From guest Wed Apr 27 08:57:09 1994 From: "Michael Jones" Message-Id: <9404270857.ZM19783@babar.asd.sgi.com> Date: Wed, 27 Apr 1994 08:57:01 -0700 In-Reply-To: "Timothy Cormier {76857}" "" (Apr 25, 5:46pm) References: <199404252146.AA05178@swlgbi.msd.ray.com> X-Mailer: Z-Mail (3.1.0 22feb94 MediaMail) To: "Timothy Cormier {76857}" , info-performer@sgi.sgi.com Subject: Remote .vs Local IRIS Performer graphics Content-Type: text/plain; charset=us-ascii Mime-Version: 1.0 Status: OR On Apr 25, 5:46pm, Timothy Cormier {76857} wrote: > Subject: :I am trying to understand why Onyx based Performer applications :cannot pipe their display to an Indigo workstation. I have set :the DISPLAY enviornment variable properly on the Onyx, and have :set xhost on the Indigo to allow the pipe. Because the graphics command stream generated by IRIS Performer is sent directly to the graphics pipeline hardware attached to the local workstation. IRIS Performer depends on such machine-tied local events as vertical retrace interrupt signals and thus can not work across a network. :I have I heard about GL, OGL and now DGL! Can someone give me an :explaination as to why I cannot display properly on another SGI :machine. You *can* display properly: X-class graphics across a network, and high performance IRIS Performer applications on a local workstation. There is just no way to send enough textured polygon definitions per second across the network. x y z = 3 floats = 12 bytes r g b a = 4 floats = 16 bytes nx ny nz = 3 floats = 12 bytes s t = 2 floats = 8 bytes TOTAL = 48 bytes/vertex 9000 tris/frame @ 30Hz => 270000 triangles/sec: 270000 x 3 x 48 = 38.88 mbytes/sec (bus speeds) 231 tris/frame @ 30Hz => 6944 triangles/sec: 6944 x 3 x 48 = 0.999936 mbytes/sec (ethernet speeds) Do you want to drive a 231-polygon display from your Onyx? -- Be seeing you, Phone:415.390.1455 Fax:415.390.2658 M/S:8L-590 Michael T. Jones Silicon Graphics, Advanced Graphics Division mtj@sgi.com 2011 N. Shoreline Blvd., Mtn. View, CA 94039-7311 From guest Fri Apr 29 07:51:44 1994 From: "Michael Jones" Message-Id: <9404290751.ZM27446@babar.asd.sgi.com> Date: Fri, 29 Apr 1994 07:51:33 -0700 In-Reply-To: hitchner@netcom.com (Lew Hitchner) "Re: Remote .vs Local IRIS Performer graphics" (Apr 29, 7:45am) References: <199404291445.HAA16962@netcom.com> X-Mailer: Z-Mail (3.1.0 22feb94 MediaMail) To: hitchner@netcom.com (Lew Hitchner), info-performer@sgi.sgi.com Subject: Re: Remote .vs Local IRIS Performer graphics Content-Type: text/plain; charset=us-ascii Mime-Version: 1.0 Status: OR Lew sent the following note and then asked me to send it out to the info-performer mailing list. How common is this type of situation? Comments? On Apr 29, 7:45am, Lew Hitchner wrote: > Subject: Re: Remote .vs Local IRIS Performer graphics :Mike Jones replied: : :> . :> . :> . :> 9000 tris/frame @ 30Hz => 270000 triangles/sec: :> 270000 x 3 x 48 = 38.88 mbytes/sec (bus speeds) : :> 231 tris/frame @ 30Hz => 6944 triangles/sec: :> 6944 x 3 x 48 = 0.999936 mbytes/sec (ethernet speeds) : :> Do you want to drive a 231-polygon display from your Onyx? : :In my experience, and I would suspect also that of many Performer :users, the example of driving a "231-polygon display" is probably :common (or, alternatively a 2310-polygon display at 3 Hz). This would :not be for production type results, but for development testing. I :found that very frequently I wanted to try out something new just to :see if "it works at all" and didn't care about frame rates or whether I :tested it with a large database. Often the "does it work at all" test :can be conducted with a toy database at low framerates. It's hard to :estimate, but I would guess that perhaps as many as 50% of my test runs :are in the "does it work" category and the rest require testing with :large database at high frame rates. In environments where there's one :"big" machine (e.g., 4 or 8 CPU Onyx) and many networked "small" :machines (e.g., Indigos), developers usually develop and test on the :small machines (single CPU) on their desktop. But, you also need to :check out your latest development on the "big machine" with its OS, :multi-CPU, and graphics hardware environment. But, in large :organizations the console of "big machine" may not always be available :on demand for every developer whenever he or she wants to test their :latest "enhancement". This is the case at NASA Ames where the main :SkyWriter often must be scheduled in advance for using it at the :console in its lab but it's nearly always