webber@aramis.rutgers.edu (Bob Webber) (08/19/87)
In article <2533@ames.arpa>, eugene@pioneer.arpa (Eugene Miya N.) writes: > Well, I'm glad that all of you Good Bodies find Disk Striping > interesting. Now, if one or two of you could just go off and build > something, we would be very interested in buying somethin. I think > we could easily get you a few dozen customers. Would be nice if it > interfaces with Crays, transfers in excess of 1 GB/sec, etc. etc...... > Oh yeah, should not cost more than a Cray itself and hold, say, > 1 Terabyte of data (100 GWords? yeah, okay for now 8-). I'm I > forgetting anything? Is Disk Striping a plausible way to approach these figures? At Siggraph'87, I was looking at a Texas Memory Systems box that contained 2 gigabytes of mass memory that it could deliver in bursts of 120 megabytes per second (100 megabytes per second average) for a paltry 1.4 megadollars. Looking at the specs for a Cray 2 system, I see that it can be configured with up to 36 Cray DD-49 disk drives (1.2 gigabytes a piece) however, apparently the fastest channel on a Cray is a 100 megabyte per second HSX channel of which you can have at most 8. The cycle time on a Cray 2 is 4.1 nanoseconds and can have up to 4 processors. So, this leaves me with the following questions: 1) how much does a fully configured Cray-2/4-256 cost? (will 75 megadollars for 50 mass memories throw me over budget?) 2) how fast do the DD-49's pump data? how much do they cost? 3) is the Cray 2 fast enough to do anything interesting with all of this data we are pumping into it? [a tetrabyte is only a 5000 by 5000 by 5000 array of 64bit words. at a gigabyte per second, it is going to take 40 minutes to cycle through it (20 minutes reading it and 20 minutes writing it back out). a) if I am running local interactions, then it is going to take days for information to propagate from one side of the array to the other just from i/o time. b) if I am running global interactions, then I need to ensure that at some time during the computation, each pair of bytes is simultaneously in the 256 megaword (2 gigabyte) common main memory. near as I can figure, the best I can do is break the tetrabyte into 1000 gigabyte chunks and bring them pairwise into memory. this is also going to take days of i/o time. so tentatively, I would say that there would be little point to making the tetrabyte Cray 2 compatible. looks more like one would want something tetrabyte compatible instead.] 5) if instead of running a large-scale application, one just wanted to run a virtual memory system on a 2 gigabyte main-memory machine, is all of this of a scale to support gigabyte users in the style that megabyte users on a SunIII have become accustomed to? ------ BOB (webber@aramis.rutgers.edu ; rutgers!aramis.rutgers.edu!webber)
cm26+@andrew.cmu.edu (Curt McDowell) (08/25/87)
> making the tetrabyte Cray 2 compatible. looks more like one would > want something tetrabyte compatible instead.] @typewriter{The disk drive on my TRS-80 Model 1 stores 46080 tetrabytes. That equals about 0.0000001676380634307861328125 terabytes. }What a difference that 't' makes!! Curt McDowell