EMC's Champion for Backup Recovery Systems in Dublin...
Even withing EMC it's well known that those old 'Legato' chaps can't spell 'Storage', let alone understand the value of Enterprise Storage in todays market....
Great punditry EMC...
HP has announced the P95000, its OEM version of Hitachi's VSP replacement for its top-end USP-V array. Yet the P9500 is smaller than HP's shipping version of HDS' predecessor, the USP-V. The XP24000 is HP's version of the USP-V high-end array, The P9500 is based on the latest development of that technology announced today as …
Having extensive experience of these things, the only way that you can scale and XP24K out to 2.26PB is to use 2TB FATA drives. Putting those in an enterprise array of this type is not, generally, a great idea. The reason for this is that they are first of all they support very low IOP ratings - half that of a 15K drive. This is compounded by being 4 times larger so the access density is 8 times worse. that the 500GB 15K drives. It gets worse - they HP documentation on these FATA drives state that they can only be used for random IO patterns 30% of the time, or you suffer abnormally high failure rates. Take that at face value, and you have only about 2% of the access density per GB on the FATA drives compared to enterprise for random IOPs (and actually managing the IO access pattern on a huge, shared enterprise array is a tricky thing to do).
If you want (relatively) low access rates on large amounts of data, then it's generally more cost effective to put those disks into an EVA8400 or the like. However, take that duty cycle seriously and be very careful what you use those FATA drives for. Bear in mind that it can take the best part of 24 hours to rebuild a RAID-5 set with 2TB drives (the only way you can max the usable storage) and, all that time, you are vulnerable to a second failure.
I do not know anybody who has maxed out and XP24K with FATA drives, and they would be mad to do so. If you use 600GB 15K drives in an XP24K then the total capacity is not much different to the P5000 and it occupies much less space and uses less power per GB for equivalent capacity and IOP count. If you need more than one chassis, buy a second one.
On any case, it is possible to use those EVAs and larger, slower disks as back-ends to boost the apparent capacity of the array.
Enterprise Arrays are not just about the absolute amount of disk capacity you can put in a box. In fact I would argue that isn't even in the top two or three priorities. First is data security and second is interruptible availability. Connect a few hundred servers to these things and you might as well pull the power supply in your data centre. Then there is maintaining high levels of reliable throughput and low latency on a mixed and changing workload so that when your data warehouse kicks in it doesn't screw up the response time in your call centres. Then you need those added value services - snapshots, array replication. Then you need capacity scalability, but you can (usually) always achieve that, with some compromises, by mapping across multiple arrays (the compromises are about ease of data migration and some array based async replication tools - and some manufacturer's arrays can cope with that). That dealt with, give me two smaller arrays with lower total power consumption and higher IOP count any time than the same capacity with some sub-enterprise disks.
Incidentally, as far as front-end FC interfaces go, there is not an Enterprise array manufactured which can approach the total bandwidth. That's a fact - all these large arrays run out of puff long before they can saturate all that FC capacity. The difficult bit is to balance the damned things so that you don't get hot spots in the SAN fabric, controllers or individual FCs.
So there is a lot more to what makes an enterprise array than the statistic of how many GB you can stuff into it, and a well-designed array will allow you to scale sideways.
"What we see in the Hitachi OEM’d P9500 is a very backward-looking design statement, this is monolithic system design circa 2005. There isn’t an inkling from anything here that the P9500 was designed for the Private Cloud concepts being adopted by customers and productised by VMware and the like."
Actually Mark youve just described most of EMC's products there!
Also what is this constant use of the cloud moniker! Those of us who actually know what clouds are dont call them clouds and certainly dont try to dress anything up to be what it isnt. I pity you if you think that everyone either needs one or is trying to build one!
The P9000 range will be what the XP range was, an array for customers who cannot and will not tolerate anything less than absolute date integrity and maximum performance! Those customers that want to build IaaS will buy the appropriate products (from the HP stable P4000 and X9000 spring to mind as a starting point). Both approaches require different design methologies if you dont understand that then I suggest you probably dont understand the question and shouldnt be commenting!
Also Chris M are you going to get a competitor to quote on all vendors launches (maybe ask a HP or Hitachi person to quote next time EMC release an array?)
"......Also Chris M are you going to get a competitor to quote on all vendors launches (maybe ask a HP or Hitachi person to quote next time EMC release an array?)" Oooh, yes please! I do love a good bit of FUD first thing in the morning - NOT! Whilst hearing supposedly top-notch storage industry bods squealling like children can be fun, I'd much rather have an independent view comparing the products from different vendors. To be honest, I wouldn't ask EMC's opinion of anyone else's array, and likewise for hp's or HDS's opinion of EMC's.
I do agree with the bit above about loading the XP24k with FATA disks to get max capacity - "are you mad" would probably be my first response! If I want lots of cheap disks anywhere I'd be much more likely to use the External Storage capabilities on the XP, keep the fast disk for my tier one storage needs on the XP, and put the second tier and third tier on a mix of FC and FATA/SATA out in attached EVAs and MSAs (or even those funny EMC boxes, which can be borged into an hp-controlled storage pool using the XP). Then again, if size is all that matters (OK, Ms Bee, I'm sure it does to you gals), I can always pool many cheaper arrays with tech like the hp SVSP or CISCO/IBM SVC.
One point missed is the new XP/P9x00 bits seem to come in standard hp racks, a plus point. The other is that they seem to be using some blades tech so hp is probably going to be able to flog these out much cheaper than EMC can push out their top-end designs. All in all, it looks interesting, though I'm not sure it makes the "XP" range safe from a future 3PAR-based development in hp.
"The P9500 does not and cannot replace it."
So because I can't put 1152 2TB SATA drives in it its not a replacement? HP and I believe HDS also now have taken the decision to use the SFF drives becuase thats where the drive manaufacturers are going in terms of product development - so when larger capacity drives are released your argument falls down completely.
Seriously does anyone really buy Enterprise Storage just to load it up with SATA drives?
In my experience with the XP/HDS storage platforms (10 years +) people buy them for typically two reasons:
1) They need the best levels of availability for storage they can get
2) They need to be able to scale performance up and out
Being able to shove in a truckload of SATA drives is rarely a consideration
Of course you neglect to mention through the EMC tinted spectacles that you can always have 247PB of virtualised lower tier storage behind the P9500, but lets not let accuracy get in the way of a good headline eh?
Gotta love the EMC quote though - thats a good sign that they are worried, I remember going on an EMC course and if the instructor had spent more time focussing on their products than fudding the competition (amusing though it was) I might have learnt something more about their products.
Facts not FUD please!
Unfortunately the blades used in the PSP aren't HP but Hitachi ones and are a custom build so no C-class type infrastructure in there. HP tell me the costs will be similar to previous gens in most cases although the entry point will be much lower.
I don't think 3Par can do all an XP can do (yet) or with the credibility in the market place for those real mission critical installations. But I confess I'm largely ingnorant on 3Pars capabilities but excited to understand what they can bring to the party.
Also contrast the Reg article on the HP launch vs HDS's - note the same disparaging comments about whether VSP is fit to lace USP's boots are nowhere to be seen.
Hidden agenda's methinks??
"Unfortunately the blades used in the PSP aren't HP but Hitachi ones and are a custom build so no C-class type infrastructure in there....." I would like to see an XP-size offering from hp built around the C-class chassis as the controller and then lots of SFF disk in accompanying (MSA?) shelves. Would take a lot of engineering work though. Anyone know if the 3PAR InForm OS and sofware can be run on C-class blades?
Biting the hand that feeds IT © 1998–2021