Date: Wed, 27 Mar 91 02:01:46 CST From: Al L Varney Subject: Re: Questions About New Service Being Installed Organization: AT&T Network Systems Well, this will be brief, since it's from memory; I've rearranged the previous discussion order somewhat. If you really MUST have more info., read the back issues of the Bell System Technical Journals. At least one issue was devoted to each switch. First, the obligatory note: ESS(tm) is a trademark of AT&T and 5ESS(tm) i0s a registered trademark of AT&T. The proper names are: 1 ESS Switch 1A ESS Switch 4 ESS Switch 5ESS Switch but I will use the obvious abbreviations below. In article goldstein@delni.enet.dec.com (Fred R. Goldstein) writes: > In article , rees@pisa.citi.umich.edu > (Jim Rees) writes... >> The 1ESS has relays in it, not to do the actual switching, >> but to switch ringing voltage and the like on to the loop. It mak) i0es a >> lot of noise, although nothing like a panel office! > The 1ESS (and the 1A, which uses a less antiquated processor) uses > reed relays to do the actual switching. They're vacuum-sealed, so > they're quieter than the old ones. I suspect that the 1 can do Caller > ID too, though Im not sure. > The 1 uses an antique CPU with ferrite sheet EPROMs and mag cores. The No. 1 ESS Switch indeed uses sealed relays for the switching fabric, but "reed relays" ??? Nope. The actual T/R path is through magnetic-latching relays, surrounded with some metal and a coil. Pulse the coil one way, the contacts close and REMEMBER to stay closed. Pulse the other way, the contacts open and REMEMBER to stay that way. No current is used to maintain either position. They are the size of a Christmas tree bulb and make little noise. The traffic-dependent noise you hear is the "wire-spring" relays that exist in the remainder of the switch, primarily in the Trunk/Junctor circuits. The 1E "CPU" consists of about 10 feet of circuits in a standard seven foot high "bay", arranged across from it's "mate" CPU. The CPUs run in lock- step, comparing results of every instruction. The memory is separate; for programs high office data, "EPROM" memory is formed from ferrite spots stuck to 6X12(?) inch sheets of aluminum. Typical office might have 40 feet of such memory, duplicated. Temporary (writable) memory is usually mag cores (32K by 23 bits + 1 parity per two foot bay). Program memory words are 37 bits wide, with an added 8 bits for Hamming-code parity (I believe automatic single-bit error correction is in the hardware). Architecture could be called "early RISC, messy" -- most instructions are one cycle or 5.5 microseconds. Capacity is roughly 35K lines. >> What's the difference between a 1 and a 1A (is it just the processor? >> Does 1A run Unix?) > The 1A goes to semiconductor memory. No. 1A ESS Switches use the same switching fabric as 1E. The circa 1973 processor is two CPUs in a six-foot wide frame, running in lock-step. Program and temporary memory are on separate busses, but look identical. Most modern version of memory puts 14 256K-by-14-bit units in a three-foot bay -- max of two bays per office allows at most four Mwords (12 Mbytes). Instruction set vaguely resembles an orthoganal version of 1E, with a typical instruction (24 or 48 bits wide) taking .7 milliseconds. Many shift/rotate/mask/insert options could be used, without added time, due to a complete 48-bit "barrel" shifter. For comparison, "clock speed" is 20 MHz; even though memory bus is 20 feet long, 700 nanoseconds can do a 48-bit read or 24-bit write. An overlapping dual-parity scheme is used on each memory word. Disk backup is used, with about 10 Mwords available. Original disk drives used 26(?) inch platters, with 100 fixed heads on each side, thus no seek overhead. Existing switches handle 90K lines. No fans in either 1E or 1A equipment, just ambient cooling. UNIX (also tm) grew up about the same time as 1A, but really !!! You don't switch 300K calls per hour on a non-MMU machine with UNIX. The OS is really a task dispenser with routines voluntarily giving up control every two or three milliseconds (sort of like Multi-Finder, no?). Much polling and processing takes place on a timed interrupt level, forced every five milliseconds. No other interrupts occur normally. >> What I'd like to know is what are 2 and 3ESS? > The 2BESS is a "suburban" office, built in the '70s to early '80s, > using (I think) reed relays like a 1A. It is basically a scaled down > version of the 1A, with a different processor. No. 2 ESS existed in 1968, so it's really scaled down from 1E. Every- thing was redesigned from the ground up, so there is essentially no shared circuitry with 1E. The processor was "strange". A 22 bit instruction word with one "long" 21 bit instruction or two 10-bit instructions; the remaining bit was = 1 only on words where transfers of control were expected to arrive. A