Hi, The thread priority fixes seem to work[1]. However there are reports that downloads of uncompressed files get corrupted when restarting Freenet:
> It also still corrupts downloads of files that were inserted without > compression on stopping and restarting Freenet. Also, it leaves behind the > "orphaned" temp file in my downloads directory when I cancel and restart > those corrupted downloads. — SSK@BKc5LtaIdeM6QMIoutwtG3tDa8iZoY8C5QN7u9dwHec,RK7kB6rAwbqQJ9XztTdNPNdd8yk1kNo79NzqEZNNIwU,AQACAAE/fms|2018-11-17|Message-1 We still have some way to go to get the new builds with Gradle and JNA back into the quality of the version before, but we’re getting there. If you want a Hackathon topic, it would be cool if you could investigate the corruption of downloads. Besides: We now have an estimate of Java 7 usage among people who use pure auto-update: I am inserting the Java 7 update for 1482 today, but for Java 8 I inserted two days ago. Therefore all those still on 1481 as of *now*[2] minus all those still on 1481 tomorrow were still using Java 7. Currently (at 3400 minutes into the release) 2 out of ~30 use 1481. If that goes to zero tomorrow, we know that we have around 7% users on Java 7 — roughly 400 users.[3] Another interesting point: around one in three of our users manually downgraded to 1480 again, so we have a high proportion of users who are tech savvy enough to manage their nodes (or who have good backups). Which I think is a good thing. Where to go from here --------------------- I want to release 1483 with the thread priority fix tomorrow. It would be cool to be able to add some improvements from the hackathon, too. If 1483 resolves the problems for Windows users, we can look into increasing the peer count again. Ideal to see the effect would be if we could collect several stats for "Success rates by HTL (remote bulk fetches)" and "Success rates by HTL (remote realtime fetches)" from the current (low) peer count and an increased peer-count. That will give us an estimate of the average hops a request needs to reach its target. To provide serious anonymity, we should have *at least* three hops on average. [2]: spreading of 1482: USK@CQFyzDofVhBmeHN5VcCMWaULAttLNY3bUBMhcJWiGks,X-frg11rtR037GuwC3ndhwHZm5WByWKeO9i2LMF13DQ,AQACAAE/watch-1482/186/ [3]: according to the probe statistics we have around 5000-6000 active nodes right now: USK@WMa1Z40iYdZZ51yctQ3toFl9zuuFEnNdsm3NejJU5KE,jCBcaNBeKD5~sSQeSkyKz737Bh5ibBGqdzfD8mgfdMY,AQACAAE/statistics/371/ Best wishes, Arne [1]: the measurements are from FMS: > Test with 1480, 1481 (original and nextgens commit without the reduced > peer-count patches), 1482 (original and with nextgens commit). > Time started with the first connected peer, then first measurement after 15 > minutes, because freenet takes some time (5-8 min) to stabilize. Second > measurement after 30 minutes. > CPU values are now averaged over (the last) 15 minutes (meaning the spikes > are included, so the values are a bit higher than the above). > > Results: > RAM usage is consistently a lot higher with 1481/1482. > Nextgens patch seems to reduce RAM usage a lot (still higher than 1480). > Network speeds are a lot lower with 1482. > Nextgens patch applied to either 1481 or 1482 reduces CPU usage significantly. > => If you release 1482, make sure you have nextgens patch applied. > > > > 1480: > - 15min: 128 MB; 16/19 peers; 59.4/56.6 KiB D/U; > - 30min: 107 MB; 15/19 peers; 52.5/48.9 KiB D/U; 2.90% CPU > > > > 1481: > - 15min: 557 MB; 16/19 peers; 45.5/40.5 KiB D/U; > - 30min: 459 MB; 18/19 peers; 50.4/45.9 KiB D/U; 4.26% CPU > > 1481 (nextgens): > - 15min: 275 MB; 15/19 peers; 47.1/44.7 KiB D/U; > - 30min: 169 MB; 15/19 peers; 49.6/48.9 KiB D/U; 3.01% CPU > > > > 1482: > - 15min: 471 MB; 10/12 peers; 19.0/17.8 KiB D/U; > - 30min: 509 MB; 10/12 peers; 19.0/18.1 KiB D/U; 2.68% CPU > > 1482 (nextgens): > - 15min: 200 MB; 9/12 peers; 19.2/18.5 KiB D/U; > - 30min: 145 MB; 11/12 peers; 19.4/18.6 KiB D/U; 2.15% CPU — SSK@vRVILhDSHOnxWBnmeLXCCtTqaXkR6Td8XoI8K4nabDs,NZesW44VfZBk4Blu7bSu8HggVNyHxDlJLyicWUMBV2Q,AQACAAE/fms|2018-11-16|Message-0 > I haven't tried the nextgens patch as yet, but I can say 1482 is marginally > better than 1481 CPU usage-wise. > > 1481 consistently used 20-40% CPU for me, with occasional spikes higher. > 1482 sits at 2-5% much of the time, but frequently spikes into the 30-40% > range for 30-60 seconds at a time. 1480 rarely went over 1.5% CPU usage. > > 1482 also still has the bug of corrupting downloads of files that were > inserted without compression when you stop and restart Freenet. — SSK@BKc5LtaIdeM6QMIoutwtG3tDa8iZoY8C5QN7u9dwHec,RK7kB6rAwbqQJ9XztTdNPNdd8yk1kNo79NzqEZNNIwU,AQACAAE/fms|2018-11-17|Message-0 -- Unpolitisch sein heißt politisch sein ohne es zu merken
signature.asc
Description: PGP signature
