Strategy

- (editor's note - this post also comes from Greg White in storage marketing. Thanks again Greg)

Data tiering – classifying and segregating data on different types of disk drives – has been around for awhile, but has been somewhat limited by drive choices. Support for additional drive types on new disk arrays, like the Dell/EMC CX4 series, could make this more relevant for a much broader audience.

Initially, you basically had two choices - Fibre Channel (FC) or SATA. Over the last year or two, SAS drives have become alternatives for FC drives in some storage systems, as SAS drives offer 15K and 10K RPM options at typically lower cost due to higher industry volumes.

Today, new drive options, like Solid State (Flash) and Low Power SATA drives, are increasing the range and spread of potential data tiers. For example, flash drives can offer up to 30 times the performance of traditional 15K FC drives in some applications. Today, these drives come at a premium to FC drives; however, over time as prices decline, more and more users may use these drives for their most performance hungry applications.

Low power SATA is at the other end of the spectrum. These drives can consume almost a third less power than traditional SATA drives, making them ideal for storing data for long term retention where performance is not critical.

This raises a few questions for discussion:

1. Do these new options - and the increasing spread between choices - make data tiering more attractive or simply muddy the waters?

2. What are the implications for managing your data, storage systems, applications, as well as other considerations, like managing your spares for example?

3. How do you classify data and implement tiering on these new drive options?

Today’s storage systems that can support flash, FC, SAS, SATA and low power SATA drives as well as virtual provisioning are the future of storage. While tools that ease the process of tiering are increasing in number and maturing, there are opportunities to leverage these new drive options to start increasing storage efficiency and resource management through tiering efforts.

- (editor note - this post comes to us from Greg White in storage marketing. Thanks Greg)

Many organizations have been looking to desktops, notebooks and servers for power and cooling savings and the associated environmental benefits of green IT solutions. What about storage?

Up to this point, storage has lagged far behind because the lowest hanging fruit has been in other areas. Server virtualization and system management have been able to provide large savings in power and cooling for organizations of all sizes.

Now, storage is beginning to take center stage in this debate. Green storage opportunities fall into three key areas: components, platforms and methodology.

New storage arrays are starting to take some of the learnings from the server guys and putting in more efficient power supplies, variable speed fans and more efficient disk drives.

In addition, the platforms are doing more to address efficiency, like utilizing virtual provisioning and de-dupe technologies to increase resource utilization and reduce the number of drives that need power – the biggest source of most storage arrays power consumption.

Lastly, organizations are looking more at how, where and for how long they store data. While tape may be cumbersome to recover from in a short period of time, it is still much more power efficient than storing data on hard drives, and where disk storage is desired, there are new options like low power SATA drives that can be deployed.

Are these new improvements in storage leading you to start factoring storage into your power and cooling planning? Is green storage a key consideration, something that is nice to have, or not even on your radar now?

Dell's FlexAddress announcement on Tuesday drew this response by Gary Thome of HP.  So what's a poor system admin to do if they want to manage SAN and LAN addresses for their blade systems?  I'd suggest they let their accountants make the decision for them. 

Here's the deal.  HP apparently had this elegant old technology from Tandem that used specialized switches to map network IDs to hot-swappable compute modules. They re-introduced this technology as Virtual Connect for their blade servers because they figured it solved the problem of maintaining consistent network IDs when swapping out compute blades. In fact, Virtual Connect works pretty well - all you need to do is buy fault-tolerant pairs of specialized chassis-insertable switches to provide the ID mappings.  That doesn't really force customers to change their network infrastructures, as some would suggest, but it does add a lot of cost to the blade solution.

Over the years Dell has been criticized for copying the technologies created by other companies.  Brad Anderson, Dell's Senior VP of Business Products said as much last November after Dell announced its acquisition of EqualLogic:

“It’s fair historically, if you go back a year or so, that you would have thought of Dell as more of a fast follower,” said Brad Anderson, senior vice president of Dell’s business product group. “But we recognize that if we want to solve our customers’ problems and provide leadership, we’re going to need to jump in very early.”  “With Michael coming back, things are a little bit different,” Anderson, said. “Company-wide, we are driving IT simplification very hard. It is our strategy; it is our mission. It’s also driving a cultural change within Dell.”

In other words, when we see a good idea now, we don't just try to imitate it, but try to figure out how to make it as simple to use and efficient as possible.  And that's what FlexAddress is all about.  HP had done a pretty good job providing a solution, but it was originally designed for mainframe environments and not today's X86-based blade servers.  As a result, Virtual Connect is over-engineered and expensive - there is no need for the specialized switches it uses.

Dell engineers figured out a much simpler way to provide persistent network IDs for blades as they are swapped or upgraded.  The network IDs for each blade are mapped to another set of network IDs in the chassis according to slot the blade is inserted into. In other words, when a blade is removed and another blade is inserted in the same slot, it automatically is mapped to the same chassis-based network IDs - which means all LAN and SAN network functions work exactly the same as before. That includes VLANs and zoning and anything else you can think of.  HP would have you think that all this has to be redone with FlexAddress, but it doesn't.

