After moving to Japan I’ve been a big fan of Apple TV and catching up on US TV shows in HD on the thing.  I suppose that I’m still in the generation that likes to own my media and not hunt the Internet for free media.  Well as a result of this while the volume of my personal files has not grow significantly, the amount of capacity used, especially for HD quality episodes, has ballooned.  Basically I just had to relocate my TV shows into another iTunes library and a separate external Firewire attached HDD.  So I’ve been poking around on how people are managing their huge iTunes libraries, and mind you mine is not that big by comparison to say Will Friedwald, but there is not a uniform manner to do this.  Also there is not really a product from Apple or an Apple partner to do this.  Frankly, I think that Apple will soon be stumbling upon a problem of how to manage large media libraries in the home.  I know that Microsoft and others have tried to make Windows Media Center a smash hit and it has little adoption to speak of.  My own personal guess for this is that not enough people are experiencing the problem of large digital media libraries so well they did not need a product.  Well I’m at a point that I do and being the Storage Muse I’m going to put down some requirements and see what people think.

• Mirrored at the HW level with RAID-1
• Resident in iTunes as managed resource
• Definable sync policies between computers and various devices from a central iTunes
• The iTunes service has all the blasted metadata for what has been purchased, it should allow someone to restore from that metadata in the event of a serious problem or data loss event, in fact I would pay for this

I’m sure that there are more requirements, but I have a suggestion for Apple, uplift Time Capsule to be a media edition version or do something with AppleTV.  I personally don’t have time to spend cooking up a solution and would rather just buy something.  So if there are any recommendations or amendments to my rather short requirements list above please drop me a line.  Also if there is a recommendation for what to do let me know that as well.

Anyway, what I hypothesized in my last post was that the reason for the relative visual closeness of btrfs and xfs comes from more and more code being offloaded to the Linux kernel.  Well the most recent update on the xfs.org web site seems to suggest that is the case below is the relevant bit.

On June 9th we finally saw the release of Linux 2.6.30. For XFS this release mostly contains the improved ENOSPC handling, but also various smaller bugfixes and lots of cleanups. The code size of XFS decreased again by 500 lines of code in this release.

The Linux 2.6.31 merge opened in the mid of the month and some big XFS changes have been pushed: A removal of the quotaops infrastructure which simplifies the quota implementation, the switch from XFS’s own Posix ACL implementation to the generic one shared by various other filesystems which also supports in-memory caching of ACLs and another incremental refactoring of the sync code. …
Source: http://www.xfs.org/index.php/XFS_Status_Updates

While I still don’t have enough data to prove my hypothesis 100% I do feel that this is a step in the right direction of proving it to be the case.  If my hypothesis is indeed true then over time it will become easier and easier to create novel file systems for Linux.

Shout out to any Linux file system or kernel developers do you feel that it is easier to create a file system on Linux than other platforms?

]]> http://blogs.hds.com/michael/2009/07/xfs-alive-and-kicking.html/feed http://blogs.hds.com/michael/2009/06/hybrid-computing-in-the-news.html http://blogs.hds.com/michael/2009/06/hybrid-computing-in-the-news.html#comments Sun, 28 Jun 2009 08:30:20 +0000 Michael Hay http://blogs.hds.com/michael/?p=398 I ran across this today on Slashdot and Network World.

“These challenges cannot be pursed independently at the component level such as processor, memory and network switches; they must be addressed as an integrated solution. Co-design of the system hardware and software that is driven by processing requirements for selected application domains is essential. Solving individual challenges will not result in viable system solutions, DARPA stated.”

The article further goes on to state that the current trend of increasing the clock speeds and explicit parallelism controls by system programmers are not desirable in the required design.  To me this really means what, yes ladies and gentlemen hybrid computing systems.  In short the designers will have to look at everything CPUs, FPGAs, software, compilers, assemblers, IDEs, etc.

Here’s a link to the original request on DARPA in PDF form.

When we look at how HNAS can either tier internally or off to external systems using NFS this is a far more robust solution to the problem because there are fewer components.  Add in the Hitachi Data Discovery Suite and you get an advanced ILM offering that uses full content indexing, federated searching and distributed task management to fulfill either internal or external tiering.  Further I feel justified in saying that this is the right approach as all you have to do is look at the most successful approach to block virtualization which is within the box virtualization and data movement that the USP-V and the NetApp vSeries both employ.  We’ve shipped thousands of controllers that take this approach.  With our HNAS, we’ve intentionally copied a winning strategy, see the image below.

As to how we compare to the competition here see the image below comparing the two approaches.

Like I said you have to be kidding me EMC is basically using RAINFinity to emulate the copy command!

