Tom, thanks a lots for your detailed explaination. I have one more question. You've mentioned there are around 200 checkpoints in a typical sample size. I found one simics checkpoint for 16 processor is at least 200MB. Then 200 checkpoints will be 40GB. That's huge, do I miss something here?
thanks a lot! Legion -------------- next part -------------- An HTML attachment was scrubbed... URL: http://sos.ece.cmu.edu/pipermail/simflex/attachments/20060228/a9dbf4c0/attachment.html From twenisch at ece.cmu.edu Tue Feb 28 14:49:33 2006 From: twenisch at ece.cmu.edu (Thomas Wenisch) List-Post: [email protected] Date: Tue Feb 28 14:49:32 2006 Subject: [Simflex] Re: question on representative queries in TPC-H In-Reply-To: <[email protected]> References: <[email protected]> Message-ID: <pine.lnx.4.53l-ece.cmu.edu.0602281343370.13...@dalmore.ece.cmu.edu> Hi Legion, On Tue, 28 Feb 2006, Liqun Cheng wrote: > Tom, > > thanks a lots for your detailed explaination. I have one more question. > You've mentioned there are around 200 checkpoints in a typical sample > size. I found one simics checkpoint for 16 processor is at least 200MB. Then > 200 checkpoints will be 40GB. That's huge, do I miss something here? No, you haven't missed anything - Simics checkpoints require a lot of space. I did a quick survey of our checkpoint area to give you more precise guidance on space: ~40 GB for the base disk images for our DB2 TPCC/TPCH installation. This includes all the disk space in the simulated system for the 100 warehouse TPC-C and TPC-H databases, which are in excess of 15GB each. 2-5GB per TPC-H query for checkpoints. This works out to be much smaller than the 40GB estimate above for two reasons. First, these checkpoints are typically smaller than 200MB because they are closer together, and the delta memory image between checkponts is smaller. Second, my estimate of 200-checkpoint samples was a little high - that is typical for transaction processing apps (TPC-C), which tend to be more variable than DSS queries (and we do have 10-20GB checkpoint sets for TPC-C). If you use matched-pair sample comparison (see our 2006 ISPASS paper or Ekman and Stenstrom's ISPASS paper from 2005), you can get away with samples of 75-100 checkpoints. However, the short version is that you definately need big disks, or better yet, a big NFS volume. Best Regards, -Tom Wenisch Computer Architecture Lab Carnegie Mellon University