accessible over the network :for quick shots at trivial tests (and then, of course, analysis of the :core dump your test generated). Also, if you are running dbx or other :CASE analysis tools remotely on the multi-CPU machine, network transfer :rate is of little concern. : : Lew Hitchner : VR consultant : (formerly NASA Ames Res. Center) >-- End of excerpt from Lew Hitchner -- Be seeing you, Phone:415.390.1455 Fax:415.390.2658 M/S:8L-590 Michael T. Jones Silicon Graphics, Advanced Graphics Division mtj@sgi.com 2011 N. Shoreline Blvd., Mtn. View, CA 94039-7311 From guest Fri Apr 29 17:12:12 1994 From: "Michael Jones" Message-Id: <9404291712.ZM28725@babar.asd.sgi.com> Date: Fri, 29 Apr 1994 17:12:04 -0700 In-Reply-To: lee@igate1.hac.com (Craig A. Lee) "Re: Remote .vs Local IRIS Performer graphics" (Apr 29, 5:00pm) References: <9404300000.AA25210@igate1.hac.com> X-Mailer: Z-Mail (3.1.0 22feb94 MediaMail) To: info-performer@sgi.sgi.com Subject: Re: Remote .vs Local IRIS Performer graphics Content-Type: text/plain; charset=us-ascii Mime-Version: 1.0 Status: OR Several people have responded to my email from Lew Hitchner. Here's one of the typical responses: > Subject: Re: Remote .vs Local IRIS Performer graphics :>Lew sent the following note and then asked me to send it out :>to the info-performer mailing list. How common is this type :>of situation? Comments? : :I think the situation described (one or a few high end graphics systems, :many low end systems, lots of development on low-end just to test out :"does it work", high-end used for full features and performance) might be :very common. Within my company, I've worked for two entirely unrelated :facilities, and both have had exactly that situation. Two other unrelated :sites also sound to me as if they operate like that. I think I should have been clearer about the question. Very many IRIS Performer sites run the same code on Indigo and Onyx machines and all types in-between. The question is if people want to run their program on one machine and have the output appear on the other. We do not support this and wonder what the reasons would be for it. Since performance considerations *dictate* that graphics appear on a local machine, what practical or application demands would be inspiration for such a requirement? -- Be seeing you, Phone:415.390.1455 Fax:415.390.2658 M/S:8L-590 Michael T. Jones Silicon Graphics, Advanced Graphics Division mtj@sgi.com 2011 N. Shoreline Blvd., Mtn. View, CA 94039-7311 From guest Sun May 1 18:21:31 1994 From: Robert Webb Message-Id: <199405020120.AA16256@godzilla.cgl.citri.edu.au> Subject: Re: Remote .vs Local IRIS Performer graphics To: mtj@babar (Michael Jones) Date: Mon, 2 May 1994 11:20:32 +1100 (EST) Cc: info-performer@sgi.sgi.com In-Reply-To: <9404291712.ZM28725@babar.asd.sgi.com> from "Michael Jones" at Apr 29, 94 05:12:04 pm X-Mailer: ELM [version 2.4 PL21] Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Content-Length: 1631 Status: OR Michael T. Jones (mtj@sgi.com) writes: > The question is if people want to run their program > on one machine and have the output appear on the other. We do not > support this and wonder what the reasons would be for it. > > Since performance considerations *dictate* that graphics appear on a > local machine, what practical or application demands would be inspiration > for such a requirement? We do not have debuggers on the many Indigos, only on the main machine. So debugging a Performer application would only be possible from that machine's console, which is already in high demand. It is far more convenient to stay seated at my desk to debug my code. Testing whether the code works at all on a multi-processor machine, performance considerations aside, would also be possible without having to get time on the main machine's console. We have Performer 1.0, and it works across the network as far as I can tell, just following the DISPLAY variable. Has this functionality been removed for Performer 1.2? Why would functionality be removed when clearly it is still desired by the users (going by the stuff that's been sent to this mailing list, and my own personal needs)? Performer just uses gl, which runs over the network, so it seems like there has been a conscious decision to stop it from doing the same. If this functionality was of no use, why was gl made to run over the network a few years ago? -- +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Rob. (webb@godzilla.cgl.citri.edu.au) +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+