Date: Sat, 01 Jul 2017 09:58:23 +0200 From: Harry Schmalzbauer <freebsd@omnilan.de> To: Alexander Motin <mav@freebsd.org>, scsi@freebsd.org Subject: Re: not a reproducable regression [Was :Re: bhyve ahcich0: Timeout on slot 0 port 0,, regression with stable/11->releng/11.1] Message-ID: <5957561F.7030906@omnilan.de> In-Reply-To: <5956A4E0.3030108@omnilan.de> References: <5956A3DF.8060109@omnilan.de> <5956A4E0.3030108@omnilan.de>
next in thread | previous in thread | raw e-mail | index | archive | help
Bezüglich Harry Schmalzbauer's Nachricht vom 30.06.2017 21:22 (localtime): > Bezüglich Harry Schmalzbauer's Nachricht vom 30.06.2017 21:17 (localtime): >> Hello, >> >> on releng/11.1 I noticed a severe performace degradation during file >> unlinking in a FreeBSD guest. Host was running quiet recent stable/11 >> before. >> >> On the host, the vm is started with >> ahci,hd:/dev/adaN >> >> The guest attaches: >> ada0 at ahcich0 bus 0 scbus0 target 0 lun 0 >> ada0: <BHYVE SATA DISK 001> ACS-2 ATA SATA 3.x device >> >> The geust has very high Sys-load during unlinking (50-75%@2 cores). >> Also, the host logs these errors: > Sorry, not the host logs these errors, but the guest, obviously. > After a reboot, I couldn't reproduce the timeouts, nor the completion time multiplication. It takes ~8 minutes like it alwas was… Sorry for late falsifying. I'm still wondering what these timeout error codes translate to. And I can still see very high CPU load in the guest during unlinking (obj-tree of usr/src). While systat reports less than 10irqs/s for AHCI0, "intr" consumes 30-800% at times (with 4 vCPUs). "bufdaemon" and "rm" consume 10-100% each. So total average is about 2 completely loaded cores for 'rm' running in a FreeBSD guest. Any hints regarding timeout codes or how to trace where CPU cycles drain are highly appreciated! Thanks, -harry
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?5957561F.7030906>