This is part of a series on the SNIA Blogfest 2010.
HDS were our third vendor of the day. Refreshed after a hasty lunch, we bussed out to North Ryde and arranged ourselves around another conference table. A notice on one wall handily informed us that guest wireless was available (huzzah!), but it required a zillion digit hexadecimal access code (boo!) so I just went with the 3 wireless broadband again, which was, mercifully, working.
I must state up front that I’ve had as close to zero experience with HDS disk arrays as matters. I’ve dealt with Sun gear when they apparently OEMed HDS gear, but I have no idea if D1000 or A1000 mini-RAID systems count, as they’re the only things I attempted to configure with the onboard RAID software. I recall one of them sucked. I was very much out of my element compared to the other vendors, but I’m always interested to learn new things.
We met Adrian De Luca, who was flying solo for his stint, brave soul. He had brought a list of our questions with him, a few of which he particularly wanted to talk about, which was just fine with us. And so, we began.
Adrian said that HDS had not observed any slowdown in storage growth during the Global Financial Crisis. HDS had apparently commissioned research by Sweeney Research, who found that in a survey of 400 odd end users in Australia/New Zealand, 67% had not stopped investing in storage technology. They had, apparently, changed the type of storage they invested in, though I don’t have any notes on what exactly that meant.
Thin Provisioning For Performance
HDS have introduced thin-provisioning capability to their gear, and reckon they get a 30% reduction in allocated space, minimum. They started with the USP-V kit, and their modular storage was the second to get the feature. And yet, said Adrian, they asked themselves “how can we do better”.
The way HDS do thin-provisioning seems to have something to do with the way they pool RAIDsets together. To do it, they group multiple RAIDsets (such as a 7+1 set, 1 being parity) into a pool. which is then used as a thin-provisioning pool. This piqued my interest, as spindle count is a big contributor to performance (or lack thereof), and I have particular experience with this at NetApp customers on ONTAP 7.1 or earlier. The XIV also uses large spindle counts to get their impressive performance specs from SATA disks, so I was keen to learn more about the way HDS do things.
They can group at a minimum of 64 RAIDsets together. I specifically asked Adrian for a number, and he didn’t have the exact figure to hand, but said it was definitely more than 64. Apparently HDS use some sort of point-to-point connectivity architecture, and not a bus, so that alters some of the characteristics of how their storage scales. Either way, 64 x 7 data disks is many IOPS worth of spindles, so that’s some serious performance mojo right there. How come we hear so little about HDS gear?
Again, I’m a bit out of my depth here, so forgive my ignorance if this doesn’t stack up somehow, and feel free to correct me in the comments.
This is where things started to get weird for me.
We started to talk about dedupe the feature, and Adrian seemed non-plussed. He said that HDS had found that it generally wasn’t a good economic choice for customers. He chose an anecdote of a customer they had apparently run the numbers for, and they would only have gotten a 5% savings on a VMware dataset of 300 VMs. We all kinda boggled a bit and asked if we’d heard him correctly.
“5%?” we were all thinking, “That’s awfully small for a data type that on any other vendor’s kit dedupes incredibly well. Like 80% well.” Adrian back-pedalled. This was a specific customer that didn’t get great dedupe. Other may well get more than that. Well, yes, I thought, but you chose that anecdote to illustrate your point.
It was all very puzzling.
Moving on, it appears that HDS do, in fact, do dedupe, both appliance based dedupe and content aware dedupe. They also dedupe with their VTL products. Which is where the OEM saga began.
HDS OEM ProtecTIER from IBM. That’s IBM’s inline deduplication gateway appliance line of products. HDS also OEM FalconStor gear. And Quantum DX1 for smaller customers. They OEM CommVault as a backup product, which does dedupe to disk, and also tape. HDS do actually make something called Hitachi Data Protection Suite, which does client side dedupe in the new version “similar to Avamar”.
We had a good long chat about OEMed things. HDS OEM from pretty much anyone. In Adrian’s words, they “don’t want to acquire 17 companies a year” and they try to “develop solutions with best-of-breed partners”. Their price for this abundance of friendliness is that they apparently want some say on the roadmaps of the products they OEM. I suppose when you’ve hitched your wagon so tightly to another company’s product, you’d quite like some say in when you can stop for the loo.
This isn’t a bad position to take per se, but it still felt like HDS were somewhat of a holding company for a bunch of other products, and that all HDS do is whack a different badge on the front before sending them out the door. That’s an unkind generalisation, but I’m trying to give you a feel for the way it came across to me at the time, as an HDS novice.
