I'm not sure I remember correctly. I was pretty sure we only had one read data and one write data line, but I could have been wrong. Those cables were about .75" thick. > The ESMD spec lays out the maximum clock rate for the data, and the > separator has to be on the drive if you're going to claim to be > ESMD. Well, I'm not even sure if the 3330 is ESMD. We called it SMD (Storage Module Drive). What does the E mean? Extended? When did it come out? >>> The later ESMD disks were pretty hot performance-wise (eg. the >>> Fujitsu Super Eagle and its successorss), >> >> Sure, but they weren't exactly the kind of drive built in the >> mid-70s. The 3330 was the "standard" drive, and it had 30 sectors per >> track, 3600 rpm. How many kB/s do you get out of that? > > How many heads are you reading in parallel? One. > I've had blood out of similar units on my hands (and blood out of my > hands on similar units 8) and I get the distinct impression that > multiple-head read activity was the norm. They may have been in the environment you're talking about. They weren't at Tandem, and I'd guess that any multi-head stuff didn't come until the mid-80s. >>> Yup. And if someone can work out how to deal with the power >>> dissipation, a slab of pseudo-static RAM the size of a 3.5" drive will >>> probably be cost-comparable inside the next 5-10 years. >> >> Assuming the disk drive people don't continue to improve their >> devices. Round about the time of the last anecdote (early 80s), the >> head of Tandem's HPRC said something to the effect that we needn't >> worry too much about disks, because they would die out in the next 10 >> years, considering the way the price of RAM was dropping. > > Sure. I think your earlier point about the basic mechanical > limitations is quite valid though; there's a basic restriction inherent > in having to fling the head assembly around. Still... Sure. In fact, I'm astounded how much disk drives have improved in the last 15 years. In 1982, Tandem introduced a 540 MB CDC SMD disk drive, the disk drive for gluttons. It was a heap of shit. It weighed a ton, was a real pig to program (it went offline for over 30 seconds to perform its power on self test, and the system had to decide whether it was meditating or dead), and it wasn't overly reliable. We pardoned it because of its high capacity. It still had the same transfer rates and positioning times that I mentioned above, probably because of its 14" construction (the last of its kind. After that, we went to little 8" Fujitsus). Try and find a new production disk drive *anywhere* with only 540 MB, 30 ms positioning, 800 kB/sec transfer rate nowadays. By comparison, even the shittiest IDE drives are a dream. Greg
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19970908151823.35891>