Gregs Server and StorageIOblog - Latest Comments

Re: What is the best kind of IO? The one you do not have to do

Greg — Sun, 06 May 2012 11:40:19 -0000

Thanks David for your perspectives and comments.

When you have seen or experineced something in the past, it can lead too Dejavu when seeing something similar or packaged differently. Likewise if you have not seen or experirenced something similar, it may appear as new and revolutionary.
Thus is often the case with technologies that evolve improving on what previous generations or their predecessors did on a proprietary or some other basis.

Thus as you point out, things were done on the IBM mainframes, as well as mid-range systems along with virtual memory extensions, paging, ramddisks, and many other tricks on those as well as other platforms ranging from PDP and VAX minis among others and later NT and Unix or Linux based systems.

It is great to see ideas or techniques, dusted off, improved, updated with new variations or generations of packaging, components, and interfaces and finding their way into products and services across the board.

Here is to having more DejaVu experiences for those who have been around for a while and revolutionary eye opening experiences for others ;)

Cheers gs

Re: What is the best kind of IO? The one you do not have to do

David Sacks — Sun, 06 May 2012 10:23:22 -0000

As one who has been in IT for a number of decades, I’m always fascinated when what is old is not. What do I mean? Many computer principles have been around for a long time, yet continue to be important even as technology evolves. “The best I/O is no I/O” is one of those.

I remember hearing that phrase in IBM at least as far back as the ‘80s. In the late ‘80s mainframe memory sizes were increasing and IBM did a lot of good work citing the benefits of using main memory to eliminate I/Os: mainly faster application response time and reduced processor utilization which could in turn lead to longer useful processor life. This was all neatly packaged under the theme “Data in Memory”.

A special form of solid state mainframe memory called Expanded Storage was developed and offered as an option to complement main memory. Expanded Storage was not as fast as main memory but was less expensive and much faster than disk storage. (Sound familiar?) A major use of Expanded Storage was as a fast paging “device”. Programming interfaces were developed so the OS and users could more effectively use main memory and Expanded Storage to get data-in-memory benefits for other data types as well.

Today, the industry is rightly excited by flash memory. Considering Data in Memory and Expanded Storage, it appears the basic principles of olden days still apply.

David Sacks, storage analyst, IBM

Re: Poll: EMC and Cisco Acadia VCE, what does it mean?

greg schulz — Tue, 17 Apr 2012 10:32:35 -0000

Note that Acadia has been renamed to VCE.

In addition to various Vblock models available from EMC and VCE, other vendors have announced or shipped their versions of converged solutions for use in legacy as well as public and private cloud deployments. These include NetApp FlexPod (also powered by Cisco UCS), HP Converged, IBM PureSystems and Oracle Exalogic among others. Note that HDS which exited the server market many years ago in the U.S. is reappearing with Intel based servers as well.

Here is some additional information about what IBM recently announced:
Part I: PureSystems, something old, something new, something from big blue
http://storageioblog.com/?p=28...

Cheers gs

Re: Unified storage systems showdown: NetApp FAS vs. EMC VNX

Greg — Mon, 16 Apr 2012 12:54:09 -0000

Thanks for the comments Lee and good to hear from starboard (aka formerly known as Reldata), hope all is well.

Cheers gs

Re: Unified storage systems showdown: NetApp FAS vs. EMC VNX

Lee Johns — Mon, 16 Apr 2012 11:45:40 -0000

Greg. Nice article. particularly the comments on "buyer beware" when it comes to SW and hardware add-ons.

I think their are a few more elements to look into with "unified" Storage.

1) What is the platform footprint. Many of the solutions you mention are in fact gateways in front of block solutions with a somewhat common management interface. While this is functional for small medium enterprises it is not optimal. An EMC VNX for instance requires 5 seperately powered boxes for a fully HA single solution. An Array, 2 Gateways and 2 management stations. Yes it is sold as a solution but apart from the pricing for addons the power requirements are going to bite you.

2) Are the platforms really sharing the resources. Just because some companies have a SAN and a NAS platform does not mean they have a solution. To my knowledge HP does not have a FC, iSCSI and NAS platform that also has a common management interface as a solution and yet you mention them in this piece. What is the criteria to be unified?

3) Do they really support the generalist in running Mixed Workloads? I do not know too many customers using todaay's established players who do not have Storage specialists to run their systems.

