kirchner@uklirb.UUCP (Reinhard Kirchner) (06/16/88)
Hello, recently I once again learned about IBM disk drives which have two head-mechanisms ( Head-Disk-Assemblies , HDA ) per spindle. They have several drives of this kind, including the big 3380. Now: - why doesn't anybody else produce drives with two HDAs ? ( or does somebody whom I do not know ?, and then why ? ) . Is it not worth the effort, patented by IBM, or what is the reason ? R. Kirchner
root@didsgn.UUCP (didsgn) (06/18/88)
In article <2351@uklirb.UUCP>, kirchner@uklirb.UUCP (Reinhard Kirchner) writes: > > Hello, > recently I once again learned about IBM disk drives which have two > head-mechanisms ( Head-Disk-Assemblies , HDA ) per spindle. > They have several drives of this kind, including the big 3380. > > Now: > - why doesn't anybody else produce drives with two HDAs ? > ( or does somebody whom I do not know ?, and then why ? ) > . Is it not worth the effort, patented by IBM, or what is the reason ? > > R. Kirchner A few other manufacturer are selling drives with two HDAs. Here is a few: AMPEX is (or was a few years ago) building a FIX/REMOVABLE disk drive with 2 HDA. One HDA is taking care of the cartrige part of the drive (16 Mb unf.) and another is taking care of the fix part of it (16 to 64 Mb unf). It is possible to activates both HDA as a single one or (and this depend on the interface&software driver) as 2 inde- -pendant ones. IBIS is building a parallel transfer drive know as the 1400. It trans- fer data at 12Mb burst/10 Mb sustain per second. It's capacity is 1.4 Gb and it contains 2 HDA with 16 heads each. Both HDA are in sync. They are using 2 HDAs to solve an inertia problem of 1 HDA carrying 32 heads. Jean-Luc Chatelain ...!gatech!rebel!didsgn!jlc Standard disclaimer applies....
cprice@mips.COM (Charlie Price) (06/20/88)
> R. Kirchner > recently I once again learned about IBM disk drives which have two > head-mechanisms ( Head-Disk-Assemblies , HDA ) per spindle. > They have several drives of this kind, including the big 3380. A head disk assembly, or HDA, is the part that *must* be replaced when you have a head crash. It is an assembly that contains a spindle with the disks mounted on it, the magnetic read/write heads mounted on a movable actuator (called a head-arm) and (at least part of) a servo positioning system to move the head arm. Several IBM "disk drives" are boxes with more than one HDA in them. The earliest I'm aware of was the 3350. This used regular old HDAs. Two follow-on big disks were the 3370 and 3380. In addition to having two HDAs, they took the step of having two separate actuators (head arms) inside each HDA. These two arms don't share any tracks, they each access half of the disk surfaces on the HDA. [Having two independent heads over the same track isn't cost effective.] These actuators are positioned independently and there are independent read/write paths to each head arm. You can read or write on the separate head arms independently. The servo is fairly fast and the data transfer rate is 3 MBytes/sec. The IBM HDA is *huge*. The head arms are located on opposite sides of the vertically-mounted HDA (the disks are vertical). It reminds me, for some reason, of a Volkswagen engine. > - why doesn't anybody else produce drives with two HDAs ? > ( or does somebody whom I do not know ?, and then why ? ) > . Is it not worth the effort, patented by IBM, or what is the reason ? Several IBM plug-compatible manufacturers produce drives that are functionally compatible with the IBM drives. I worked for Storage Technology Corp (also StorageTek) for several years so I know the most about their drives. The StorageTek versions of the HDA use the same basic geometry, but their actuators are located side-by-side on one side of the disk platters. Other manufacturers do similar things. I believe the Amdahl drives, manufactured by Fujitsu, use 4 separate HDAs (something like the Super Eagle HDA). Why go to all this trouble? I don't know why IBM chose this approach, but one serious problem that they wanted to address was the problem of having too much data being serviced by a single head arm. Much of what IBM mainframes do is "data intensive" and these systems need speedy access to data. Disks are terribly slow compared to the processors; particularly physical positioning time. To have fast IO you want many disks and you want to keep them all busy. The larger the amount of data under a given head arm, the more likely that you will get into a situation where the system bottleneck is access to that data. For some reason, IBM still wanted to build physically large HDAs with a large number of platters and a lot of data so they made the math come out right by having two actuators. The original 3380 had 1.25 Mbyte HDAs, the 3380E has 2.5 MByte HDAs, and the current drives (number unknown) are 3.75 Mbyte HDAs. The current technology really favors building the same sort of system out of a larger number of smaller HDAs. It isn't clear to me that IBM's choice made technological sense. STC built the HDA with two actuators because at the time you couldn't sell a plug-compatible product to an IBM customer if it wasn't very similar to the IBM offering. Charlie Price cprice@mips.com {decwrl|ames}!mips!cprice MIPS Computer Systems / 930 Arques / Sunnyvale, CA 94087 / (408) 720-1700
