Will@cup.portal.com (Will E Estes) (01/24/91)
I need to get an idea for the maximum number of users that I can get to connect to an IBM mainframe using TCP/IP. I'm interested in factors that limit performance (e.g., I'm assuming that probably 50% of the host CPUs time is going to be spent just running the TCP connection for each user); I am particularly interested in any bottlenecks that would physically prevent more than a given number of users from connecting. I'm assuming that a single 3172 probably supports a finite number of concurrent TCP connections to the IBM host, but could you use multiple 3172s, or multiple gateways from some other manufacturer, that would allow you get many hundreds of concurrent connections? Today I might only need a few hundred connections. Over time this number might approach 1000 concurrently connected users, so I need a scalable architecture. Any thoughts on this are appreciated. Thanks, Will Estes (apple!cup.portal.com!Will)
MAB@CORNELLC.CIT.CORNELL.EDU (Mark Bodenstein) (01/26/91)
On 23 Jan 91 23:45:32 GMT you said: >I need to get an idea for the maximum number of users that I can >get to connect to an IBM mainframe using TCP/IP. ... We currently have about 200 users connecting to one of our IBM mainframes using version 1 of the IBM VM TCP/IP software and one 8232 front end processor. The CPU used is perhaps 5% of a 3090J processor. This same 8232 supports perhaps 350 megabytes per day (approx 50,000 files) sent via the VMNET (RSCS over TCP/IP) protocol, plus various other services (e.g. SMTP, FTP, and a campus information service.) > ... I am particularly interested in >any bottlenecks that would physically prevent more than a given number >of users from connecting. Our current configuration will support up to around 300 telnet connections, based on the virtual memory available to the server, the amount it uses per connection, and the other uses here of TCP. The next version of the IBM software, currently available but not yet in production here, increases this limit to around 2000 with a 370 mode service virtual machine, and I believe to around 4096 (the current operating system limit for this kind of login) with an XA mode service virtual machine (this latter capability has been announced on the network but not yet distributed, as far as I know.) > I'm assuming that a single 3172 probably >supports a finite number of concurrent TCP connections to the IBM >host, but could you use multiple 3172s, or multiple gateways from >some other manufacturer, that would allow you get many hundreds of >concurrent connections? A 3172 is about an order of magnitude better performer than the 8232 which I mentioned that we use. It should not be the limitation on number of users connected within the ranges we're discussing, for ordinary terminal sessions. > Today I might only need a few hundred connections. >Over time this number might approach 1000 concurrently connected users, >so I need a scalable architecture. Any thoughts on this are >appreciated. You should be able to do this with the current technology. Mark Bodenstein (mab@cornellc.cit.cornell.edu; 607-255-8059) Cornell University