« The Excitement Builds: EMC World 2012 Preview | Main | VNX: Charting The Changes »

May 11, 2012

TrackBack

TrackBack URL for this entry:
http://www.typepad.com/services/trackback/6a00d83451be8f69e20163057933e8970d

Listed below are links to weblogs that reference VMAX: Charting The Changes:

Comments

David Sacks

Hello, Chuck.

(Disclosure: I work for IBM. These comments are mine.)

I'm always pleased to see storage industry announcements that reflect how active the industry is and which foster healthy competition.

However, I was disappointed to see the slide at the bottom of your blog comparing DS8800 to Symmetrix VMAX 20K and 40K includes misleading data. Here are some items in the table that caught my attention.

1) The slide claims that maximum "cache memory" is 384GB for DS8800, 1TB for VMAX 20K, and 2TB for VMAX 40K. Here are just two ways this is misleading. First, these capacities are for processor memory, only a portion of which in all these systems is used for cache unless Symmetrix design has changed to put its system data somewhere else. Second, as acknowledged in EMC's press release for the new VMAX models, only half of VMAX "DRAM" (which I am assuming is referring to the data cache) is "usable" (EMC's term) because Symmetrix VMAX mirrors all data in cache. In contrast, DS8800, like most disk systems, mirrors only writes in cache, not reads, because there is generally no benefit to mirroring reads since read data is accessible on RAID-protected disk even if there is a cache failure. Considering the slide has a row that specifically refers to "Useable [drive] capacity" for data, it seems appropriate that the cache row should refer to usable capacity as well.

2) The slide states that DS8000 Easy Tier "Movement frequency" is "24 hours". That is inaccurate. Easy Tier moves data using various durations appropriate to the reason for the migration. While some migrations are done after 24 hours of analysis - a time period carefully chosen based on a study of customer data access patterns - Easy Tier performs other migration operations (called rebalancing, warm demotion, and cold demotion) continuously and over significantly shorter periods of time than 24 hours.

3) The slide gives VMAX credit for supporting "heterogeneous array replication" with "Yes, with RecoverPoint integration." But the slide does not give credit to DS8800 which can support heterogeneous array replication using the IBM SAN Volume Controller (SVC). Moreover, this SVC function does not need special integrated support in the disk system, unlike the write splitter in Symmetrix for RecoverPoint.

The table row about "relative cache miss performance" also caught my attention, but it had no explanation of what was behind the numbers so I am ignoring that in this comment.

David Sacks, IBM Senior Storage Consultant

Chuck Hollis

Hi David

Thank you for your comment and your disclosure.

1) I agree that a practice of claiming usable capacity for storage cache would be a good idea. Difficult for two reasons: one being precedent, the other being variability depending on how the storage device is used. But not impossible. Your understanding of VMAX caching is incorrect, I believe.

2) I believe the widely-used default for the DS8800's cycle of "analyze-and-move" is 24 hours. If there is some other widely-used default, please let me know. Of course, any default can be overridden if the customer makes the effort and assumes the risk.

3) The slide is incorrect I believe in that the VMAX does heterogenous replication via the new FTS (federated tiered storage) feature included at no charge. The VMAX also supports the RecoverPoint splitter, but is an external capability as you point out. The claim should be referencing FTS, not RecoverPoint. The IBM 8800 has no such equivalent feature if this is the case.

-- Chuck

John Ford

David, what I learned in class was that only the writes are mirrored for protection and that excess cache is used for reads and they are not mirrored. There is no reason to mirror them. And yes why wouldn't you want to have cache hits for reads rather then hitting the disk?

DumaDum

Hi Chuck
Good post which very succinctly brings out the tremendous innovations that EMC has done to the VMAX product line. However are these innovations really relevant in a real customer scenario?? For example, which applications would benefit from a 2TB cache?? While you may argue that this improves the probability of a read hit, this also depends on the ability of the cache algorithms to prefetch data into cache. Large cache is not going to help if the application involves large number of sequential writes anyways. So does Vmax has something special to justify the investment in large cache or is it just a marketing statement?
Scalability to 3200 disks: Looks great on paper but again does VMAX have the capability to linearly scale performance with the addition of disks considering that the number of backend ports is the same in both 20K and 40K?
Considering that VMAX uses Intel CPUs and we know that Intel has a much smaller lifecycle, would there be investment protection if a customer had to buy a 40K with 4 engines and wanted to buy additional engines say 2 years down the line?
Also surprising is that EMC is using MLC based flash drives. In enterprise class storage systems, I would assume you to use SLC based drives which are more reliable rather than MLC.
You also mentioned that David's understanding of reads being mirrored in cache is wrong. However in the below website, it mentions that of the 2TB cache only 1TB is usable. Can you please clarify?
http://www.emc.com/about/news/press/2012/20120521-02.htm