mat@amdahl.uts.amdahl.com (Mike Taylor) (06/20/88)
In article <2440@winchester.mips.COM>, cprice@mips.COM (Charlie Price) writes: > > Two follow-on big disks were the 3370 and 3380. > In addition to having two HDAs, they took the step of having > two separate actuators (head arms) inside each HDA. > For some reason, IBM still wanted to build physically large HDAs > with a large number of platters and a lot of data so they made > the math come out right by having two actuators. > The current technology really favors building the same sort > of system out of a larger number of smaller HDAs. It isn't > clear to me that IBM's choice made technological sense. They probably had a manufacturing technology for the 14" platters that they didn't want to replace. I expect their sums said that using old platters and developing new head technology - the thin-film head - was the best economic choice. -- Mike Taylor ...!{ihnp4,hplabs,amdcad,sun}!amdahl!mat [ This may not reflect my opinion, let alone anyone else's. ]
cprice@mips.COM (Charlie Price) (06/21/88)
In article <8dly9ed7S71010boqkE@amdahl.uts.amdahl.com> mat@amdahl.uts.amdahl.com (Mike Taylor) writes: >In article <2440@winchester.mips.COM>, cprice@mips.COM (Charlie Price) writes: >> >> Two follow-on big disks were the 3370 and 3380. >> In addition to having two HDAs, they took the step of having >> two separate actuators (head arms) inside each HDA. >> For some reason, IBM still wanted to build physically large HDAs >> with a large number of platters and a lot of data so they made >> the math come out right by having two actuators. >> The current technology really favors building the same sort >> of system out of a larger number of smaller HDAs. It isn't >> clear to me that IBM's choice made technological sense. > >They probably had a manufacturing technology for the 14" platters >that they didn't want to replace. I expect their sums said that >using old platters and developing new head technology - the thin-film >head - was the best economic choice. >-- >Mike Taylor ...!{ihnp4,hplabs,amdcad,sun}!amdahl!mat > A good thought, and perhaps that is part of it. It turns out that they made a questionable call if that was the basis for the decision. I believe the 3370 uses the old industry standard 14" platters (it has been a while) but it is a lower density product than the 3380. Supposedly the original work on the 3380 suggested that the standard platters might not by stable enough to get the desired track density. The disks are 1) not perfectly flat in the first place and 2) deform some in a real HDA when they are clamped into a fixed stack at the inner diameter, heated up, and spun at 3600 RPM. You can tolerate a certain amount of "surface terrain" on a disk, but as your track density goes up the head flying height perforce gets lower and the sensitivity to surface variations increases. IBM decided to go ahead with standard platters because changing to a new kind of platter would cost a lot of money in tooling and such -- like $20Meg. After a while, they decided they had to spend the $20Meg and thereby slipped the delivery of these disks (like 15 months or more). The platters they use are quite thick -- like 1/4 inch. Perhaps they wanted to take the conservative route and this ended up being more expensive than they expected. On the other hand, these new platters are probably a fair bit better for reliable high capacity drives. IBM certainly didn't lose money on 3380. StorageTek was quite worried about this substrate change for a while because they just didn't have the resources to create a whole new substrate (STC mostly buys disks from other people, like Dysan.) In the end, STC was able to use the standard disks and do reasonably well. HDA reliability related to disk stability has never been a problem in their 8380 HDAs. Charlie Price cprice@mips.com {decwrl,ames}!mips!cprice
)) (06/21/88)
