March 18, 2010
XtremeData update
I talked with Geno Valente of XtremeData tonight. Highlights included:
- XtremeData still hasn’t sold any dbX stuff (they’ve had a side business in generic FPGA-based boards paying the bills for years). Well, there may have been some paid POCs (proofs of concept) or something, but real sales haven’t come through yet.
- XtremeData does have three prospects who have said “Yes”, and expects one order to come through this month.
- XtremeData continues to believe it shines when:
- Data models are complex
- In particular, there are complex joins
- In particular, two large tables have to be joined with each other, under circumstances where no product can avoid doing vast data redistribution
- XtremeData insists that all the nice things Bill Inmon – including in webinars — has said about it has not been for pay or other similar business compensation. That’s quite unusual.
- XtremeData is coming out with a new product, codenamed the Personal Data Warehouse (PDW), which:
- Is ready to go into beta test
- Should be launched in a month and a half or so
- Will have a different name when it is launched
Naming aside,
- The XtremeData PDW consists of XtremeData software running on a Cray CX1 box.
- Thus, the XtremeData PDW will plug into a 20 amp wall power socket. It consumes 1600 watts.
- The XtremeData PDW also inherits the Cray CX1’s noise cancellation feature.
- Bottom line on the form factor: The XtremeData PDW is meant to be stuck in the corner of a business analyst’s office, not a computer room.
- The XtremeData PDW will have 16 1 TB disks (going up in size later), for 5 TB of uncompressed user data.
- Pricing isn’t finalized for the XtremeData PDW, but it will be around XtremeData’s usual figure — $20K/TB of uncompressed user data.
- XtremeData hasn’t “released” compression yet, but it’s “ready to go.”
- The XtremeData PDW will not include FPGAs, unlike other XtremeData dbX appliances. It will just run the XtremeData dbX software on 8 Nehalem chips.
- XtremeData calls this a “3-node” machine. I didn’t bother asking why it wasn’t 4-node. (Perhaps there’s a head node of some kind that properly isn’t counted.)
Some comparative notes:
- A Netezza Skimmer has similar size and price to the XtremeData PDW, seems to draw less power, has less uncompressed user data capacity (but already has compression), is also in essence a three-node system (I think), and of course has a lot of software connectivity. If XtremeData can match Netezza’s compression, the XtremeData PDW will have a 2X or so price/TB advantage over Netezza Skimmer – but Netezza’s compression is of course a moving target. I don’t know how happy Skimmer is outside a computer room.
- Kickfire manages similar amounts of data on a smaller box (5 rack units vs. 7), drawing less power (600 watts vs.1600), also with a lot of BI and ETL tool connectivity.
Categories: Analytic technologies, Benchmarks and POCs, Data warehouse appliances, Data warehousing, Database compression, Kickfire, Market share and customer counts, Netezza, Pricing, XtremeData
Subscribe to our complete feed!
Comments
5 Responses to “XtremeData update”
Leave a Reply
Netezza’s Skimmer has 1 S-Blade with 8 CPU cores and 8 FPGA cores, so I would say that’s 8 Nodes not 3, agree ?
By that logic, how many nodes would you say 32 Nehalem cores + 0 FPGA cores is? (The XtremeData/Cray box.)
(Apologies if I’m being too pedantic)
Power supply size is not equal to power consumption. The power supply size is simply the maximum that the power supply can deliver in total across the different voltages that it provides.
The actual power draw will be a function largely of the CPU, video, disks, RAM as well as the actual activity level of the server.
The Cray CX1 is basically a small blade device, with capacity for up to four blades with two CPUs each. While you could stick it in an office, pricey hardware like this belongs in a proper server room, especially since there’s not much advantage in having it deskside.
Well, I’d assume the power supply size has something to do with the peak power drawn.
But yes — if a Machine A finishes its work faster, on average, than Machine B, then Machine A hopefully will draw less power overall.
[…] a response to a question I messaged over, XtremeData tells me they have actual customers now. Press releases to […]