Date: Sun, 24 Mar 2013 08:43:48 -0700 (PDT) From: Paul Pathiakis <pathiaki2@yahoo.com> To: Bill Totman <bill.totman@gmail.com>, "davide.damico@contactlab.com" <davide.damico@contactlab.com> Cc: "freebsd-performance@freebsd.org" <freebsd-performance@freebsd.org> Subject: Re: FreeBSD 9.1 vs CentOS 6.3 Message-ID: <1364139828.17881.YahooMailNeo@web141405.mail.bf1.yahoo.com> In-Reply-To: <514F099D.9040005@gmail.com> References: <CAF3xD3nDDUYB94TS9AUUQ=CPztB1S8mU4fRwNB5GupmK8MgwXg@mail.gmail.com> <ebfc033f-d8db-4162-86be-db013e3d7eb2@email.android.com> <1363998883.22604.YahooMailNeo@web141401.mail.bf1.yahoo.com> <5d10e2a0b0f6477293459a26df1fc272@sys.tomatointeractive.it> <514F099D.9040005@gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
I haven't worked with CentOS since 6.0.=A0 I work on many other variants at= this time.=A0 I'm more than happy to take a look after I get my company of= f the ground....=A0 (A couple more months or land my next contracting gig)= =0A=0A=0AAnyhow,=A0 unlike with jails, it seems no matter what type of VM I= use, there's always 'overhead' in using virtual machine software.=A0 Perfo= rmance is 99.99% there with a 'real' installation.=A0 Whether it's VMware, = VirtualBox, or Zen, there's the issues of things that have always seemed to= cause a minor 'hurt'.=A0 It's also annoying when I have to dictate how muc= h memory, how many cores, disk space and everything else with regards to ea= ch virtual machine.=A0 Running the virtual software takes resources, from t= he OS, and each VM under control has to be 'given' bounds as to how many re= sources it can use. (I'm told that an ESX server is better at this, however= , that server must have a core OS that uses resources as well.)=0A=0A=0ASo,= lately, I started working with jails....=A0 for everything.=A0 There seems= to be no measurable issues with their use.=A0 Does anyone have any compari= son on jails versus various VM software?=A0 I'm not just talking the VM sof= tware running an OS in real-time and no negligible loss of performance.=A0 = I'm talking about the what's being taken from the core machine running the = software.=A0 That's overhead.=A0 The software consumes resources (memory, c= pu cycles, etc) and creates a certain amount of overhead for each VM create= d.=0A=0A=0AJails seem to be highly maintainable, easy to use, the resource = management of CPU, memory and other types are handled by the OS and not an = additional layer of software running on the host that becomes responsible f= or all this juggling.=A0 So, from my perspective, it seems jails remove a l= ayer of indirection over VM software.=A0 (Of course, arguably, jails are li= ghtweight VMs.)=A0 I'm just starting to become knowledgeable and a 'fan' of= jails.=0A=0AI'm also a little 'aged' and I never understood the need for V= M software as UNIX has always been capable of juggling (time slicing) task = courtesy of the job scheduler and the like.=A0 I can see that a mainframe, = mini, and Windows OS that were not designed to be capable of time sharing w= ould need them, but not UNIX.=0A=0AIf I wanted a 'rough' analogy, I can equ= ate VM software is to jails as UFS is to a ZFS pool.=A0 I think of it this = way:=A0 with VM software I have to understand the resources I will need and= I will create boundaries according to a 'best guess' scenario with jails I= create the environment and all the jails get access to all the available r= esources to the machine and allow a robust UNIX-like (I really hate writing= that given FBSD's roots :-) ) to handle something it's always been capable= of handling from it's design.=A0 This is akin to having to setup UFS versu= s ZFS.=A0 UFS you have to have an idea of how big the partitions are and ch= oose bounds (and it's "not fun" when you have to re-partition), however, wi= th ZFS every partition grows within the bounds of the pool until it is exha= usted, at that point, add storage to the pool.=A0 (With a jail, at that poi= nt, if it's anything but CPU cores, just add resources to the machine - if = it is CPU, it's time for a new CPU or maybe a second machine.)=A0 =0A=0A=0AI hate to say this but I'm fin= ding jails 'highly superior' to VM software and now that I hear that we can= run Linux in a jail, I'd be very curious to do that, too.=A0 =0A=0A=0A<ran= t>=0A=0AOne last thing, I see VMs almost as a 'development tool' that peopl= e just recklessly took to the next level.=A0 It's a lot of fun to create VM= s on a desktop machine that you are doing development on to see what change= s occur before putting software into production, but, like most things of t= he last 25 years (high capacity disk drives, plummeting memory prices, and = the ongoing speed increases of CPUs), people have become lazy in doing thin= gs the right way.