Maybe there are a few corner case advantages for Virtual Connect, such as lab environments where changes to blade configurations are more random and have less serious consequences. But in production environments where the rule of "if it ain't broke don't fix it" applies, the types of changes made to blade modules tends to be one to one replacements and upgrades.  FlexAddress is perfect for this. 

And one other thing.  HP says that Virtual Connect was designed to enable the system admin to function independently of the network and SAN people.  Oh really?  And how are the proper tie-ins made from the Virtual Connect switching modules to the LAN and SAN switches anyway  - by divination?  C'mon guys - there is no free lunch. 

So let's see now.... should I buy one SD card or a pair of expensive switches that cost approximately 10X as much?  Hmmmm.....

Did you know that The Wharton School at the Univ of Pennsylvania was the world's first collegiate business school - founded in 1881? I didn't - not the first time that's happened.

They've just opened a new data center - replacing IBM BladeCenter gear with PowerEdge M600 blades and Dell EqualLogic PS5000X storage arrays.

Their goals were to run a Linux cluster (for research) with the blades and expand mailbox quotas for students and faculty - from 300MB to 1GB.

Interesting notes:

• Wharton Computing IT professionals were able to unload and set up one of the Dell M1000e blade enclosures in four hours, compared with a 24-hour process using the previous hardware. Wharton's Joe Cruz blogged about it.  "the new hotness" - love that.
• The Dell enclosure was delivered in one box, replacing technology that previously arrived in about 60 boxes, a reduction of approximately 45 cubic feet of packaging material. With that in mind - check out this video.
• They've also implemented PowerEdge R900 servers that are providing a small hardware footprint and intended to deliver stronger performance than the previous generation of servers, with the goal of quintupling the computing power per node in Wharton’s Microsoft® Exchange 2007 environment.

According to Dan Alig, senior IT director of Wharton Computing. “Dell blade servers and Dell EqualLogic storage will allow us to do more with less and manage our budget as effectively as we manage our technology resources.”

That's good business. And I don't even have an MBA - yet...

There's a lot of talk about blades out there - especially by Dell and HP. I’ll write more soon about the affordability of Dell’s M-Series vs. the HP Shorty. In the meantime, I think you’ll enjoy this quick :40 second video which takes a look at how the two systems stack up … literally.

I’ve always found it interesting that one of the main things you hear from HP about Dell’s M-Series blade server solution is that it is our “third venture” into the blade market. This is true, as is the fact that HP has introduced four different blade models into the market (PowerBar – HP; E-class; P-class; C-class). Quite the omission on their “Real Story about Dell and Blades” page. What do they have against innovation?

Blade technology has evolved significantly over the last 5 years and Dell has not just evolved with it, but we’ve anticipated it with our M-Series blades. Planned chassis evolution, which is what Dell has done, is a good thing. It means you are keeping up with or anticipating future technology needs.

My newest favorite comes from HP’s “Shorty”, the c3000 and the marketing around it being “…built just for small sites with big compute and storage needs…” (The new BladeSystem c3000). I think this claim begs the following points or questions:

1. If you have “big compute needs,” wouldn’t you need the basics in enterprise class capabilities like redundant network fabrics…or is ensuring you can access your data just not important?

2. The “plug it into a 120v standard wall socket and be up and running” attitude borders on criminally funny. Most standard wall sockets are 15amp, but I’ll give HP the 20amp benefit of the doubt. You can run four standard BL460c blades in the chassis with that, perhaps five. Six blades at 60% capacity with two Intel L5420 procs and 8GB of RAM is right at the de-rated, best practice, limit on a 20amp circuit with 15.79 amps. A power spike to 100% will blow the circuit. Shorty doesn’t have redundant power so if the circuit goes, the chassis goes. Best practice for Shorty should be to only use half the chassis max (4 blades), and that’s with low voltage processors with a maximum 8GB of RAM each. Not the best RAM capacity for virtualization.

3. A better compare for customers is the M-Series configuration below using 208v, which only draws about 18 amps on a SPEC benchmark load. (A 208v window air conditioner is normally on a 30amp circuit.) You can easily put in 4 times the servers (16), each with twice the RAM of the Shorty config above (16GB), using a little more than twice the total power of the Shorty config. If you drop the blade count to parity, you are a solid winner with more RAM, more I/O and true enterprise class features.

4. M1000e configuration:

• redundant Chassis Management Controllers;
• the Avocent iKVM;
• 6 switch modules (3 redundant fabrics);
• 9 fans;
• 6 power supples; and,
• 16 M600 blades, each with two L5420 Intel procs and 16GB of ram, two LOMS and two dual port mezzanine I/O cards (six total I/O ports in three redundant pairs)