My take on FCoE - wait and see.  Sure many industry "experts" say we will see everything converge tomorrow and that FC will be replaced in the future.  As you noted the same was said of iSCSI and the market still favors FC today (based on market #’s).  The reality with iSCSI was/is that there is a place and time for it.  My money says we are years (3+) before FCoE will be a major player.  There are three key things that must occur for FC to be repalced.

1) FCoE products are readily available from a variety of vendors

2) The equipment and inteconnects are cost effective (10GbE is not real cheap today, but many are working to make it so)

3) Storage Admins grow comfortable with the technology after it has proven itself to be reliable and stable.  

Even if all that proves out you still have the 800lb elephant sitting in the corner....politics.

In many organizations there are two networks today; 1)The FC network infrastructure typically run by the Storage Admins and 2) The IP network run by a separate group.  The big draw with FCoE is that you can converge the two networks and the associated cabling.  My gut feeling is that many SA's are not going to give that up to the IP guys just yet.  Maybe i'm wrong, only time will tell.  Some say that the potential cost savings from the cabling will cause many business to make this a reality, but again that was a big draw with iSCSI.  The cost difference of copper vs. Fiber has not proven to be a real deciding factor,except in the SMB space.  I’m sure someone will come along and say “my xyz company of 10trillion servers all run on iSCSI”, bu the sales number’s say different.  In the end despite the large cost difference you have two networks today and very well may have two networks tomorrow.

On the subject of SSD drives - fantastic technology.....for the right purpose.  The cost per GB is very high today and will likely remain so for the next several years, I am guessing.  On top of that it, it is too new of a technology that will need to prove itself to in the industry.  Many have some exposure to the early desktop SSD's which are not all that impressive.  Enterprise SSD is different, but the perception of high failure rates from the desktop SSD is already out there.

Personally I would like to see an array with a hybrid approach.  In my dream world the array controllers would be smart enough to watch i/o patterns on LUN’s and dynamically transition those with higher iop requirements over to a SSD based disk pool.  Kind of like a layer of secondary controller cache if you will.  This should not only be dynamic, but perhaps even allow the admin to set some sort of Quality of Service policy on a given LUN.  If no QoS policy is set in the array then it dynamically manages the LUN’s.  If there is a policy set then it will apply some sort of preference based on those settings to the defined LUN’s and run them from the SSD drive pool.  Perhaps the array management software could be smart enough to even make recommendations on which LUN’s should have a higher priority set on them.  This recommendation to the admin would be based on past performance data that it would collect and analyze on a continual basis.

Great input Mike... thanks for jumping in!  

Calvin

Past experiences with large OLTP db made me seriously doubt on the SAN hability to "detect the io patern".

==> I think this is a "lab" feature that had not be really reviewed

Facts:

-They are 2 very different SSD family : SLC (1 nand = 1 bit) and MLC (1 nand = 4 bits)

-MLC family is NOT enterprise oriented (but can be 4x cheaper)

-As of today (without real volume), SSD (slc) are 500€ per 64GB (slc type of course)

-Intel annouced his X25-E SSD (32GB first then 64GB) with WRITTEN specifications (250MB/s read, 170MB/s write, 0.08ms latency) for Q4-2008

-Any 1TB OLTP db intensively used (let's say "from 800 concurrent users") will gain from relying on 15x64GB instead of dozens of sas 15k array

The question is : For what situations and when does the SSD iops clearly allow benefits ?

Thanks for the feedback. I agree with all your points.

For the time being, only SLC based SSDs can be used in enterprise applications, largely due to the wear leveling issue and performance characteristics.

The performance improvement that SSD brings to OLTP type of applications is very tangible. The significant latency reduction can lead to a cascaded benign effect: lower latency --> reduction of parallel transactions waiting in queue --> reduction of DB lock contention --> reduction of transaction lock holding time --> reduction of transaction latency. Among many immediate benefit, a larger cluster can be built.

The reliability characteristics are less certian. After all, not many real OLTP applications have run on SSDs long enough for the any meanful data gathering.

The exact savings and cost justfications, of course, vary based on the actual applications and business configurations.

I am with a market research firm that performed a survey of users and makers of SSDs.  Our take-away was that there are indeed applications where SSDs can help, and these applications will see some replacement of HDDs with SSDs, but the impact on the HDD business will be low.  In other words, our findings agree with Jieming Zhu's.  Meanwhile, we found reason to expect the storage hierarchy to undergo significant changes over the next decade.  This is an interesting time in the market.  (see more at www.Objective-Analysis.com/Reports)

Hi Jim - thanks for the comment and nice to see that your research validates HP's approach.  There are a few - or I should say one - storage vendor that has been over-hyping solid state and then claiming that somehow HP is stalling.  Thanks again!