Mailing-List: contact dev-help@hbase.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@hbase.apache.org
Received-SPF: pass (athena.apache.org: local policy)
DomainKey-Signature: a=rsa-sha1; q=dns; c=nofws;
  s=s1024; d=yahoo.com;
  h=X-YMail-OSG:Received:X-Mailer:References:Message-ID:Date:From:Reply-To:Subject:To:In-Reply-To:MIME-Version:Content-Type:Content-Transfer-Encoding;
  b=lw9mFE3bOqx2HDeC4n0DCw6wTdzIc3JXDwD6fvLEOAmfCHi+G2avkVcI8S4Wdh1WcNv9swgn4iSamSEWseMfKTLfdZVRC7lkk6hzq+N2cwF22wEDrrPXUe4S6J4mWRzU9dV8HX6wmKiRpjETIiSEevssfqRP/xkC25V/JBjw+D4=;
References: <1335914771.79923.YahooMailNeo@web121706.mail.ne1.yahoo.com>
 <CADY20s7rmTinW5m3q2dvMhPPj9e_7pu7T8wVwpWBmHGVQjASVA@mail.gmail.com>
 <1335936048.37273.YahooMailNeo@web121703.mail.ne1.yahoo.com>
 <CAKYwJ9ytcQKOtDin776TpoJ8ceuad16DscWKQzpM45=BNRjO4Q@mail.gmail.com>
 <CALte62ybPHr1UHXY8NPXP6UyJejWvjV=Y5rWVKx3BE1JxOG-RQ@mail.gmail.com>
 <CAKYwJ9w-K6uHZ2gjACiT6ci_hHsGLuJxWOeLZ+QKU4fOH+odYQ@mail.gmail.com>
 <CALte62z_Son4ZcW7Veh2ZQ7qKqsZszaFYRdVixJf9p6HJStm5g@mail.gmail.com>
 <CAPpvZ9Bxy44m-SDvMtGdSVzr1WdjS5yf=-2K61AXGSOfqMZeMQ@mail.gmail.com>
 <CAKYwJ9yFm8nCpc7+Ann_yLCVim_zzwgZHgQj_iFFtQNJzRAZJg@mail.gmail.com>
 <CAKYwJ9yPsJtiD3BvmOqMq+GGS4Z2J2fONGkjk1AGEO576ZSoZA@mail.gmail.com>
 <CALte62y+m4UmrCnXGkckLZzNaR1ZhQ-V2Z1vPOAq-Th+qSF8FA@mail.gmail.com>
 <CAKYwJ9zXFNn60xg71ueZ5gnbNRSfyigWnFzjHnQJMr9PW=6Cmw@mail.gmail.com>
 <CALte62xjG7U3HwT7zjCG+3jM_a38rrwe3ZmPuySUJNsEPB74xg@mail.gmail.com>
 <CAKYwJ9yFCWzyXyuJK5mbDJC6FfNg8rdWHu7s3TCx5ycjaaarvQ@mail.gmail.com>
 <CADY20s7JabcVQi8PhTcMA5jGrOGB0t+fim
 ONH0nWMP2dE9q-9w@mail.gmail.com>
 <CAKYwJ9x1Ek4eUcUhPrqk7EHJs5=_TDtB1i06WAV=NJj5b56P=w@mail.gmail.com>
Message-ID: <1336508493.42409.YahooMailNeo@web121703.mail.ne1.yahoo.com>
Date: Tue, 8 May 2012 13:21:33 -0700 (PDT)
From: lars hofhansl <lhofhansl@yahoo.com>
Reply-To: lars hofhansl <lhofhansl@yahoo.com>
Subject: Re: ANN: The third hbase 0.94.0 release candidate is available for
 download
To: "dev@hbase.apache.org" <dev@hbase.apache.org>
In-Reply-To: 
 <CAKYwJ9x1Ek4eUcUhPrqk7EHJs5=_TDtB1i06WAV=NJj5b56P=w@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Transfer-Encoding: quoted-printable

Hmm... So our "performance release" is slightly slower than 0.92.=0AWith al=
l the optimizations that went into 0.94 I find that a bit hard to believe.=
=0A=0ACan you tell us more about the testing? How many machines, setup, was=
 that test IO or CPU bound, etc?=0AAnything else of note?=0A=0AThanks for d=
oing this!=0A=0A-- Lars=0A=0A________________________________=0AFrom: Ellio=
tt Clark <eclark@stumbleupon.com>=0ATo: dev@hbase.apache.org =0ASent: Monda=
y, May 7, 2012 11:07 AM=0ASubject: Re: ANN: The third hbase 0.94.0 release =
candidate is available for download=0A=0ASorry everything is in elapsed tim=
e as reported by Elapsed time in=0Amilliseconds.=A0 So higher is worse.=0A=