Full disclosure I am with Starboard Storage Systems www.starboardstorage.com

Lee Johns

Re: Optimize Data Storage for Performance and Capacity Efficiency

balesio AG — Mon, 02 Apr 2012 14:34:40 -0000

Reducing your data footprint from the start is key to the most efficient data center. All advantages you will see from Native Format Optimization (NFO) will be passed on to other tiers along your data lifecycle:
http://storagegaga.com/nfo-for...

If you really want to do something about data storage growth, then don't use patches like backup dedupe, tiering, etc. before you haven't optimized the data footprint of your files in the first place.

Re: Is 14.4TBytes of data storage for $52,503 a good deal? It depends!

Brian McCarthy — Fri, 23 Mar 2012 08:06:13 -0000

Greg all good questions. Having been in the VAR business for 9 years and with 50% of our sales in the Public sector, I find "most" go right to price and damn the TCO and ROI it's all about price, be it Higher Ed or K-12. With that said most find themselves with non-branded or trailing edge technology or beta release products from startups that never made it.

The HP rep use to laugh and tell me how this one County School on Florida's west coast was "the dumping ground for all his of discontinued stock" As for the man-weeks of added management and labor for the non-feature rich SANs purchased on price, as one CIO of a local State College once said "that's what we have staff for, I don't care how long it takes them". As for the Cloud well it's too soon to tell other then USC moving 8 PBs to the Nirvanix Cloud, but my experience says it will be one of the Clouds top verticals with cloud storage street prices running sub-$.10 / GB.

Re: Is 14.4TBytes of data storage for $52,503 a good deal? It depends!

balesio AG — Thu, 22 Mar 2012 16:45:48 -0000

Greg,
Very true. Cost per storage capacity and even the concept of Total Cost of Ownership (TCO) of storage are insufficient because there are many externalities wher you can't possibly put a price tag on. Example: What is all storage capacity worth if you can't get the information (=file) from A to B because you sacrificed investments in Network for your big storage box?
IT Admins who think about this will realize that it all comes down to only one decisive cost driver: information itself (often wrapped in a file). So only if you start with the single files and perform storage optimization there, e.g. via Native Format Optimization (NFO), you can be sure that you maximize storage utilization all the way down to the archive. If you then still need 15 TB as a school, you really need it, but it is more likely that they would have then a concrete need of only half of it, like this NJ school district:
http://balesio.com/pdf/caseStu...

Chris Schmid, COO balesio AG

Re: Is 14.4TBytes of data storage for $52,503 a good deal? It depends!

balesio AG — Thu, 22 Mar 2012 16:36:15 -0000

Greg,
Very true, lowest cost per capacity is insufficient, even Total Cost of Ownership (TCO) is an insufficient concept as there are so many externalities where nobody can put a price tag on. Example: What does it matter to have the highest capacity if you can't get the information (=file) from A to B because you saved on network/bandwidth for the sake of having more storage capacity?
IT admins who think about the whole concept will often find one single decisive cost driver: information itself, wrapped in a file. Storage technologies such as Native Format Optimization (NFO) which work on this cost driver will result in smarter storage utilization all the way down to the archive.

Chris Schmid, COO balesio AG (cs@balesio.com)

Re: Why SSD based arrays and storage appliances can be a good idea (Part I)

Greg Schulz — Wed, 14 Mar 2012 23:52:02 -0000

Roy you bring up some good points.

Certainly, with SSD in drive form factors, the drive itself
is a field replaceable unit (FRU) and thus when there is a failure, just like
with a HDD; the entire drive would have to be replaced.

However like a HDD, when it fails, and assuming that the SSD
drive is mirrored or in some other way protected as with an HDD, then the availability
should be no less, and if the appropriate technique used, performance does not
have to be impacted either.

Of course, as with HDDs, how the devices are configured,
raid or protection and other options will have an impact on performance, availability
and capacity.

Thus a LUN does not have to be impacted depending on how
configured.

There is a notion that drive form factor SSDs are more
likely to fail which Im not convinced of yet given they use the same or similar
nand flash dies, and in some cases FTLs.

Granted, the type of SSD drive, SLC vs. eMLC vs. MLC along
with storage system flash management, wear leveling and other firmware
enhancements to a controller can make a difference.

