ric@ace.sri.com (Richard Steinberger) (04/13/91)
I have a few networking questions that I could use some help answering. I'm still reading Stevens' Unix Coimmunications book and am not a guru. Thanks in advance to all who reply. Configuration: We have a uVAX with TCPIP SW (from DEC, UCX I think) and an attached Aptec interface. The uVAX needs to xfer files to another (unix) computer, and at the moment this computer is a Multiflow (basically a Unix mini-super based on VLIW architecture). In the future, the Multiflow might be replaced by a SUN, DEC or other machine acting as a fileserver. The usual procedure is that the unix computer retrieves files via FTP or by using the APTEC bus. FTP (TCP over enet) transfer rates have been as high as about 100K bytes/second. Aptec rates have been less that that, but the company that developed the Aptec interface promises that they now can transfer files at 2 - 3 times FTP rates. (Aptec sells what they call IO computers. Their forte' is the rapid transfer of data from computer to computer, generally VAX to VAX, but they have an interface for SUN and some [incomplete, if you ask me. Allegedly Mitre has a turnkey file xfer package] interface SW). Question 1: Is FTP considered the fastest reliable way to transfer medium to large files (500KB - 10MB) between a VAX/VMS machine and a BSD-4.3 unix machine (using ethernet as the physical medium)? Would adding another enet interface card to each computer and connecting a second enet cable make any sense? I.e., could a second cable allow multiple (2) file transfers simultaneously, and could a vax and unix machine use both interfaces at once? Would this be transparent to the end user? Are there other (enet-based HW and/or SW) possibilities? A colleague at another company mentioned that something he called a "DECnet-Internet Gateway" would allow file xfer at 2 - 3 times FTP rates. Perhaps he means DECnet-Ultrix network SW. Is this mush? Or can DECnet xfer rates dramatically exceed FTP over enet (or is it the reverse, or is it more complicated than that)? Question 2: I would like a repeatable way to put a known traffic load on an ethernet-based LAN? Is there something more scientific than continuously transfering a large file from one host to another? Are there any (bundled) utilities in VMS or MultiNet that can display some quantifiable measure of network traffic (updating at a selectable interval)? Or, conversely, would it be better to attach a network analyzer (or buy monitoring SW)? Question 3: This is about fileserver performance. Does anyone know of any PD fileserver benchmark tests. I would like to propose a simple (but hopefully not simple-minded) test strategy for evaluating the performance of a few candidate fileservers. We are primarily concerned with a server's ability to provide (via NFS) files to 2 or 3 clients with "minimal" delays. This same server would simultaneously be receiving files via FTP (or other enet SW) and perhaps doing some computation (to reformat the files). Is there something more elaborate than doing one (or several) rcp(s) to/from server to client and measuring the time? What is likely to be the performance "bottleneck"? The ethernet itself, or the fileserver's ability to retrieve and store data from/to disks? Again, thanks to anyone who can provide help, guidance or suggestions on these questions. I do appreciate it. Suggestions for books to read also welcome. regards, ric steinberger ric@rml2.sri.com ric@ace.sri.com 415-859-4300