=0AThe standard deviation on 0.92.1 writes is 4,591,384 so Write 5 is a lit=
tle=0Aoutside of 1 std dev.=A0 Not really sure what happened on that test, =
but it=0Adoes appear that PE is very noisy.=0A=0AOn Mon, May 7, 2012 at 10:=
47 AM, Todd Lipcon <todd@cloudera.com> wrote:=0A=0A> Is higher better or wo=
rse? :) Any idea what happened on the "Write 5" test?=0A>=0A> On Mon, May 7=
, 2012 at 10:42 AM, Elliott Clark <eclark@stumbleupon.com>=0A> wrote:=0A> >=
 http://www.scribd.com/eclark847297/d/92715238-0-94-0-RC3-Cluster-Perf=0A> =
>=0A> > On Fri, May 4, 2012 at 7:42 PM, Ted Yu <yuzhihong@gmail.com> wrote:=
=0A> >=0A> >> 0.94 also has LoadTestTool (from FB)=0A> >>=0A> >> I have use=
d it to do some cluster load testing.=0A> >>=0A> >> Just FYI=0A> >>=0A> >> =
On Fri, May 4, 2012 at 3:14 PM, Elliott Clark <eclark@stumbleupon.com=0A> >=
> >wrote:=0A> >>=0A> >> > With the cluster size that I'm testing YCSB was s=
tressing the client=0A> >> > machine more than the cluster.=A0 I was satura=
ting the network of the=0A> test=0A> >> > machine.=A0 So I switched over to=
 pe; while it doesn't have a realistic=0A> >> work=0A> >> > load it is bett=
er than nothing.=0A> >> >=0A> >> > On Fri, May 4, 2012 at 3:07 PM, Ted Yu <=
yuzhihong@gmail.com> wrote:=0A> >> >=0A> >> > > Thanks for the update, Elli=
ot.=0A> >> > >=0A> >> > > If I read your post correctly, you're using PE. y=
csb is better=0A> >> measuring=0A> >> > > performance, from my experience.=
=0A> >> > >=0A> >> > > Cheers=0A> >> > >=0A> >> > > On Fri, May 4, 2012 at =
3:04 PM, Elliott Clark <=0A> eclark@stumbleupon.com=0A> >> > > >wrote:=0A> =
>> > >=0A> >> > > > So I got 94.0rc3 up on a cluster and tried to break it,=
 Killing=0A> >> masters=0A> >> > > and=0A> >> > > > killing rs.=A0 Everythi=
ng seems good. hbck reports everything is=0A> good.=0A> >> >=A0 And=0A> >> =
> > > all my reads succeed.=0A> >> > > >=0A> >> > > > I'll post cluster ben=
chmark numbers once they are done running.=0A> >>=A0 Should=0A> >> > > > on=
ly be a couple more hours of pe runs.=0A> >> > > >=0A> >> > > > Looks great=
 to me.=0A> >> > > > On Thu, May 3, 2012 at 10:36 AM, Elliott Clark <=0A> >=
> eclark@stumbleupon.com=0A> >> > > > >wrote:=0A> >> > > >=0A> >> > > > > I=
 agree it was just a micro benchmark with no guarantee that it=0A> >> > rel=
ates=0A> >> > > > to=0A> >> > > > > real world. With it just being standalo=
ne I didn't think anyone=0A> >> > should=0A> >> > > > take=0A> >> > > > > t=
he numbers as 100% representative.=A0 Really I was just trying to=0A> >> > =
shake=0A> >> > > > out=0A> >> > > > > any weird behaviors and the fact that=
 we got a big speed up was=0A> >> > > > > interesting.=0A> >> > > > >=0A> >=
> > > > >=0A> >> > > > > On Thu, May 3, 2012 at 12:15 AM, Mikael Sitruk <=
=0A> >> > > mikael.sitruk@gmail.com=0A> >> > > > >wrote:=0A> >> > > > >=0A>=