Update

Storagezilla Mark, posted an interesting comment on this post.  Since there might be some question as to where I got my facts from, here is the location which is publicly available on EMC’s site:  http://www.emc.com/collateral/hardware/specification-sheet/h1685-celerra-filemover.pdf.  Note I would post a screenshot from the materials, but I don’t want t violate EMC copyright.  That’s why I drew the picture above and added to the post.  I’ve asked Storagezilla Mark for a publicly available document, a tcpdump or other factual evidence that proves his point and at that point I’d be most happy to make a correction, etc.  Until then my logic stands!

Also he inserted an implied question on BlueArc and I answered that in the comments section.

I’ve been intentionally planting seeds across several posts for sometime now.  Specifically, I’ve been trying to set the stage about pointing out the relatively static nature of general purpose processors versus the dynamic nature of hardware accelerated software on FPGAs.  I also know that by talking about this in this way, I’m pushing ahead of where customers are thinking.  It is my opinion that the architectures I’ve been describing set the stage for building in real automation that the body of enterprise users have been questing for.  As I’ve also lived through and run design sessions with users and I want to relate a point: lots of automation is requested, but when asked about implementation time frames and if they trust the automation, there is well hesitation on both fronts.  Users like automation, but couched as tangible usability improvements or on a roadmap, but the message is clear: users always want to be in control while reaping the benefits from the force multiplication effect — one person can do the job of two to three people. What I’m going to talk about is meant to set the stage for creating infrastructures and architectures can can provide real automation not hyper usability, however, I fully recognize that this is not something that will not happen in short order.

For many years, since my university days really, I’ve been a fan of borrowing themes from the natural world for the realm of computer science and engineering.  One constant phenomena available in the natural world is the dance between the environment and the organisms which live in concert with the environment.  Basically as one changes, the other changes in response, forcing a change in the originator of the first change, and the cycle continues on and on. It is like dancing where the leader and the follower move synchronously with one another making something of beauty.  Because of the relatively slow rate of change in general purpose processors we will consider them as the environment while software in a computer system we will think of as the organism.  Today the organism, software, is forced into an unnatural rate of change and further the software cannot see, at least in its short execution time, the environment/hardware change in response its activities.  However through raising things like interrupts the hardware can lead the software to perform certain activities or trigger call-backs, etc.  I guess this is more like dancing whilst standing on top of someone’s feet, in that the software really doesn’t have the freedom or flexibility to push back as a peer partner in the dance. What I am getting at is that there is a key missing piece in most modern systems allowing the software to exist as a full partner to the hardware changing the behavior or configuration of the hardware and participating fully in being changed by the hardware.  In short the dance!

Again, hybrid computing devices can answer the call, and yes I expect my esteemed colleagues to chafe under this point.  Hybrid computing systems offer fixed computing elements (think ASICs and General Purpose CPUs) and flexible computing elements (think FPGAs).  With the combination it is possible to contemplate software that can react to the hardware and push on the hardware to change and optimize configurations for the task at hand.  For those that understand neural networks which evolve the best approach to the problem being worked, augmenting static hardware with FPGAs would be a natural extension of the neural network approach, but the system should, in theory, be able to achieve improved speed and I would argue emergent properties. In this way the software is dancing with the hardware and each is changing to meet the other. With the notion of neural networks there is the human brain and all and within the human brain there are elements like the medulla oblongata which is basically hard wired to handle autonomic functions and is much like an ASIC in that it cannot change configuration and it is absolutely required for life to function.  However the cerebral cortex is literally changing its physical nature as we learn so again the natural world provides an example of what is possible.  As I talked about in a previous post, some of the skills that can be brought into muscle memory like typing get stuck into deep parts of our brain so that we don’t have to think about actually doing them.  I imagine that this model could be the source of the automation perhaps learned by traditional software through neural network techniques and then offloaded to FPGAs and hardware accelerated software which can emerge new autonomic behaviors, and voila. (Yeah I know I really need that PowerPoint compiler right about now, this is the automation I was talking about.)

In the long term, like say in the next 10 years or more, I believe that in the storage and processing space we will see the continued adoption of hybrid computing architectures.  Further I would expect to see software that can best take advantage of these hybrid systems.  The combination of the software and hardware will evolve and dance together so that it can solve key problems better than merely a static hardware software combination alone.  The system will learn to solve problems offload key learned algorithms, accelerate application I/O, figure out the right placement of LBAs, etc.

By the way the selection of Hitachi’s technologies already are hybrid in nature.  We include FPGAs, general purpose CPUs, and ASICs in our systems already.  Again we do a lot of things under the covers that I cannot discuss on this forum, but let’s just say that we are advancing the cause quietly and carefully.