On the other hand, give the large population base of drive
form factor SSDs installed, including some that have been deployed for many
years, it would not be surprised that the sample size is larger thus more
drives to fail, however that should also be looked at in terms of the total
population.

PCIe nand flash cards are a bit newer, thus their population
or sample size are smaller and thus there should be fewer failures, however
again, there should not be a higher percentage of failures for the population.

Likewise if a PCIe nand flash card fails, depending on the configuration,
it too can be impacted by a failure, and like drive form factor SSDs, can also
be protected or impacts minimized with replication and other techniques.

Not to mention the card would have to be replaced.

As you point out, a component in an appliance or storage
system could fail, how that system or appliance can handle the fault and
support replacement should be considered.

In other words, it all depends particularly if apples to
oranges vs. apples to apples comparisons.

Thanks again for your comments and perspectives.

Cheers

Re: Why SSD based arrays and storage appliances can be a good idea (Part I)

roy — Wed, 14 Mar 2012 14:16:34 -0000

Greg, One "form factor" concern expressed by storage administrators in the HDD world is the size of the drive. A drive is a minimum unit of replacement in the field. When a drive fails, there is a cost that is incurred. This cost includes the fact that the storage device (drive, LUN) continues to run with a lower level of protection and the required operations to bring the storage device back to the regular level of protection. The prevailing notion was that drives that are larger than 1TB can be problematic given the fact that the larger the minimum unit that needs replacement the greater the cost to bring the storage service back to guaranteed levels.

Appliances that go with a DIMM/PCIe approach are now available with 100TB capacities. And while they have the advantage of smaller form factors, the negative is that should a single Flash Card need to be replaced, the entire unit needs to be taken offline. The loss and recovery thereof tied to 100TB of data is a lot more expensive than 1TB and hence can become a significant negative when one factors in failure scenarios.

Appliance vendors argue that the actual failure rate of Flash is a lot lower than HDD and so the down time is a lot less than legacy environment. There could be some truth to this argument. Even so, given that existing perceptions of legacy environments are the reality that we have to contend with, it will take a few years before monolithic appliances that enclose 10s of TB of DIMM/PCIe based Flash cards can be viewed as Tier-1 storage.

Any comments that are contrary to the above will be appreciated.

Re: Are social media and networking a waste of time?

Greg — Sun, 11 Mar 2012 11:46:14 -0000

Thanks Gordon for your comments and perspectives, sounds like you get it.

Cheers gs

Re: Are social media and networking a waste of time?

Gordon Jablonski — Sat, 10 Mar 2012 22:49:02 -0000

Hi Greg,

Your ask the all important question, is it all worthwhile. With the changes now days to social media, it is vitally important that you create a social profile. This is done through interacting with people from all walks of life - in the areas you are interested in - and commenting sensibly to create a good connection with the person who wrote the article.
I belong to Xeeme, as I had so many different social accounts (and growing) that I just couldn't keep a check on them all. It's a free service and helps you keep your social networking in line.
The information you learn from these sites is priceless and keeps you up to date with the latest on the internet, and in business in general. A Big YES - social media is necessary...ask Google!!!

Re: AWS (Amazon) storage gateway, first, second and third impressions

John Mann — Sat, 25 Feb 2012 10:07:12 -0000

Thanks for the post and information! I think education is
important for us so we must prepare the best education for our generation by
sharing such great information with each other!

free
cell phone

Re: Researchers and marketers dont agree on future of nand flash SSD

OzGuvnor — Mon, 20 Feb 2012 16:03:16 -0000

There is a definite need for Tier 0 in todays enterprise arrays in order to offset the ever decreasing IOPs/TB you get per spindle. To all intensive purposes it doesnt matter to the end-user who owns the storage service what goes on inside the SSD box, so long as it demonstrably achieves the service levels we need to provide to our IT customers. If the current generate of tech declines in reliability too much as it shrinks then there will probably be a lag of a few years until the next generation of tech is market ready. Meanwhile business goes on.

Re: Researchers and marketers dont agree on future of nand flash SSD

Greg — Mon, 20 Feb 2012 15:54:22 -0000

Thanks Alex for your good perspectives, always great hearing from "experts" ;)...