 >> > > > >> Hi guys=0A> >> > > > >> Looking at the posted slide/pictures f=
or the benchmark the=0A> >> > > > >> following intriguing me:=0A> >> > > > =
>> 1. The recordcount is only 100,000=0A> >> > > > >> 2. workoloada is: rea=
d 50%, update 50% and zipfian distribution=0A> >> even=0A> >> > > > with=0A=
> >> > > > >> 5M operations count, the same keys are updated again and agai=
n.=0A> >> > > > >> 3. heap size 10G=0A> >> > > > >>=0A> >> > > > >> Therefo=
re it might be that the dataset is too small (even with=0A> 3=0A> >> > > > =
versions=0A> >> > > > >> configured we have =3D 3(version)*100,000(keys)*1K=
B (size of=0A> >> record) =3D=0A> >> > > 300=0A> >> > > > >> MB=0A> >> > > =
> >> of "live" dataset ?=0A> >> > > > >> And approximately the number of st=
ore files will be 5x10^6 (op=0A> >> > > > >> count)*1KB(record size)/256MB(=
max store file size=0A> (Default))=3D>20=0A> >> > store=0A> >> > > > >> fil=
e, even taking factor of 10 for metadata (record key, in=0A> store=0A> >> >=
 > files)=0A> >> > > > >> we=0A> >> > > > >> will get 200 files.=0A> >> > >=
 > >> if a major compaction is running it will shrink all the=0A> storefile=
=0A> >> > to a=0A> >> > > > >> single small one.=0A> >> > > > >> What I try=
 to say is - if the maths are correct - (please note=0A> >> that=0A> >> > i=
=0A> >> > > > did=0A> >> > > > >> not take into account compression which j=
ust make things=0A> better),=0A> >> > can=0A> >> > > we=0A> >> > > > >> rel=
ate on such scenario for performance benchmark with such=0A> small=0A> >> >=
 > > dataset=0A> >> > > > >> and such distribution?=0A> >> > > > >>=0A> >> =
> > > >> Regards=0A> >> > > > >> Mikael.S=0A> >> > > > >>=0A> >> > > > >> O=
n Thu, May 3, 2012 at 1:10 AM, Ted Yu <yuzhihong@gmail.com>=0A> >> wrote:=
=0A> >> > > > >>=0A> >> > > > >> > I am surprised to see 0.92.1 exhibit suc=
h unfavorable=0A> >> performance=0A> >> > > > >> profile.=0A> >> > > > >> >=
 Let's see whether cluster testing gives us similar results.=0A> >> > > > >=
> >=0A> >> > > > >> > On Wed, May 2, 2012 at 3:07 PM, Elliott Clark <=0A> >=
> > > eclark@stumbleupon.com=0A> >> > > > >> > >wrote:=0A> >> > > > >> >=0A=
> >> > > > >> > > Sure, sorry about that.=0A> >> > > > >> > >=0A> >> > > > =
>> > > http://imgur.com/waxlS=0A> >> > > > >> > >=0A> >> > > > >> > >=0A> >=
> > > > >> >=0A> >> > > > >>=0A> >> > > >=0A> >> > >=0A> >> >=0A> >>=0A> ht=
tp://www.scribd.com/eclark847297/d/92151092-Hbase-0-94-0-RC3-Local-YCSB-Per=
f=0A> >> > > > >> > >=0A> >> > > > >> > > On Wed, May 2, 2012 at 3:01 PM, T=
ed Yu <=0A> yuzhihong@gmail.com>=0A> >> > > wrote:=0A> >> > > > >> > >=0A> =
>> > > > >> > > > Elliot:=0A> >> > > > >> > > > Thanks for the report.=0A> =
>> > > > >> > > > Can you publish results somewhere else ?=0A> >> > > > >> =
> > > Attachments were stripped off.=0A> >> > > > >> > > >=0A> >> > > > >> =
