X-Original-To: pgsql-performance-postgresql.org@localhost.postgresql.org Received: from localhost (unknown [200.46.204.144]) by svr1.postgresql.org (Postfix) with ESMTP id 17A4D3A5D5E for ; Wed, 12 Jan 2005 16:57:59 +0000 (GMT) Received: from svr1.postgresql.org ([200.46.204.71]) by localhost (av.hub.org [200.46.204.144]) (amavisd-new, port 10024) with ESMTP id 44371-02 for ; Wed, 12 Jan 2005 16:57:52 +0000 (GMT) Received: from wproxy.gmail.com (wproxy.gmail.com [64.233.184.202]) by svr1.postgresql.org (Postfix) with ESMTP id C19393A5D32 for ; Wed, 12 Jan 2005 16:57:48 +0000 (GMT) Received: by wproxy.gmail.com with SMTP id 71so579300wra for ; Wed, 12 Jan 2005 08:57:47 -0800 (PST) DomainKey-Signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:message-id:date:from:reply-to:to:subject:cc:in-reply-to:mime-version:content-type:content-transfer-encoding:references; b=HLfw12p9cs51qoJGMlSvRI3R1ruGY82KEem1p49/tNf8SQFQfkvOz134e4aml0GOWmH331QQmeeP77MATgSKpukEJNVpibqAzas4EKs3XOiWs/Cywbrwh3FZTeV/QWUTdL7hYuclq5T7Po/6EL8JjGqdwXJg3p5DHm37pqMGMu8= Received: by 10.54.50.57 with SMTP id x57mr522047wrx; Wed, 12 Jan 2005 08:57:47 -0800 (PST) Received: by 10.54.18.72 with HTTP; Wed, 12 Jan 2005 08:57:47 -0800 (PST) Message-ID: <33c6269f050112085763aff4a6@mail.gmail.com> Date: Wed, 12 Jan 2005 11:57:47 -0500 From: Alex Turner Reply-To: alex@neteconomist.com To: Greg Stark Subject: Re: which dual-CPU hardware/OS is fastest for PostgreSQL? Cc: Merlin Moncure , pgsql-performance@postgresql.org In-Reply-To: <87zmzg0ycq.fsf@stark.xeocode.com> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit References: <6EE64EF3AB31D5448D0007DD34EEB3412A75A3@Herge.rcsinc.local> <87zmzg0ycq.fsf@stark.xeocode.com> X-Virus-Scanned: by amavisd-new at hub.org X-Spam-Status: No, hits=0.024 tagged_above=0 required=5 tests=RCVD_BY_IP X-Spam-Level: X-Archive-Number: 200501/151 X-Sequence-Number: 9841 Infact the cache hit ratio that Oracle suggests is the minimum good value is 95%. Anything below that is bad news. The reason is pretty obvious - RAM transfer speed is around 3.2G/sec these days, whilst even the best array isn't going to give more than 400MB/sec, and that's not even starting to talk about seek time. anything below 90% is not going to keep even the best disc hardware saturated. I know that our dataset is 99% cached, and therefore better CPUs/better RAM has a huge impact on overall performance. Alex Turner NetEconomist On 11 Jan 2005 10:39:01 -0500, Greg Stark wrote: > > "Merlin Moncure" writes: > > > heh, our apps do tend to be CPU bound. Generally, I think the extra CPU > > horsepower is worth the investment until you get to the really high end > > cpus. > > I find that while most applications I work with shouldn't be cpu intensive > they do seem end up being cpu bound quite frequently. What happens is that 90% > of the workload has a working set that fits in RAM. So the system ends up > being bound by the memory bus speed. That appears exactly the same as > cpu-bound, though I'm unclear whether increasing the cpu clock will help. > > It's quite possible to have this situation at the same time as other queries > are i/o bound. It's quite common to have 95% of your workload be frequently > executed fast queries on commonly accessed data and 5% be bigger data > warehouse style queries that need to do large sequential reads. > > Incidentally, the same was true for Oracle on Solaris. If we found excessive > cpu use typically meant some frequently executed query was using a sequential > scan on a small table. Small enough to fit in RAM but large enough to consume > lots of cycles reading it. > > -- > greg > >