P.S.  I had originally intended to span this over several posts, but as I kept going preserving the integrity of the concept and line of thinking is key.  So hopefully you Mr. Reader have hung in until the end.

P.P.S.  Here is a collection of companies, groups and consortiums who are also building hybrid computing systems with FPGAs.  I reference them because I want people to understand that this is not just unique to Hitachi it is a phenomenon, enjoy.

1. XtremeData - … XtremeData’s key IP comes in the form of innovative hardware components (FPGA-based In-Socket Accelerators™) coupled to an extensive software and firmware stack.  … In 2005, strong market demand steered XtremeData into productizing and selling the FPGA In-Socket Accelerators™ – originally intended only for in-house use. …  Our intent is to continue on this path: enabling our customers in financial services, video imaging, military, medical, and bioinformatics to create their own hardware-accelerated appliances.
2. Netezza - Today, Netezza is the data warehouse appliance leader, combining storage, processing, database and analytics into a single system that delivers 10-100x the performance, at half the cost and one-third the power of other approaches. Think of it as a Ferrari, with the price and efficiency of an economy car.
3. Exegy - The hardware-acceleration technology behind the Exegy Ticker Plant provides a competitive edge to the high frequency, colocated trader who requires a ticker plant …
4. Nallatech - For over 15 years, Nallatech has been at the forefront of technological and market innovation in FPGA computing in the Embedded Defense and High Performance Computing markets.
5. FHPCA - The FPGA High Performance Computing Alliance (FHPCA) is developing high-performance computing solutions using Field Programmable Gate Arrays(FPGAs) to deliver new levels of performance into the technical computing market.
6. Altera’s Consortium for FPGA based HPC - Altera Corporation (Nasdaq: ALTR) today announced the development of a new university program to support academic research into high-performance computing. AMD, Sun Microsystems and XtremeData are participating in the program that will donate $1 million in workstations and development software to universities. Using the workstations, participating universities will be able to research and drive the adoption of FPGA co-processing for high-performance computing applications such as medical imaging, data analytics, text searches, network security, bioinformatics and energy.

He has several posts on the topic in general:

The most recent is 2.012
Next is here 2.011
Finally here 2.010

Well this is not getting under my skin personally at all.  In fact I’m quite flattered as EMC again and again chooses to copy Hitachi.  In the most recent article Barry is also asserting that we are censoring his comments and I can say that is far from the truth.  HDS doesn’t do this, we only censor SPAM which leads to nasty web sites and malware, which I’m guessing Barry’s site does as well.  Further I am glad that Barry spends serious quality time with customers I do to in fact just went to see on this last week in Japan.

The last thing that I want to point out is that I’ve been offering Barry a small olive twig. Basically my recent thesis is that all systems are now hybrid computing devices consisting of ASICs, general purpose CPUs, and potentially FPGAs.  I think that V-Max is this very kind of system, even EMC admits to creating a link ASIC for node communications along with their usage of Intel, etc.  I think that this is something which is largely overlooked in computing today and it is so important that I’ve been dedicating posts to it.  I have one more coming out soon.

Michael

P.S.  Barry the recent usage or hinting at derogatory four and three letter words is not doing your arguments justice.  I urge you to please refrain from using them, I’ll commit to not using them if you will.

P.S.  Barry and I worked this out on his blog. We will continue to joust much like Linus and Tanenbaum: http://www.dina.dk/~abraham/Linus_vs_Tanenbaum.html.

Update:

Well according to CNET it is confirmed that Steve had the transplant and has returned to work already.  Again good luck in the recovery.

The images above are taken directly from the author’s, Răzvan Musăloiu-E, site that takes us on this journey into the Linux file system internals.  His exploration is done by looking at the number of external symbols per file system type.  I think that there are a number of interesting points from the analysis. One is that the fewer the symbols the more self contained the file system is.  Further the more modern file systems like btrfs, ext4 and xfs all have many external symbols. Perhaps this means that more and more of the code needed to generate a file system is making its way into the kernel and associated libraries resulting in it being easier to create new file systems on the Linux platform?  Finally if you look at some of the visuals you will find that btrfs and xfs are often in close proximity to one another, I’m not sure what this means, but it is interesting none the less.

Update:

I just got permission to use some of Răzvan images and conversed with the author of the study via email, I do want to point out the following point:

Me: … the increase of dependence on external symbols is there any conclusion we can draw from this?  In particular I’m wondering if more functions have been finding their way into the Linux kernel making it easier to create file systems on Linux than other platforms.

Răzvan: If things would be really easy then I would expect the interactions between the file systems and kernel to be limited in number and see that most of the system are using most of them. This is not the case though. Just by looking from outside we can see that there are quite a lot of them. There is the chance that these big number is also inflated by inlining.

Thanks for the permission and discourse Răzvan!