bit-twiddlers toy. 10K lines?? (The processor was also used to drive the "Automatic Intercept System" [AIS], the one that says "The number you have reached, nyen-nyen-one-pause-six-six-six-six has been changed. The new ...". This was my first project with AT&T.) No. 2B ESS Switch was just a re-worked version of the No. 3 ESS processor with mico-code interpreting the original 2E instructions (but faster than the original hardware). I believe it gave a 50%+ increase in capacity. > The 3ESS is a very small analog office, of which very few were built > (ca. 1980). Don't know numbers, but there were quite a few in more "rural" areas. The "3A" processor -- no relation to the "3B" line -- was small and fast. I believe this was the first to use mico-code; 1E/1A/4E/2E don't. Don't know much else, except a whole office could fit in a semi-trailer (with MDF!) for emergency use. Several were tested on the trailer, shipped and then slid into place with attached air pallets. >> And what kind of hardware does a 4ESS have (I've never seen one)? > As someone else noted, the 4ESS is a different beast, a big digital > toll switch. Well, actually a Tandem switch, but BIG anyway. Same processor as 1A, with a totally digital switch. These are rated at 100K Trunks, 600K+ calls/hour. There was also (past tense, I believe) the No. 101 ESS switch, an early PBX. This used a processor from another project, with a unique PAM fabric (pulse amplitude modulation). Essentially, every line/trunk had an appearance on a single wire, with a different combination connected at an 8KHz rate. This allowed noise-less switching and many connections to a single line/trunk without loading problems. This same fabric was used in AIS, to allow many people to listen to "six" at the same time. "Six" was a single trunk connected to a repeating .5 second recording. Adjusting the volume on those trunks was boring!! Oh, oh, another long article. Maybe I'll do 5ESS later, Pat. In closing, I've had the pleasure of programming all of these switches except the No. 3 ESS switch. They all had something worth learning as far as designing to a particular goal. In most cases, the capacity of the switch drove the design. Al Varney, AT&T, Lisle, IL --- Corrections based on comments from att!citi.umich.edu!rees (Jim Rees) --- Jim, Jim> Very interesting! Thanks. You're welcome! Al> The No. 1 ESS Switch indeed uses sealed relays for the switching Al> fabric, but "reed relays" ??? Nope. Jim> Are you sure? I could have sworn they used reeds. I can't even imagine Jim> how you could build a magnetic latching relay without reeds. I goofed in the statement -- it should have been "...uses sealed REEDS ..." So, you are correct. "Reed relays" implied to me a particular implementation of a relay. The "relay" was the "nope" part. The original crosspoints used sealed reed contacts with plates of "Remendur" around them. Remendur is a cobalt/iron/vanadium magnetic material with a "square loop" characteristic. In that sense, they are a "reed relay" with memory. The "release" state has the plates magnetized in an opposing manner, such that the center of each plate is one pole, and the ends are another. The "operate" state has the plates magnetized in an aiding manner, such that the ends are opposite poles, and the reeds are in a magnetic path parallel to the plates. A cost/size-reduced version of the old "ferreed" crosspoint is the "remreed" crosspoint, with the only magnetic material needed being the reeds themselves. These sealed reeds are semipermanent magnets that attract or repel each other, based on the last current pulse that switched them. Sorry about the confusion -- while they are absolutely "reed" switches, they are not a "relay" in the usual manner. Why? The crosspoints have a very short lifetime if they are switched "wet" (with current flowing across the contacts), so the battery/ground supplies have to be interrupted with real relays before a network connection is set up. Thus they are a switch, but not a relay in the sense of connecting a large current in response to a smaller one. In general, old network connections remain until the next use of a crosspoint matrix, when the pulses that close a particular crosspoint also open any previously-closed crosspoint open all the other crosspoints in the same "row" and "column" of that matrix. Thus "releasing" a network connection just means removing the inputs and outputs (externally) and marking the input/output points as idle. Hunting an idle path consists of finding an idle input & output within a matrix. Al> Instruction set vaguely resembles an orthoganal version of 1E, with a Al> typical instruction (24 or 48 bits wide) taking .7 milliseconds. Jim> You mean .7 microseconds, right? Yes, of course. My "goofs/email message" ratio is getting bigger. And "orthoganal" is really "orthogonal" Al Varney AT&T att!ihlpf!var