> > > On Wed, May 2, 2012 at 2:59 PM, Elliott Clark <=0A> >> > > > >> eclar=
k@stumbleupon.com=0A> >> > > > >> > > > >wrote:=0A> >> > > > >> > > >=0A> >=
> > > > >> > > > > I ran some tests of local filesystem YCSB. I used the=0A=
> 0.90=0A> >> > > > client=0A> >> > > > >> for=0A> >> > > > >> > > > > 0.90=
.6.=A0 For the rest of the tests I used 0.92 clients.=0A> >> The=0A> >> > >=
 > >> results=0A> >> > > > >> > are=0A> >> > > > >> > > > > attached.=0A> >=
> > > > >> > > > >=0A> >> > > > >> > > > > 0.90 -> 0.94.0RC3 13% faster=0A>=
 >> > > > >> > > > > 0.92 -> 0.94.0RC3 50% faster=0A> >> > > > >> > > > >=
=0A> >> > > > >> > > > >=A0 This seems to be a pretty large performance=0A>=
 improvement.=0A> >> > >=A0 I'll=0A> >> > > > >> run=0A> >> > > > >> > > so=
me=0A> >> > > > >> > > > > tests on a cluster later today.=0A> >> > > > >> =
> > > >=0A> >> > > > >> > > > > On Tue, May 1, 2012 at 10:20 PM, lars hofha=
nsl <=0A> >> > > > >> lhofhansl@yahoo.com=0A> >> > > > >> > > > >wrote:=0A>=
 >> > > > >> > > > >=0A> >> > > > >> > > > >> Thanks Todd.=0A> >> > > > >> =
> > > >>=0A> >> > > > >> > > > >> I agree with doing source code releases g=
oing forward.=0A> >> > > > >> > > > >>=0A> >> > > > >> > > > >> For that, w=
ould it be sufficient to just vote against=0A> an=0A> >> > SVN=0A> >> > > >=
 tag?=0A> >> > > > >> > > > >> Tarballs can then be pulled straight from th=
at tag.=0A> >> > > > >> > > > >>=0A> >> > > > >> > > > >> -- Lars=0A> >> > =
> > >> > > > >>=0A> >> > > > >> > > > >>=0A> >> > > > >> > > > >>=0A> >> > =
> > >> > > > >> ----- Original Message -----=0A> >> > > > >> > > > >> From:=
 Todd Lipcon <todd@cloudera.com>=0A> >> > > > >> > > > >> To: dev@hbase.apa=
che.org; lars hofhansl <=0A> >> > lhofhansl@yahoo.com=0A> >> > > >=0A> >> >=
 > > >> > > > >> Cc:=0A> >> > > > >> > > > >> Sent: Tuesday, May 1, 2012 9:=
35 PM=0A> >> > > > >> > > > >> Subject: Re: ANN: The third hbase 0.94.0 rel=
ease=0A> >> candidate=0A> >> > is=0A> >> > > > >> > > available=0A> >> > > =
> >> > > > >> for download=0A> >> > > > >> > > > >>=0A> >> > > > >> > > > >=
> +1 from me, I took it for a spin on the local=0A> filesystem=0A> >> > wit=
h=0A> >> > > > >> some=0A> >> > > > >> > > YCSB=0A> >> > > > >> > > > >> lo=
ad.=0A> >> > > > >> > > > >>=0A> >> > > > >> > > > >> Here is my signature =
on the non-secure tarball.=0A> >> > > > >> > > > >> -----BEGIN PGP SIGNATUR=
E-----=0A> >> > > > >> > > > >> Version: GnuPG v1.4.10 (GNU/Linux)=0A> >> >=
 > > >> > > > >>=0A> >> > > > >> > > > >>=0A> >> > > > iEYEABECAAYFAk+guTIA=
CgkQXkPKua7Hfq9YSQCeMnCQ4XFqLjw+PF8IXNPDug+t=0A> >> > > > >> > > > >> h90Ao=
J+q4YSg4JbfiCmaXenadWSRU1of=0A> >> > > > >> > > > >> =3DCdfZ=0A> >> > > > >=
> > > > >> -----END PGP SIGNATURE-----=0A> >> > > > >> > > > >>=0A> >> > > =
> >> > > > >> I didn't check out the secure tarball.=0A> >> > > > >> > > > =
>>=0A> >> > > > >> > > > >> I think for future releases we should do the vo=