=A0 When things were tight, people thought at the 'assembl= er' level to program lean, mean and fast, C was a boon as it's kind of a le= vel 2.5 Von Neuman language, you can access low-level but it's structured l= ike a level 3 language.=A0 Now, people don't really think of machine resour= ces.=A0 They just hack together things and hope the compiler catches their = mistakes.=A0=A0=A0 *shrug*=0A</rant>=0A=0AP.=0A=0A=0APS - (I'd post my cred= entials, but, basically, I'm a Systems Architect that has a past with emplo= yment or consulting with many major corporations.=A0 My job is creating sys= tems of systems that are highly scalable and modular and can be nimble in m= oving from one tech to another.=A0 I've been exposed to almost every *NIX t= ype of OS, Windows and other OS variants....=A0 I'm still impressed with FB= SD as I'm a CS and I have watched it always try to be cutting edge and alwa= ys implement the correct technologies and refuses to compromise by releasin= g a<Version>.0 that 'kinda works'That's how I've lived my career.=A0 Kudos = to all the people working on it!)=0A=0APPS - I have to make my living using= all variants of *nix including Debian, CentOS, RH, SuSe, etc.=A0 Also, Ibe= lieve I was the first one to create a SAN in 1993 while at EMC.=A0 I've wor= ked with many Linux variants.=A0 However, when I look at who is really usin= g BSD.... Cisco, Juniper, NetApp, and many major manufacturers base their *= NIX products on it and kind of give away Linux for free but they are really= happy to get consulting hours at $200-$400/hr to work it....=A0 So, I'm no= t a 'fan boy' , I'm somebody who respects the mindset of the 'best tech to = solve the problem'.=0A=0A=0A=0A________________________________=0A From: Bi= ll Totman <bill.totman@gmail.com>=0ATo: davide.damico@contactlab.com =0ACc:= freebsd-performance@freebsd.org =0ASent: Sunday, March 24, 2013 10:11 AM= =0ASubject: Re: FreeBSD 9.1 vs CentOS 6.3=0A =0AOn 3/23/13 3:44 AM, Davide = D'Amico wrote:=0A> Il 23.03.2013 01:34 Paul Pathiakis ha scritto:=0A>> Hi,= =0A>> =0A>> There are several things about this that are highly suspect.=0A= >> =0A>> First, wipe out the hardware RAID. The processor doing RAID=0A>> c= omputation is, probably, MUCH slower than a core on the CPU. Even if=0A>> i= t's RAID-1 (Simple Mirror) this RAID card is performing tasks that is=0A>> = does not need to do including replicating writes to two targets from=0A>> t= he controller or checking it's cache, battery, etc. If it's possible=0A>> t= o disable the onboard cache, do it.=0A> =0A> Hi Paul,=0A> thanks for your s= uggestions (some of them I've applied before starting any consideration, li= ke disabling all on-disk caches or controller buffers) I'll try next monday= .=0A> =0A> Anyway, the fact is that using the same hardware configuration (= raid1+raid10) I saw that a centos 6.x outperformed freebsd 9.1.=0A> Another= test I made yesterday was: on the same hardware I installed vmware esx 5.x= and created a vm with centos inside it. The result was really impressive: = the centos vm outperformed the 'real' freebsd 9.1 too and checking vmware p= erformances graphs I didn't see any huge need for a massive throughput (I s= aw values from KBps to 10MBps), instead I saw a big use of CPU (using OLTP = tests with a concurrency of 32 threads it's performaces began to slow down)= .=0A> =0ASo, what happened when you installed FreeBSD 9.1 in the VM? How di= d the 'fake' FreeBSD 9.1 compare 1) to the 'real', and 2) to either of the = CentOS installations?=0A=0A-bt=0A> I don't know is using some magic value f= or HZ or setting some trick with scheduler, I could gain something: I hope = so, because I don't want to "pinguinate" my farm :)=0A> =0A> Thanks,=0A> d.= =0A> =0A> _______________________________________________=0A> freebsd-perfo= rmance@freebsd.org mailing list=0A> http://lists.freebsd.org/mailman/listin= fo/freebsd-performance=0A> To unsubscribe, send any mail to "freebsd-perfor= mance-unsubscribe@freebsd.org"=0A=0A_______________________________________= ________=0Afreebsd-performance@freebsd.org mailing list=0Ahttp://lists.free= bsd.org/mailman/listinfo/freebsd-performance=0ATo unsubscribe, send any mai= l to "freebsd-performance-unsubscribe@freebsd.org" From owner-freebsd-performance@FreeBSD.ORG Sun Mar 24 16:01:06 2013 Return-Path: <owner-freebsd-performance@FreeBSD.ORG> Delivered-To: freebsd-performance@freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:1900:2254:206a::19:1]) by hub.freebsd.org (Postfix) with ESMTP id B959263B for <freebsd-performance@freebsd.org>; Sun, 24 Mar 2013 16:01:06 +0000 (UTC) (envelope-from amvandemore@gmail.com) Received: from mail-wi0-x233.google.com (mail-wi0-x233.google.com [IPv6:2a00:1450:400c:c05::233]) by mx1.freebsd.org (Postfix) with ESMTP id 5A1921DF for <freebsd-performance@freebsd.org>; Sun, 24 Mar 2013 16:01:06 +0000 (UTC) Received: by mail-wi0-f179.google.com with SMTP id hn17so3193223wib.6 for <freebsd-performance@freebsd.org>; Sun, 24 Mar 2013 09:01:05 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20120113; h=mime-version:x-received:in-reply-to:references:date:message-id :subject:from:to:cc:content-type; bh=5nZ8bn0UzjEUPZt1R5v/7a7mnUO9IeDU7UsyRFs9Q3U=; b=sd1th+4cmbjP8mmdNOX+LHG9jmMaL0f/G1ZnZ1PmRfv0VudDsaGwDbxrZf+vw5AfZ5 kMVCaLiQ6WWOUujai0eJsW4T512gbe9VJvTZ5XzMLwAWTA/ewRDfoM4Pur7FXp/TzHR6 ggTxJL/yAwd1cPl218L3z8qxhg1AZ9Fb455ewESotRGpwCM//X1EUPxGZWGNfgGJeb6c 1ByOUxhsdki/n+xCwkp0PKtiYo7pgjzyb75UdAjOIX5uxtoDPrO9tijDkkrXKmcbm4f2 mSmf5dNuRuhMaNtGhtE94RDZeF6J72wl/6zZpeLH3/Q7S2jBb1shootzE/jsWWdFkVZy 5AKw== MIME-Version: 1.0 X-Received: by 10.180.109.82 with SMTP id hq18mr12856083wib.0.1364140865477; Sun, 24 Mar 2013 09:01:05 -0700 (PDT) Received: by 10.194.140.20 with HTTP; Sun, 24 Mar 2013 09:01:05 -0700 (PDT) In-Reply-To: <20130323213406.93cc3baddf69d5d71f10365e@neosystem.cz> References: <514C1E5F.8040504@contactlab.com> <20130323213406.93cc3baddf69d5d71f10365e@neosystem.cz> Date: Sun, 24 Mar 2013 11:01:05 -0500 Message-ID: <CA+tpaK2JK3xhEc_RrOCAdEB1vvapEHE=VqvY5=kSM-Bkhy07PA@mail.gmail.com> Subject: Re: FreeBSD 9.1 vs CentOS 6.3 From: Adam Vande More <amvandemore@gmail.com> To: Daniel Bilik <daniel.bilik@neosystem.cz> Content-Type: text/plain; charset=ISO-8859-1 X-Content-Filtered-By: Mailman/MimeDel 2.1.14 Cc: freebsd-performance@freebsd.org X-BeenThere: freebsd-performance@freebsd.org X-Mailman-Version: 2.1.14 Precedence: list List-Id: Performance/tuning <freebsd-performance.freebsd.org> List-Unsubscribe: <http://lists.freebsd.org/mailman/options/freebsd-performance>, <mailto:freebsd-performance-request@freebsd.org?subject=unsubscribe> List-Archive: <http://lists.freebsd.org/pipermail/freebsd-performance>; List-Post: <mailto:freebsd-performance@freebsd.org> List-Help: <mailto:freebsd-performance-request@freebsd.org?subject=help> List-Subscribe: <http://lists.freebsd.org/mailman/listinfo/freebsd-performance>, <mailto:freebsd-performance-request@freebsd.org?subject=subscribe> X-List-Received-Date: Sun, 24 Mar 2013 16:01:06 -0000 These are interesting results. Did you try tuning any of the jemalloc options in /etc/malloc.conf? On Sat, Mar 23, 2013 at 3:34 PM, Daniel Bilik <daniel.bilik@neosystem.cz>wrote: > On Fri, 22 Mar 2013 10:03:27 +0100 > Davide D'Amico <davide.damico@contactlab.com> wrote: > > > Hi, I'm doing performance tests on a DELL R720, follows dmesg: > > ... > > I will use this server as a mysql-5.6 dbserver so I have a root > > partition using a hw raid1 and a /DATAZFS partition, follows > > configuration: > > ... > > Well, it seems to be interesting coincidence... We've just finished > benchmarking MySQL with various (m)allocators. The goal was to test > tcmalloc, but when the system was up and running, we've taken the > opportunity to benchmark also other alternatives... including jemalloc. > All tests were performed on default MySQL 5.5.28 running on Debian Wheezy. > Between the tests nothing was touched on the machine or the system, just > allocators were changed (ie. mysqld restarted). > > Results for different test modes are available here... > > http://neosystem.cz/benchmark/mysql/ > > It seems there is notable performance penalty for read-only transactions > when MySQL is using jemalloc. The more concurrent threads are running, the > more is jemalloc losing to other allocators. The penalty is also there for > read-write transactions, but not that significant (error bars in the > histograms also show that results for read-write tests tend to be very > unstable). OTOH in non-transactional tests, jemalloc seems to be in par > with others, and under specific load can even outperform some of them. > > In your original post, there is not mentioned in what mode you've performed > OLTP test, but according to numbers I suspect it was "complex", ie. > transactional. Can you repeat tests (both on CentOS and FreeBSD) with > --oltp-test-mode=nontrx and/or simple? > > -- > Daniel Bilik > neosystem.cz > _______________________________________________ > freebsd-performance@freebsd.org mailing list > http://lists.freebsd.org/mailman/listinfo/freebsd-performance > To unsubscribe, send any mail to " > freebsd-performance-unsubscribe@freebsd.org" > -- Adam Vande More
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?1364139828.17881.YahooMailNeo>