Forums › BB Series Discussions › Performance regression for APIs newer than 3.0.11
- This topic has 2 replies, 2 voices, and was last updated 6 years, 5 months ago by Andrew.
catalin_roParticipantcatalin_ro October 7, 2016 at 12:09 am
With the APIs that came after v3.0.11 there is a serious performance hit in zero-span mode, especially when using two BB60C simultaneously (Spike used for both of them). The test PCs are roughly equivalent – i7-2600/16GB/Radeon HD6850 and i7-4710HQ/16GB/nVidia.
Spike 3.0.21 (with API 3.0.11) shows about 10% CPU usage per receiver (so 20% total CPU usage when using both of them) when not decimating (27MHz BW/40MHz sampling rate).
Starting with Spike 3.0.22 (API 3.0.12 and newer) shows about 15% CPU usage for the same settings, but the total CPU usage jumps to 60% when using both receivers. Those 60% are equally split between the two Spike instances.
The test PCs are running Windows 10 Home, 64-bit, fully updated, including the newest drivers.
AndrewModeratorAndrew October 7, 2016 at 1:11 pm
Check out your other thread for my response to the 2 streaming receiver question.
As for performance regressions. I am re-visiting this issue for a short time. As we have discussed in earlier emails, the bbGetIQ function adds a small amount of additional overhead. At this point I have only seen a marginal regression on one Win10 laptop in our office. All other PCs I have tested exhibit a non-observable CPU increase by switching to the bbGetIQ functionality. In the latest 3.0.14 API, the bbFetchRaw function is simply routed to the bbGetIQ function, and will incur the small overhead.
I will use the Win10 laptop as a test bed for some potential optimizations.
In the mean-time, as we previously discussed, you can continue to use the 3.0.11 API version and the bbFetchRaw functionality.
AndrewModeratorAndrew October 7, 2016 at 4:47 pm
Wanted to give you an update.
I may have found the cause of the performance regression. I will need to investigate further on Monday, but it appears to not even be related to the addition of the bbGetIQ function, but a separate unrelated change to the IQ processing. I will confirm my findings on Monday and hopefully be able to get you an update to try soon. Fingers crossed.
You must be logged in to reply to this topic.