Date: Thu, 18 Feb 2016 16:16:07 +0100 From: Borja Marcos <borjam@sarenet.es> To: Marius Strobl <marius@freebsd.org> Cc: =?utf-8?Q?Henrik_Lidstr=C3=B6m?= <freebsd@lidstrom.eu>, Hajimu UMEMOTO <ume@mahoroba.org>, Peter Jeremy <peter@rulingia.com>, freebsd-stable@freebsd.org Subject: Re: 10-STABLE hangups frequently Message-ID: <6FF1912A-3861-4496-A93F-680FA6479DE5@sarenet.es> In-Reply-To: <20160218002403.GA95789@alchemy.franken.de> References: <ygeegcvpmv1.wl-ume@mahoroba.org> <20160202200738.GA78969@server.rulingia.com> <yge8u32pptt.wl-ume@mahoroba.org> <56B761A4.7010901@restart.be> <56BD9FBA.3040002@lidstrom.eu> <20160218002403.GA95789@alchemy.franken.de>
next in thread | previous in thread | raw e-mail | index | archive | help
> On 18 Feb 2016, at 01:24, Marius Strobl <marius@freebsd.org> wrote: >=20 >=20 > Could those of you experiencing these hangs with ZFS please test > whether instead of reverting all of r292895, a kernel built with > just the merge of r291244 undone via the following patch gets > rid of that problem - especially on amd64 - and report back? > https://people.freebsd.org/~marius/r291244_reversal_10.diff Fist, a warning: as my report is very fuzzy, take it with a grain of = salt. That said, since two months ago more or less, I have been having = performance problems with a =E2=80=9Cplayground=E2=80=9D machine I use = for several purposes, like searching Netflow data, graphing performance = data from several machines with Orca, etc. =E2=80=94=E2=80=94=E2=80=94 CPU: Six-Core AMD Opteron(tm) Processor 2431 (2400.15-MHz K8-class CPU) Origin=3D"AuthenticAMD" Id=3D0x100f80 Family=3D0x10 Model=3D0x8 = Stepping=3D0 = Features=3D0x178bfbff<FPU,VME,DE,PSE,TSC,MSR,PAE,MCE,CX8,APIC,SEP,MTRR,PGE= ,MCA,CMOV,PAT,PSE36,CLFLUSH,MMX,FXSR,SSE,SSE2,HTT> Features2=3D0x802009<SSE3,MON,CX16,POPCNT> AMD = Features=3D0xee500800<SYSCALL,NX,MMX+,FFXSR,Page1GB,RDTSCP,LM,3DNow!+,3DNo= w!> AMD = Features2=3D0x37ff<LAHF,CMP,SVM,ExtAPIC,CR8,ABM,SSE4A,MAS,Prefetch,OSVW,IB= S,SKINIT,WDT> SVM: NP,NRIP,NAsids=3D64 TSC: P-state invariant real memory =3D 8589934592 (8192 MB) avail memory =3D 8270192640 (7887 MB) Event timer "LAPIC" quality 400 ACPI APIC Table: <SUN X4x40 > =E2=80=94=E2=80=94=E2=80=94 I was running lots of stuff (creating a lot of memory pressure) but the = system coped with it anyway. After one update (I use to track -STABLE) I = begun to have problems. Something I noticed was, the system begun = killing processes because of a lack of memory while the swap partition = wasn=E2=80=99t full at all. The wired memory would skyrocket as well. After reducing the clutter (ie, removing what I really didn=E2=80=99t = need) the server has been stable, but the performance (for example, when = performing large Netflow searches) has decreased remarkably. I have rebuilt -STABLE today with that patch, and something has = definitely changed. Performance is much better and memory consumption is = not growing like before, with the same NFsen, Orca and other stuff = running. I know this is not a serious report, I can=E2=80=99t do it with that = server, but I hope it rings a bell or somewhat confirms a data point. Borja.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?6FF1912A-3861-4496-A93F-680FA6479DE5>