In article <2351@uklirb.UUCP> kirchner@uklirb.UUCP (Reinhard Kirchner) writes: >Hello, >recently I once again learned about IBM disk drives which have two >head-mechanisms ( Head-Disk-Assemblies , HDA ) per spindle. >They have several drives of this kind, including the big 3380. >Now: >- why doesn't anybody else produce drives with two HDAs ? > ( or does somebody whom I do not know ?, and then why ? ) I believe that the CDC Hydra (good name!) has four HDA's, but I could be mistaken. On a 27Mbyte/second channels no less. Yipes! >. Is it not worth the effort, patented by IBM, or what is the reason ? The main reasons are that each additional set of heads, head electronics and data pathways is *very* expensive, and that you don't gain anything unless your controller and/or drivers are very intelligent, and usually only under pretty high load. Personally, I'd normally prefer to have twice the capacity on two separate but overlapping seek drives than occasionally getting up to twice the thruput on a single disk. Mind you, a Hydra would make a nice swapping or paging disk - almost as good as a head-per-track disk or a drum. -- Chris Lewis, Spectrix Microsystems Inc, Phone: (416)-474-1955 UUCP: {uunet!mnetor, utcsri!utzoo, lsuc, yunexus}!spectrix!clewis Moderator of the Ferret Mailing List (ferret-list,ferret-request@spectrix)
cprice@mips.COM (Charlie Price) (06/22/88)
In article <2440@winchester.mips.COM> cprice@winchester.UUCP (Charlie Price) writes: >The original 3380 had 1.25 Mbyte HDAs, the 3380E has 2.5 MByte HDAs, >and the current drives (number unknown) are 3.75 Mbyte HDAs. I meant, of course, GigaBytes. Charlie Price cprice@mips.com {decwrl|ames}!mips!cprice MIPS Computer Systems / 930 Arques / Sunnyvale, CA 94087 / (408) 720-1700
srg@quick.COM (Spencer Garrett) (06/22/88)
A big reason for not putting in two HDA's is that you would then have to align them. One big advantage of non-removable media is that as long as the heads don't move (WRT one another) it doesn't matter where they are. You just torque them down and format the drive. It is possible to get around this with embedded-servo information, but then changing heads requires a (short) seek, which does nasty things to filesystem throughput. (The cylinders get very small, equal to one track apiece.) It's a shame, too, because having multiple access arms could otherwise give you the best of both worlds (one big drive, for allocation efficiency and flexibility, versus several drives with the ability to seek independently).
rick@svedberg.bcm.tmc.edu (Richard H. Miller) (06/24/88)
In article <2440@winchester.mips.COM>, cprice@mips.COM (Charlie Price) writes: > > > The current technology really favors building the same sort > of system out of a larger number of smaller HDAs. It isn't > clear to me that IBM's choice made technological sense. > STC built the HDA with two actuators because at the time you > couldn't sell a plug-compatible product to an IBM customer > if it wasn't very similar to the IBM offering. Well, one reason could be floor space and environmental conditions. In large mainframe shops, as you pointed out, the job is data intensive and many shops require many gigbytes of storage. It is more economical of floor space, power and air conditioning to have a smaller number of very large HDA's than a large number of smaller HDA's since in many cases the drive external to the HDA would require about the same amount of material. Another reason is architectual limitations of the O/S (number of addresable drives on a channel, number of channels, etc.) It is easier to modify the I/O routines to support higher density drives, multiple arms and the like than to redesign the entire IO complex to add additional addressing. Richard H. Miller Email: rick@svedberg.bcm.tmc.edu Head, System Support Voice: (713)799-4511 Baylor College of Medicine US Mail: One Baylor Plaza, 302H Houston, Texas 77030
" Maynard) (06/25/88)
In article <2451@winchester.mips.COM> cprice@winchester.UUCP (Charlie Price) writes: >>The original 3380 had 1.25 Mbyte HDAs, the 3380E has 2.5 MByte HDAs, >>and the current drives (number unknown) are 3.75 Mbyte HDAs. ^^^^=3380K >I meant, of course, GigaBytes. Each HDA in a 3380 (of any series) has two actuators, each with an equal amount of cylinders accessible. The capacity is increased by increasing the number of cylinders: the 3380A/B/Ds have 885 cylinders of 15 tracks of 47476 bytes each, the 3380E/Js have 1770 cylinders, and the 3380Ks have 2655 cylinders. Each 3380 box has two HDAs, or four logical volumes; you can hang four boxes on a string, for a total of 16 volumes of over 30 GB per string. (Side note: my not-too-small shop only has 17 GB total of 3350 space...) -- Jay Maynard, EMT-P, K5ZC...>splut!< | Never ascribe to malice that which can uucp: uunet!nuchat! | adequately be explained by stupidity. hoptoad!academ!uhnix1!splut!jay +---------------------------------------- {killer,bellcore}!tness1! | Birthright Party '88: let's get spaced!