Thanks for the insight Bob. Its really a pity that USB performance is slow because the rest of the product is so impressive. Are any improvements are in the pipeline as far as the USB subsystem is concerned that would improve performance at all, and can any tweaking be done in meantime?
Regards, David. Bob Doolittle wrote: > The reason for slow USB performance on Sun Ray is that, in order to > remove the need for multiple custom firmware drivers, the USB protocol > is expressed directly over the network between client and server > (except for keyboard and mouse which are handled in firmware). It's > the server that processes the "raw" USB stream. USB protocol is > defined for an extremely low-latency bus, and as such has very small > maximum transfer sizes, and the protocol is synchronous. When > translated to a higher-latency bus like ethernet, the overall > throughput is greatly reduced by the inter-packet latencies. > > -Bob > > David Markey wrote: >> Ok it seems that using samba doesn't make a difference when copying off >> a USB key mounted on the server. It took 26 minutes via samba and about >> the same via RDP redirection to copy a 330MB file from my USB key. >> >> The biggest bottleneck lies somewhere between the DTU and the SunRay >> server, or I heard that the pcfs implementation on Solaris has issues >> aswell. >> >> >> It's really quite strange that the USB throughput over a 100mbit pipe is >> as low as ~12MB/min, I did a quick test and i was getting 11.21MB/s over >> FTP to the sunray server(from the same location that my DTU is) so i'm >> ruling out a slow network. >> >> Does everyone else routinely experience USB performance similar to this? >> >> Thanks. >> >> David >> >> >> >> >> >> >> >> >> >> David Markey wrote: >> >>> Not really, But i did do the samba test 1st so in theory RDP had the >>> advantage by having the file buffered for the 2nd test? >>> >>> >>> >>> John Francis wrote: >>> >>>> 2009/3/28 David Markey <[email protected]>: >>>> >>>>> It took 7 seconds via samba to copy the file and 29 seconds via RDP. >>>>> >>>> Did you take into account file system buffers? For example, before >>>> running either test did you do a "cat myfile > /dev/null" to make sure >>>> it was likely pre-loaded into RAM? >>>> >>>> >>> _______________________________________________ >>> SunRay-Users mailing list >>> [email protected] >>> http://www.filibeto.org/mailman/listinfo/sunray-users >>> >> >> _______________________________________________ >> SunRay-Users mailing list >> [email protected] >> http://www.filibeto.org/mailman/listinfo/sunray-users >> > _______________________________________________ SunRay-Users mailing list [email protected] http://www.filibeto.org/mailman/listinfo/sunray-users