ting=0A> >> > against a=0A> >> > > > >> source=0A> >> > > > >> > > tar=0A> =
>> > > > >> > > > >> (ie an svn export) since we now produce multiple=0A> >=
> binaries,=0A> >> > > and=0A> >> > > > >> it's=0A> >> > > > >> > > > easie=
r=0A> >> > > > >> > > > >> to verify that a source tar matches SVN, etc.=0A=
> >> > > > >> > > > >>=0A> >> > > > >> > > > >> -Todd=0A> >> > > > >> > > >=
 >>=0A> >> > > > >> > > > >>=0A> >> > > > >> > > > >> On Tue, May 1, 2012 a=
t 4:26 PM, lars hofhansl <=0A> >> > > > >> lhofhansl@yahoo.com>=0A> >> > > =
> >> > > > >> wrote:=0A> >> > > > >> > > > >>=0A> >> > > > >> > > > >> > Th=
e third 0.94.0 RC is available for download here:=0A> >> > > > >> > > > >> =
> http://people.apache.org/~larsh/hbase-0.94.0-rc3/=0A> >> > > > >> > > > >=
> > (My gpg key is available from pgp.mit.edu. Key id:=0A> >> > > 7CA45750)=
=0A> >> > > > >> > > > >> >=0A> >> > > > >> > > > >> > HBase 0.94 is a perf=
ormance release, and there are=0A> some=0A> >> > > > >> > interesting=0A> >=
> > > > >> > > > new=0A> >> > > > >> > > > >> > features as well.=0A> >> > =
> > >> > > > >> >=0A> >> > > > >> > > > >> > It is wire compatible with 0.9=
2.x. 0.92 clients=0A> should=0A> >> > work=0A> >> > > > >> with=0A> >> > > =
> >> > > 0.94=0A> >> > > > >> > > > >> > servers and vice versa.=0A> >> > >=
 > >> > > > >> >=0A> >> > > > >> > > > >> > You can do a rolling restart to=
 get your 0.92.x=0A> HBase=0A> >> up=0A> >> > on=0A> >> > > > >> this=0A> >=
> > > > >> > > > >> 0.94.0RC.=0A> >> > > > >> > > > >> >=0A> >> > > > >> > =
> > >> > The full list of changes is available here:=0A> >> > > > >> > > > =
>> >=0A> >> > > > >> > > > >> >=0A> >> > > > >> > > > >>=0A> >> > > > >> > =
> >=0A> >> > > > >> > >=0A> >> > > > >> >=0A> >> > > > >>=0A> >> > > >=0A> =
>> > >=0A> >> >=0A> >>=0A> https://issues.apache.org/jira/secure/ReleaseNot=
e.jspa?projectId=3D12310753&version=3D12316419=0A> >> > > > >> > > > >> >=
=0A> >> > > > >> > > > >> > Please take this RC for a spin, check out the d=
oc,=0A> etc,=0A> >> > and=0A> >> > > > >> vote=0A> >> > > > >> > > +1/-1=0A=
> >> > > > >> > > > >> by=0A> >> > > > >> > > > >> > May 8th on whether we =
should release this as 0.94.0.=0A> >> > > > >> > > > >> >=0A> >> > > > >> >=
 > > >> > Thanks.=0A> >> > > > >> > > > >> >=0A> >> > > > >> > > > >> > -- =
Lars=0A> >> > > > >> > > > >> >=0A> >> > > > >> > > > >>=0A> >> > > > >> > =
> > >>=0A> >> > > > >> > > > >>=0A> >> > > > >> > > > >> --=0A> >> > > > >>=
 > > > >> Todd Lipcon=0A> >> > > > >> > > > >> Software Engineer, Cloudera=
=0A> >> > > > >> > > > >>=0A> >> > > > >> > > > >>=0A> >> > > > >> > > > >=
=0A> >> > > > >> > > >=0A> >> > > > >> > >=0A> >> > > > >> >=0A> >> > > > >=
>=0A> >> > > > >=0A> >> > > > >=0A> >> > > >=0A> >> > >=0A> >> >=0A> >>=0A>=
=0A>=0A>=0A> --=0A> Todd Lipcon=0A> Software Engineer, Cloudera=0A>