A theme you touch on that I find interesting is that basic processing and comonents need to evolve, that includes starting on the motherboards of servers being used for both compute and storage along with how PCIe is being used. Both take time to evolve, then ramp up volume to where pricesses become affordable, in the meantime, workarounds will exist as will oppourtunities to close those gaps with software, drivers, shims, plugins, card modules, adapters, drives, systems etc...

Hope all is well.

Cheers gs

Re: Researchers and marketers dont agree on future of nand flash SSD

AlexSons_NL — Mon, 20 Feb 2012 15:07:16 -0000

As always with a new technology it is first hyped, hyped bigger, hyped to the hype whereafter expectations will settle rapidly at much lower levels. Gartner built a life on it and called it the Hype Cycle (http://www.gartner.com/technol....

For me is the current use for SSDs falls apart in two methods:
1. used as near to the application as possible, on this method FusionIO and some others thrive
2. used as Tier 0 in storage systems with the NetApp solution (SSD as cache extension) as an odd but viable variant on this method

In my opinion the first method makes best use of SSD characteristics but is somewhat limited of use. For example VDI can greatly benefit, or special purpose environments like Facebook or Google.

The second method will show way less performance gains at -current- skyhigh prices. "Huh" I expect you to say? Well, for general purpose storage systems to benefit from SSD technology they need a thorough revamp of its processing power and (PCI) bus technology to be able to make the best use of SSD or even better NAND. Such an technology overhaul takes time: as a proof of this, you still won't find a major vendors storage system in the marketplace which is completely fit for SSD usage. Sure, some disk bays may be used for SSD, but you can't add a lot of them before buses or CPU/ASICs become saturated.

(I've read numerous articles on SSDs or related. I've tested myself SSDs in a Hitachi VSP and found the performance gains a bit disappointing and their cost disastrous. And I have an OCZ Vertex3 SSD in my own PC so I am an experience expert too ;)

Re: Researchers and marketers dont agree on future of nand flash SSD

Greg — Mon, 20 Feb 2012 14:39:13 -0000

For those who are interested, here is a link to a
related discussion-taking place over on the LinkedIn SSD forum:

http://www.linkedin.com/groupI...

Re: Researchers and marketers dont agree on future of nand flash SSD

Schistad — Mon, 20 Feb 2012 08:31:08 -0000

Actually, this just further pounds home the fact that tiered storage is an absolute necessity.

What this report actually says is that you have to choose capacity or performance, but cannot have both (for a given sum of money). This creates an incentive to store your data ever more intelligently, and only pay for performance where it is needed.

And of course, the article is exclusively focused on NAND flash, which happens to be the oldest and hence most mature technology but possibly not the best long term.

Re: Researchers and marketers dont agree on future of nand flash SSD

fletch00 — Mon, 20 Feb 2012 02:59:00 -0000

"While the growing capacity of SSDs and high
IOP rates will make them attractive in many applications,
the reduction in performance that is necessary to increase
capacity while keeping costs in check may make it dif-
ficult for SSDs to scale as a viable technology for some
applications." - some have argued (on /.) this is negating probable advances in tech to conteract the capacity/performance effect - the study assumes no advances in 10+ years?

Re: Speaking of speeding up business with SSD storage

Greg — Sat, 11 Feb 2012 09:27:21 -0000

Here is a link to the first in a series of guest posts I am doing about "spinning up to speed on SSD" and associated topics appearing over at "The Virtualization Practice"

SSD options for Virtual (and Physical) Environments: Part I Spinning up to speed on SSDhttp://www.virtualizationpr...

Cheers gs

Re: IT and storage economics 101, supply and demand

Greg — Sat, 11 Feb 2012 09:26:56 -0000

Re: EMC VFCache respinning SSD and intelligent caching (Part I)

Greg — Sat, 11 Feb 2012 09:26:09 -0000

Re: EMC VFCache respinning SSD and intelligent caching (Part II)

Greg — Sat, 11 Feb 2012 09:25:53 -0000

Here is a link to the first in a series of guest posts I am doing about "spinning up to speed on SSD" and associated topics appearing over at "The Virtualization Practice"

SSD options for Virtual (and Physical) Environments: Part I Spinning up to speed on SSDhttp://www.virtualizationpr...

Cheers gs

Re: AWS (Amazon) storage gateway, first, second and third impressions

Greg — Fri, 10 Feb 2012 12:34:58 -0000

Good to hear from Twinstrata and thanks for the comments Andy.

Hope all is well.Cheers gs