X-Received: by 10.236.150.114 with SMTP id y78mr6503571yhj.7.1402505410700; Wed, 11 Jun 2014 09:50:10 -0700 (PDT) X-Received: by 10.50.164.129 with SMTP id yq1mr677536igb.15.1402505410493; Wed, 11 Jun 2014 09:50:10 -0700 (PDT) Path: news.cambrium.nl!textnews.cambrium.nl!feeder1.cambriumusenet.nl!feed.tweaknews.nl!209.85.216.88.MISMATCH!s7no539021qap.1!news-out.google.com!qf4ni19599igc.0!nntp.google.com!a13no1285237igq.0!postnews.google.com!glegroupsg2000goo.googlegroups.com!not-for-mail Newsgroups: comp.databases.oracle.server Date: Wed, 11 Jun 2014 09:50:09 -0700 (PDT) In-Reply-To: <0e14a914-3194-435e-b185-b91f47eca3f1@googlegroups.com> Complaints-To: groups-abuse@google.com Injection-Info: glegroupsg2000goo.googlegroups.com; posting-host=155.72.28.11; posting-account=PtUHNAoAAACxzjf46HeljEY0yw_sq9w0 NNTP-Posting-Host: 155.72.28.11 References: <0e14a914-3194-435e-b185-b91f47eca3f1@googlegroups.com> User-Agent: G2/1.0 MIME-Version: 1.0 Message-ID: <56dc19a1-330e-448c-a601-64ae3f8560e6@googlegroups.com> Subject: Re: in-memory option From: Arup Nanda Injection-Date: Wed, 11 Jun 2014 16:50:10 +0000 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: quoted-printable Xref: news.cambrium.nl On Wednesday, June 11, 2014 1:15:46 AM UTC-4, vsevolod afanassiev wrote: > Not much new info in the presentation by Larry Ellison compared with the = Open World presentation. It would be nice to know: >=20 > - how much the In-Memory Option will cost - I assume it will be separatel= y licensed option like Partitioning >=20 > - how to estimate size of a table in compressed columnar format >=20 > - what happens if a table grows and no longer fits in memory?=20 >=20 > - how long it takes to populate the in-memory columnar tables after datab= ase restart >=20 > - the columnar in-memory store - is it going to be re-sized by automatic = shared memory management? or it will remain static? >=20 > - in all the 'real-life' examples - did they compare performance on the s= erver with the same amount of memory? Or they added memory for the In-Memor= y Option? Disclaimer: I don't work for Oracle; so I can't answer questions like cost,= licensing and if a specific feature will work exactly that way when the pr= oduction version comes. Please consider my responses with that background. > - how to estimate size of a table in compressed columnar format Exact space depends on the types of compression. There is an estimate funct= ion available to estimate the space usage, very similar to the space usage = in Hybrid Columnar Compression. =20 > - what happens if a table grows and no longer fits in memory?=20 No worries. Remember, this is an *additional* cache; not a replacement. The= table will be subject to the same approach as in a non-InMem option databa= se. Besides, you may not want to put all the tables in InMem store. Those n= ot in there will be subjet to the old behavior.=20 > - how long it takes to populate the in-memory columnar tables after datab= ase restart Depends on the table size. You will be able to see the amount loaded and pe= rcentage remaining. In my tests I saw a 1 TB column store getting loaded in= 1 hour. YMMV. > - the columnar in-memory store - is it going to be re-sized by automatic = shared memory management? or it will remain static? No; it's different from ASMM.=20 > - in all the 'real-life' examples - did they compare performance on the s= erver with the same amount of memory? Or they added memory for the In-Memor= y Option? I did not add memory; I removed it from the buffer cache to give it to the = InMem cache. Thanks.=20 Arup Nanda @ArupNanda arup.blogspot.com