On 11/14/2012 2:59 PM, Eric Keller wrote: > On Wed, Nov 14, 2012 at 12:57 PM, Kent A. Reed <[email protected]>wrote: > >> As for the silly glxgears test, I think the original intent was to use >> it as a quick exercise of the X/OpenGL/driver path and perhaps as a >> benchmark of the graphics performance if the path works (although >> graphics guys are quick to tell us it isn't a benchmark). >> >> For latency testing purposes, it might help smoke out problems with the >> graphics subsystem---some drivers and some chips grab hold of DMA for >> example. >> > glxgears can slow down the computer, but I've never really seen it blow up > the latency numbers much....
I have seen it cause a blow up but not recently. When it did, typically web browsing did too because the browser also exercises the graphics subsystem hard. Trouble is, good or bad behavior depends on the specific circumstances, so we're forced to run everything when we exercise a new system. That's really all the latency-test instructions say: try everything you can think of to see if you can get to the worst-case numbers and here's a list of things we've had trouble with in the past---glxgears, videos, web surfing (which implies network access), directory listings, etc. > ... The worst latency I got on the last computer I > stress tested was starting a Java program off of the cd drive. > Historically, the worst latency problems on the PC architecture were from > disk accesses. All the other problems are really driver related as far as > I can tell I've had more problems with various USB devices (and not only USB mass-storage devices) than with internal disks. Again, the details of board/bios, cpu, drivers, specific device, whether or not legacy USB support is enabled, plugging the device in before or after booting, etc., mattered. Unfortunately, saying the problems are driver-related only identifies the source of the problem, it doesn't resolve the problem. There's still vendor-proprietary magic which we can't change in drivers for many graphics subsystems; even with open-source hardware drivers we face the problem that changes get made in them whenever a developer gets a wild hair. Sometimes the changes are pure improvement; more often the changes are a mixed blessing. And, yes, I do always expect my bread to fall jelly-side down on the floor:-) Regards, Kent ------------------------------------------------------------------------------ Monitor your physical, virtual and cloud infrastructure from a single web console. Get in-depth insight into apps, servers, databases, vmware, SAP, cloud infrastructure, etc. Download 30-day Free Trial. Pricing starts from $795 for 25 servers or applications! http://p.sf.net/sfu/zoho_dev2dev_nov _______________________________________________ Emc-developers mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/emc-developers
