1.015: stranger danger
If you have children, hopefully you've taught them about Stranger Danger at a very young age - prevention and awareness are the most powerful weapons we have to protect our families and friends.
And if Symmetrix DMX could talk, it surely would be yelling at the top of it's blower fans:
NO! I don't know you! You are not my Dad!
Let it hereby be known that Moshe Yanai is not the father of Symmetrix DMX.
No, despite the public assertions of IBM to the contrary, Moshe had virtually nothing to do with the creation of Symmetrix DMX. And on behalf of the hard-working, dedicated engineers, developers and patent-holders who did in fact design and deliver DMX to the market in February of 2003, it is high time to set the record straight.
Moshe's responsibilities for Symmetrix development ended in 2001, long before DMX production even began. And Symmetrix DMX was a radical change in virtually every dimension from the 5 generations of Symmetrix that preceded it under Moshe's watch.
I'm pretty sure Moshe didn't like anything very much about Symmetrix DMX.
And I will point out that Moshe left EMC in 2002 (for whatever reason), which was well before DMX was even introduced to beta sites or discussed under NDA with analysts and prospects.
I was the marketing lead for the launch of Symmetrix DMX, so I know first-hand that he wasn't involved.
So, as you can imagine, I find it curious that IBM is attempting to mislead everyone who will listen that the success of DMX is a feather in Moshe's cap - evidence of his storage prowess and foresight.
On behalf of DMX I say: thanks for the recognition as the market leader. But he still ain't my dad!
actually...
IMHO, the success of DMX is in spite of Moshe.
From the outset, Symmetrix DMX has contradicted many of the design tenets that Moshe seems to embrace even still today. Some will even say that the Symm 5 (the 8000 series) embodiment of those design tenets was no longer competitive in 2001 or 2002, and that Symmetrix fell behind the competition and nearly cost EMC it's market-leading position.
And others will note that the bus-based Symm 5.5 was the last Symmetrix designed and built entirely under Moshe's watchful eye, and that Moshe's vision for Symm 5.6 (also bus-based) was in fact scrapped in favor of the Direct Matrix Architecture.
And the observant will note that EMC not only changed the name, but it changed the entire look and feel of Symmetrix with the introduction of the DMX, as is evidenced by this historical perspective of the product line (from 2005).
As I said, with "the DMX" (as Joe Tucci calls it), EMC introduced a radical new trajectory for Symmetrix customers, different from its predecessor on multiple dimensions (not just the color scheme).
And today's DMX-4 is even more different from the last Symmetrix that Moshe built, as the table below highlights:
 Symm 5.5 |  DMX-4 | |
| Design | Monolithic | Modular, incrementally scalable |
| Max | 64 | 133 |
| Memory | 128GB raw, | 512GB raw, selectively mirrored (256GBu) |
| Max Drives | 96 (single bay) | DMX-4 950: 120 (single bay), 360 (2 bays) |
| Drive | SCSI | 4GB Fibre Channel & 3GB SATA-II |
| Supported | 15K: 36 & 73GB | Enterprise Flash Drives: 73, 146GB |
| RAID | RAID 1, | RAID 1, RAID 10, RAID 5, RAID 6 |
| Max Internal | 1.6GB/s | 128GB/s |
| Cache | Single Queue | Multi Queue TBC (LRU+LFU Hybrid) |
| Storage | Full Device Only | Full, Thin and pre-allocated Thin Devices |
| Maximum | 8,192 | 64,000 |
| Remote | Sync & Adaptive | Sync, Async, Adaptive, Cascaded |
| Local | Full Mirror, | Full Mirror, Full Clone, |
| Connectivity | 1GB FC, ESCON | 1-2-4GB FC & FICON, ESCON, |
| Quality | Permacache | Dynamic Cache Partitioning, |
| Management | CLI, | CLI, Industry-standard SMI-S, Symmetrix Management Console, EMC z/OS Storage Manager (EzSM), Control Center, |
That's quite a lot of changes since Moshe last worked on Symmetrix, don't you think?
Truth be told, this table doesn't even begin to tell the story of how different DMX is than the Symmetrix that Moshe used to know. Over the past 5 years, the scope of architectural changes, code re-implementation, component optimization, and feature enhancements that have gone into Symmetrix are nothing short of mind boggling compared to what Symmetrix was before.
Symmetrix DMX ain't the Symmetrix
that Symmetrix's daddy would have built!
nextra ain't no symmetrix, either.
In fact, it seems that the XIV Nextra that IBM bought is trapped in the pre-DMX world of storage.
At the very least, the product that they bought (as opposed to the one that they'll announce under the IBM brand later this quarter, I'm told) - the product they bought bears many of the same trademarks that plagued the Symmetrix 8000 series.
Remember the old joke "With Symmetrix, you can have any type of RAID protection you'd like, just so long as you like mirroring (RAID 1)?"
Same is true for the Nextra.
Oh, they'll argue that SATA drives are cheap, and thus it's not a problem. Plus, they'll say, Nextra's RAID-X gains so many benefits by mirroring at the 1MB blob level instead of at the disk.
As I've noted before, this was the initial positioning of EMC Centera as well. But guess what - people don't like paying more than 2x for their storage than they need to. RAID 5 and RAID 6 are perfectly viable, and in fact they provide better protection against data loss caused by 2 drive failures than does Nextra's RAID-X. That's right, even though it mirrors each blob, RAID-X has a higher probability of data loss than does RAID 5 (3+1).
But mirrored disk isn't the only Moshe legacy to be found in Nextra.
Being built out of essentially what are commodity PCs, the memory in the Nextra storage nodes isn't mirrored. Long a thorn in the side of Symmetrix, mirrored memory is pretty much standard in every storage platform today - and not just in the enterprise. (Selectively mirrored global memory was added to DMX-3 in 2005).