While I have made a few observation, again let me mention that these are great innovations unmatched by your competition. My point is that not all are relevant in actual customer scenarios. Some of these specifications are just "Great to Know" type..... So I guess EMC is somewhere missing the real picture.

Chuck Hollis

DumaDum

It's good practice to disclose whether or not you work for an EMC competitor, which is clearly the case here. I'm guessing you're affiliated with Hitachi, yes?

VMAX cache is primarily used to soak up OLTP writes.

It also contains what's known as a "track table" which manages device access among other things. In large configurations, the track table can get quite large, hence the difference between the two numbers.

The 3200 drive config is useful in tiered scenarios where flash drives and disk drives of different geometries are used to provide a range of automatically tiered storage services. VMAXes (and the DMXes and the Symm 3,4,5s before them) have demonstrated ridiculously long lifetimes in customer environments that are totally dependent of Intel's roadmap.

VMAX was the first enterprise array to use SSDs back in 2008 (SLC at the time), since then eMLC has evolved to provide an additional attractive options.

All of the information above is public and easily obtainable by anyone with more than a passing interest. Like downloading a spec sheet.

-- Chuck

Ajaz Ahmed

Is there any document that points out the relative merits of using the different Management software for VMAX. What are the advantages of using Prosphere vs using Unisphere, SMC or say Ionix Control Center

Chuck Hollis

Hi Ajaz

Yes, there are pros and cons to each management approach. The differences revolve around you, your environment, and what you're trying to get done.

Some shops are just looking for array-level management, so all of our products come with what we term an "element manager". SMC and portions of Unisphere fall into that category for VMAX and VNX respectively.

Bigger enterprise storage shops want to manage large farms of servers and arrays -- that's where ProSphere (formerly ControlCenter) fits in.

Just about every shop want integration upwards into the stack, exposing storage information and functionality to other roles: VMware or Windows administrator, email admin, database admin, etc. There exists a dizzying selection of add-ins and modules on top of the core storage array management products.

Finally, the data protection folks usually want their own tools for backup and/or remote replication. Enter DPA (data protection advisor) and the new AppSync for more modest environments.

Yes, a lot of tools -- but there are many different roles and needs in IT shops, and storage tends to touch many of them, hence the need to expose the same functionality in different ways to different people.

And, no, I am not aware of a document that lays this all out simply and cleanly. Maybe one needs to be written!

I hope this helps

-- Chuck

Ajaz Ahmed

My understanding was that the Prosphere is a updated version of SMC, while Ionix is the updated version of ECC. Please correct me if I am wrong. Chuck which Management tool is the most commonly used by EMC Customers or are they still using Symcli as thier preferred tool. Now if this is sensitive data then you dont have to answer that. I also read in one of the other blogs that the VMAX engines dont have to reside next to each other and they can reside elsewhere. I am seeking clarification of how "elsewhere" is defined. Somewhere within the same data center or within synchronous distances. It would be a mind blowing feature if each engine can reside at Async distances.

The comments to this entry are closed.

Chuck Hollis


  • Chuck Hollis
    Chief Strategist, VMware SAS BU
    @chuckhollis

    Chuck has recently joined VMware in a new role, and is quite enthused!

    Previously, he was with EMC for 18 years, most of them great.

    He enjoys speaking to customer and industry audiences about a variety of technology topics, and -- of course -- enjoys blogging.

    Chuck lives in Holliston, MA with his wife, three kids and four dogs when he's not travelling. In his spare time, Chuck is working on his second career as an aging rock musician.

    Warning: do not buy him a drink when there is a piano nearby.
Enter your Email:
Preview | Powered by FeedBlitz

General Housekeeping

  • Frequency of Updates
    I try and write something new 1-2 times per week; less if I'm travelling, more if I'm in the office. Hopefully you'll find the frequency about right!
  • Comments and Feedback
    All courteous comments welcome. TypePad occasionally puts comments into the spam folder, but I'll fish them out. Thanks!