Bug 1158029
| Summary: | [RFE] Spice: Improve performance over high latency WAN links | ||
|---|---|---|---|
| Product: | Red Hat Enterprise Linux 7 | Reporter: | Evgheni Dereveanchin <ederevea> |
| Component: | spice | Assignee: | Frediano Ziglio <fziglio> |
| Status: | CLOSED WONTFIX | QA Contact: | SPICE QE bug list <spice-qe-bugs> |
| Severity: | medium | Docs Contact: | |
| Priority: | medium | ||
| Version: | 7.4 | CC: | amarchuk, bdoran, cfergeau, dblechte, ederevea, fziglio, inetkach, jbuchta, jpullen, lpeer, lsurette, marcandre.lureau, meverett, mtessun, obockows, pzhukov, rbalakri, srevivo, tpelka |
| Target Milestone: | rc | Keywords: | FutureFeature |
| Target Release: | 7.4 | ||
| Hardware: | Unspecified | ||
| OS: | Unspecified | ||
| Whiteboard: | |||
| Fixed In Version: | Doc Type: | Enhancement | |
| Doc Text: | Story Points: | --- | |
| Clone Of: | Environment: | ||
| Last Closed: | 2017-05-24 10:59:57 UTC | Type: | Bug |
| Regression: | --- | Mount Type: | --- |
| Documentation: | --- | CRM: | |
| Verified Versions: | Category: | --- | |
| oVirt Team: | Spice | RHEL 7.3 requirements from Atomic Host: | |
| Cloudforms Team: | --- | Target Upstream Version: | |
| Embargoed: | |||
|
Description
Evgheni Dereveanchin
2014-10-28 11:17:11 UTC
I want to add that the shaper (reproducer step 2) can be applied on client side as well, so no changes to RHEV side are required to test this. Got some investigations. I instrumented spice-server adding message counter. I tested with a connection having 100ms latency and 20Mbit. As expected the number of commands per second is much higher than Windows (70 versus 200 of RHEL7). I applied the network improvements patches I sent 18 months ago and got clear improvements, the commands goes up to 500 (a bit less) doing continued scrolling and responsiveness is better. The reason of so much commands compared to Windows is that on RHEL7 Mesa implementation split images in chunks of 64K pixels (more or less). This increase the number of needed commands. Also this make video detection and streaming in general much worse basically causing spice-server code to create multiple streams. Beside the network optimizations possible I think would be worth collapsing (so doing the reverse of what Mesa is doing) so to have bigger images and less commands. Got a couple of promising patches specifically for RHEL7. Trying to get also some better statistics. worth pointing out the upstream series to improve mesa (also reducing memcpy and traffic in guest): drisw/glx: use XShm if possible Current status: - mesa patch does not fix the issue (https://bugzilla.redhat.com/show_bug.cgi?id=1030024#c19); - my workaround patch which is working and tested are rejected for 7.4; - for 7.5 we are going to support Virgl instead of Qxl so the problem will disappear. |