We somehow got to talking about cloning, or an equivalent feature, specifically in the context of VMware. Adrian kindof dodged the question, and said that it was a little complex to explain (A complex story on cloning? Why?) and may require diagrams. Possibly unbeknownst to Adrian, I am a ridiculous fan of the diagram and have been known to trample slow-moving middle managers in my rush to the whiteboard, given half a chance. If I were a band, I would be Love of Diagrams. Alas, there were no whiteboards in the room (!!??), so we arranged an easel and a flipchart. Remember those?
Adrian drew a picture of their NAS solution (OEMed from BlueArc), which can sit in front of their USP or AMS storage, and is apparently used to make multiple copies of VMware instances somehow. I appear to have gotten a little bit lost here, as I have no notes or recollection of exactly how this worked, though I do recall being puzzled (again) that you could only do copies/cloning with NAS based storage for VMware, which has had a challenging relationship with NFS storage prior to version 4.0. I recall Jumbo Frames not being supported even up to version 3.5.
We spoke about this little exchange in the bus afterwards, and the other bloggers noted with amusement the way I had pressed Adrian for an answer on this particular issue. Apparently he’d been signalling that he’d quite like to move on to something else and I, totally oblivious, had called his bluff. How sad that when I accidentally become a hard hitting journo who asks the tough questions that I can’t even write something decent about what it was about.
Anyhow, moving on.
Hitachi CommandSuite, or HCS, is the centre of the HDS manageware universe. It is positioned as the central tool for holistic storage management, of both HDS kit, and other vendor’s storage.
Adrian said that customers are adopting tiered storage, but the question becomes how do you manage, move, administer, etc. all this “round, brown, spinning stuff.” If you have tiered storage you will, sooner or later, want to move it around non-disruptively. Fair point. HCS is HDS’s answer to this conundrum.
Adrian talked about how HDS want to help customers by having good management tools that integrate with all their OEMed products. Sorry, make that most of their OEMed products, or integrated mostly with all their OEMed products. Wherever possible within the confines of finite resources and all that.
This led into a discussion on an HDS feature called Dynamic Tiering, which is storage tiering at the sub-LUN level. It’s a licensed feature, and there was a bit of back and forth with people on Twitter about the exact kit models that Dynamic Tiering was available on. I’m not entirely clear from my notes, so I’d refer to the #sniafest twitter-stream.
Note that Dynamic Tiering and Dynamic Provisioning are totally different things and that this is very important for HDS people.
We rounded out the day with some quick notes on vStorage API support and VAAI for HDS arrays. It is apparently much improved for VSP, USP-V and AMS arrays.
I apologise if this HDS writeup is a bit thin on the detail compared with my recounting of the EMC and IBM sessions. As I stated, I’m not an HDS boffin, so I found it hard to ask intelligent questions, or to properly comprehend the answers to the questions I did ask.
The whole session was very much in the nature of a fireside chat with Adrian, who was very pleasant and did his level best to give us the information we asked for. We didn’t make it much past question two on his list, so perhaps there would have been more breadth to our discussion if we’d not been quite so inquisitive about the beginning few topics.
HDS appear to form close ties with the business side of their customers. Adrian mentioned that they have business analysis who work with customers to understand what they’re trying to do, and how HDS products can help. That’s a really smart strategy. Even though the dedupe example Adrian used was a poor choice, I was impressed by the fact that HDS had bothered to test their assumptions about whether the customer would get decent bang for their buck. Far too often companies buy technology because it’s new shiny and never check to see if it was worth the cost. If HDS is helping customers to make IT purchasing a business decision instead of a geeks with new toys decision, that can only be a good thing.
Sadly, the HDS story was not well told on the day. That was always going to be a risk for vendors trying out this new thing, and I hope the outcome doesn’t sour HDS on the idea of Blogfests or tech field days. If anything I hope it spurs them to do better next time. After all, we’d had an entire morning to get used to the idea, and Adrian was somewhat dumped in the middle of it; we’d hit our stride and he was just warming up. It’s not clear to me if the somewhat haphazard HDS story is the real one, or just an artifact of the day.
Merch: None. Afternoon tea was provided: some percolated coffee, a fruit platter, and some pastry things including something that looked very much like a delicious cheesecake slice, which I specifically chose not to try, to my continued regret. I did have a couple of melon slices and half a kiwi-fruit, which were nice. I don’t do percolated coffee.