Async replication? Not in the current Nextra (added to Symmetrix after Moshe left).
Non-disruptive upgrades and hot code loads under full I/O workloads? Uh-uh - Nextra no can do (a foundation feature of DMX from Day 1).
Incremental scalability? Sure, if you don't mind starting with 180 drives, and scaling (next year) in increments of 180 disk drives (DMX scales from 32-2400 drives in increments of ONE).
Maximized storage utilization? Sorry - at best, only 45% of the storage you buy in a Nextra is usable - the rest is reserved for mirroring your data and as spares in case a drive or a (non-redundant) storage node fails. (That means you'll need twice as many disks on a Nextra to get the same capacity as with Symmetrix using RAID 5 or RAID 6)!
And that's before we even get into Mainframe (nope), Series I (nope), Ethernet-based replication (nope), in-the-box tiered storage (nope), 3rd party replication (nope), QoS management (nope), 3-site replication with disaster restart and incremental recovery (nope), third party application integration (nope), industry standard APIs (nope), eLab host/HBA/SAN qualification ...
Well, you get the picture.
Nextra is a long way from addressing the market requirements and multiple storage tiers that Symmetrix DMX fulfills today. And even though we have yet to see actual Nextra pricing, I suspect that you'll be able to add 90 1TB SATA drives to your DMX-4 for less than you'll be able to buy a 180-drive Nextra for (unless, of course, IBM gives them away like they are doing today with the 43-month-old-and-counting DS8000 - but that's another story altogether).
Some would even go so far as to suggest that Moshe has built a great new storage solution that will compete well.
Compete with Symm 5, maybe...
it's independence day
Tomorrow is July 4th, and here in the United States we annually celebrate our independence and give nod to our forefathers who documented it in our Declaration of Independence.
And just as we Americans owe a lot of who we are to our British forefathers, so too does the entire Symmetrix community owe a lot to Moshe and the team he led through the first 12 years or so of Symmetrix' life. I mean not to detract from his contributions in any way - they were significant in their day. And I almost have to applaud him for tricking IBM into buying Nextra in the first place (although I doubt there are many DS8000 users who would join me, given the obvious uncertain future for that has-been platform).
But just as the United States of America is no longer referred to as a British Colony, Symmetrix DMX (and with the community of people behind it) deserve the same respect.
We are no longer that which Moshe made.
We are today what we have made of ourselves.
Happy Independence Day!
technorati tags: Symmetrix,DMX-4,DMX,EMC,IBM,Moshe Yanai,XIV,Nextra,tiered storage,enterprise storage,Symmetrix DMX,DS8000
Interesting to see the comparisons. Can a DMX really sustain 128GB/s ? Is that worth mentioning? By my 'simple' maths that would need 320 4Gbit host ports (using marketing MB)...
Interestingly missing from the comparison is a host side IO/s and MB/s performance comparison???
Posted by: Barry Whyte | July 03, 2008 at 05:43 PM
"Non-disruptive upgrades and hot code loads under full I/O workloads? Uh-uh - Nextra no can do (a foundation feature of DMX from Day 1)."
That's really not 100% true in this DMX customer's real world experience. But then what is meant "non-disruptive upgrades" generally needs to defined carefully. It largely depends on how big a code jump you are making (in-family vs out-of-family). The picture seems to be a bit better with the DMX3 and now DMX4 iterations.
And lets just accept for a moment that the actual microcode update is non-disruptive at the frame level. If you have to touch and patch 100's of hosts attached to that storage and do patching -- including HBA firmware upgrades(*) which = reboots which may = production application downtime depending on your HA situation -- then code upgrades can really be quite disruptive.
it's not like EMC is the only vendor guity of this. But "non-disruptive" upgrades sounds like marketing gobbeldy-gook to people who actually have to live with the consequences of those "features"
(*) e-lab navigator tends to just track the most recent kernel/drive/HBA firmware vendor releases. So even if relatively little has changed within those drivers -- perhaps just an incremental version change -- if the versions you're running don't match what's in the compatibility matrix, you pretty much have to go through the host-side process.
In other words e-lab will just take the latest code for an external vendor's driver, certify it, and that's _the_ supported version. even if the .x.yy version still works just fine and was _the_ supported version in a previous support matrix, it's no longer "supported."
Posted by: Scott Harney | July 03, 2008 at 05:52 PM
I would disagree to an extent with you scott. I've done several upgrades to my dmx without having any disruptions, so I would, at least from my point of view have to say the statement is true. Now, I haven't had a change that required firmware upgrades on my hba's...I'm not aware of any pending either....But there's always the first time.
Posted by: william bishop | July 07, 2008 at 02:14 PM
And I have to disagree with you William, I've done out of family upgrades which have caused downtime and hence are disruptive. It is probably true that the majority of upgrades are non-disruptive but it is also true that some can be disruptive. There are also some other maintenance activities which can be very disruptive as well.
No vendor is perfect however, I've also got scars from DS upgrades going wrong as well. Just to keep a sense of balance.
Posted by: Martin Glassborow | July 08, 2008 at 04:09 AM
Speaking of balance, just to point out that my comparison above was in relation to the ability to upgrade the code on the storage array without having to shut down (or stop all I/Os from) all of the dependent hosts simultaneously during the upgrade.
DMX can do it; Nextra cannot.
(One could also argue that neither DS8000 nor USP/USP-V can be upgraded non-disruptively either, since both require you to severely limit I/O demand during the upgrade process, and failure to do so reportedly can cause either system to halt unexpectedly...YMMV).
Posted by: the storage anarchist | July 08, 2008 at 06:59 AM