Here is an interesting view on flash drive developments... http://www.eetimes.com/showArticle.jhtml?articleID=209100803&cid=NL_eet -- richard
On Fri, Jul 18, 2008 at 3:11 PM, Richard Elling <Richard.Elling at sun.com> wrote:> Here is an interesting view on flash drive developments... > http://www.eetimes.com/showArticle.jhtml?articleID=209100803&cid=NL_eet > -- richardThanks Richard. I saw that announcement in a couple of other techsites during the week. This looks like a golden opportunity for Sun to lead, rather than lag, the market in a new, highly anticipated, market segment. If you look at the overall I/O throughput in Mb/Sec over the years and compare it with the advances in server memory size or SPEC-int rates over the years, the I/O throughput curve looks *almost* flat - as the delta between the other two curves continues to widen. What is even worse, is that to improve the I/O curve, especially in terms of I/O Ops/Sec (IOPS) is very, very expensive. Flash drives - or even a RAM/Flash drive combo - have the capability to dramatically change the IOPS curves for the first time in 10 years. And almost everyone I know, wants to own and deploy a flash based storage system as soon as the price/performance ratio is "reasonable". Historically RDBMS (or other transaction based systems) were limited by CPU "horsepower". This is no longer the case - now it''s simply a matter of IOPS as multi-core CPUs are widely available at commodity prices. And solid state storage systems will dramatically increase transactions/Sec for any RDBMS type system with no other changes to the production platform. I can''t wait for the market to reach commodity price/performance..... Regards, -- Al Hopper Logical Approach Inc,Plano,TX al at logical-approach.com Voice: 972.379.2133 Timezone: US CDT OpenSolaris Governing Board (OGB) Member - Apr 2005 to Mar 2007 http://www.opensolaris.org/os/community/ogb/ogb_2005-2007/
On Fri, 18 Jul 2008, Al Hopper wrote:> If you look at the overall I/O throughput in Mb/Sec over the years and > compare it with the advances in server memory size or SPEC-int rates > over the years, the I/O throughput curve looks *almost* flat - as the > delta between the other two curves continues to widen. What is even > worse, is that to improve the I/O curve, especially in terms of I/O > Ops/Sec (IOPS) is very, very expensive. Flash drives - or even aWhile I certainly agree with you regarding IOPS, I would not agree that I/O performance gains have been flat. Sequential I/O performance has been increasing over the years similar to memory performance. 220MB/second is a lot faster than 3MB/second. We are now in a situation where per-core CPU performance has stalled (actually retreated by 30%), RAM memory performance has almost stalled, and disk seek times have completely stalled. Solid-state devices will definitely help dramatically with IOPS (by eliminating the seek) but it seems that the vast majority of solid state devices will still be much slower at sequential write I/O than traditional hard drives. It is fully conceivable that in a couple of years, hard drives will be supporting 400MB/second with 4TB of storage each. As far as saving power goes, a recent study found that FLASH drives were consuming considerably more power than the hard drives they replaced. This study likely applies to laptop/desktop applications rather than server applications. Bob =====================================Bob Friesenhahn bfriesen at simple.dallas.tx.us, http://www.simplesystems.org/users/bfriesen/ GraphicsMagick Maintainer, http://www.GraphicsMagick.org/
Except the article was redacted. The reason the battery life decreased was because the throughput increased so much that it drove up the cpu usage up, thus bringing down battery life. It just goes to show how SEVERELY io bound we currently are. The flash itself was using LESS power. --tim On 7/18/08, Bob Friesenhahn <bfriesen at simple.dallas.tx.us> wrote:> On Fri, 18 Jul 2008, Al Hopper wrote: > >> If you look at the overall I/O throughput in Mb/Sec over the years and >> compare it with the advances in server memory size or SPEC-int rates >> over the years, the I/O throughput curve looks *almost* flat - as the >> delta between the other two curves continues to widen. What is even >> worse, is that to improve the I/O curve, especially in terms of I/O >> Ops/Sec (IOPS) is very, very expensive. Flash drives - or even a > > While I certainly agree with you regarding IOPS, I would not agree > that I/O performance gains have been flat. Sequential I/O performance > has been increasing over the years similar to memory performance. > 220MB/second is a lot faster than 3MB/second. We are now in a > situation where per-core CPU performance has stalled (actually > retreated by 30%), RAM memory performance has almost stalled, and disk > seek times have completely stalled. > > Solid-state devices will definitely help dramatically with IOPS (by > eliminating the seek) but it seems that the vast majority of solid > state devices will still be much slower at sequential write I/O than > traditional hard drives. It is fully conceivable that in a couple of > years, hard drives will be supporting 400MB/second with 4TB of storage > each. > > As far as saving power goes, a recent study found that FLASH drives > were consuming considerably more power than the hard drives they > replaced. This study likely applies to laptop/desktop applications > rather than server applications. > > Bob > =====================================> Bob Friesenhahn > bfriesen at simple.dallas.tx.us, http://www.simplesystems.org/users/bfriesen/ > GraphicsMagick Maintainer, http://www.GraphicsMagick.org/ > > _______________________________________________ > zfs-discuss mailing list > zfs-discuss at opensolaris.org > http://mail.opensolaris.org/mailman/listinfo/zfs-discuss >
On Fri, 18 Jul 2008, Tim wrote:> Except the article was redacted. The reason the battery life > decreased was because the throughput increased so much that it drove > up the cpu usage up, thus bringing down battery life. It just goes to > show how SEVERELY io bound we currently are. The flash itself was > using LESS power.Any study will show that (similar to humans) computers make little use of their brains. The CPU spends a large part of its time idle waiting for data be copied from RAM to second level cache, and then to first level cache. The RAM sits idle waiting for data to be copied from disk or network. That is why Sun''s SPARC T1/T2 CMT technology does not suck in servers. Unfortunately, except for the database random access cases and synchronous write cases, the solid state memory will still not make much difference since it is so much remarkably slower than system RAM. If RAM is good for 10GB/second (or 20GB/second for fastest system), the FLASH device might be good for 100MB/second. Fortunately for us, most computers in the world spend 99.99% of their time waiting for the human to do something. Bob =====================================Bob Friesenhahn bfriesen at simple.dallas.tx.us, http://www.simplesystems.org/users/bfriesen/ GraphicsMagick Maintainer, http://www.GraphicsMagick.org/
Bob Friesenhahn wrote:> Any study will show that (similar to humans) computers make little use > of their brains. >Surely there''s no reason to continue spreading that myth. http://faculty.washington.edu/chudler/tenper.html -Grant