Thinking Inside the Box

RE: Change, but no Change

Scott_Schweitzer — Thu, 27 Aug 2009 21:10:27 GMT

What role do intelligent programable processor based 10Gb Ethernet NICs play in the above design? We're talking about NICs that can easily steer traffic to various host memory or processor cache buffers via OS bypass. Wouldn't this move them from outside the snippet processor to inside it? How might this improve performance for Netezza?

RE: Shooting the Curl in San Diego

pfrancisco — Tue, 04 Aug 2009 22:18:21 GMT

Funny. I went into the post looking for a really cool quote about surfing. I did find some but they just seemed way too serious for such a festive occasion. Plus, I never fail to LOL at Anchorman - so when I landed on quotes & images from the movie in my Google search, it was a must!

RE: Shooting the Curl in San Diego

karinakarina3 — Tue, 04 Aug 2009 17:26:49 GMT

I LOVE that you quoted Anchorman here.

RE: Change, but no Change

shawnkfox — Sat, 01 Aug 2009 01:24:49 GMT

3X network performance vs. prior models implies that there are 2 gigabit Ethernet ports per blade but that is not shown anywhere in the diagrams. Is this correct or am I just bad at math?

Nothing is stated about the network connection between the blades and the host either. The current model has 2 gigabit Ethernet ports connected to the SPUs which effectivly limits data loads to 500GB/hour at best. The 2TB+ per hour load speed implies that either the host now uses 10 gigabit Ethernet to connect to the blades (and to the external network as well), or perhaps that data is streamed directly to the blades from external systems which then rebroadcast data to the other blades, thus preventing the bottleneck in the host server. How was the system designed to get beyond 500GB/hour data load rates?

RE: "Netezza Underground" - NPS at your bookstore

dbirmingham — Thu, 18 Dec 2008 22:06:14 GMT

Hey, thanks for the feedback.

RE: Setting the Record Straight

karinakarina3 — Wed, 17 Dec 2008 23:15:59 GMT

Crodgers - great comment! I would encourage you to comment on Greg's blog too, like Mark did. Your points are very valid and they should be heard!

RE: Setting the Record Straight

crodgers — Wed, 17 Dec 2008 22:38:07 GMT

I see the point that the blogger is making, but he is not looking at it from a customer's perspective. I think most corporations look at it from the business case to upgrade. It is based on current in-house conditions (can't make batch load windows anymore, monthly sales analysis takes too long, customers are waiting for answers that should be quicker based on perception, etc). So, businesses will shop around for what fits their needs and the comparisons are against what they currently have. I think that is fair in the sense that it tells you what your new capability will be with the new hardware/software/whatever (lessened batch window by x hours, reports in x fewer days, etc). If Oracle doesn't want to benchmark and claim stats on systems they have replaced, then that is their (marketing) loss.

A much bigger point not mentioned by the blogger, but brought up in a response and here is cost. That is where you can show a whole other side to the comparisons. Go ahead and think of the challenge... get two systems that are a "fair comparison" on a hardware level. Can that be done? NZ has 1:1 cpu to disk store. Is that comparable to HP/O DM? Some could argue not, since there is not the same ratio in DM. But, let's get over that (otherwise there will never be a fair comparison).

If you could build out a comparable machine/system that could perform the same throughput as the NZ server, see how they perform. Spewing data is only a part of the process. Intelligently selecting the relevant data to deliver is a bigger part. Tools to help that part of the process can be costly on some systems (indexes, preaggregated tables, etc) - costly in both size/footprint and maintenance.

So, let's say we can build a comparable system that can hold the same amount of user data (let's say a 5 TB fact table - not too small, yet not the biggest out there), and perform queries at about the same rate, and deliver data at about the same rate. It would be everyone's "latest and greatest". IF you can get these great comparable systems together, what would the comparison be like in terms of

1) footprint (racks),

2) logistical cost (electric and cooling),

3) operational cost (number of experienced DBAs, time/bandwidth/cycles spend keeping the data in order, etc), and, oh yes

4) purchase cost (both initial and annual maintenance)?

Where would various vendors stand on these points? Would that be a fair comparison? Would it take IBM 8 times the footprint and 5-10 times the annual cost to perform the same as a comparable NPS? Would it take Oracle 4 times the footprint and 5 times the annual cost to produce the same performance as an NPS?

Food for thought...

RE: Setting the Record Straight

crodgers — Wed, 17 Dec 2008 21:59:25 GMT

Actually, "The Power to Question Everything"(TM) IS a slogan from Netezza. It is still on the splash screen of the NPS Admin Gui (4.5.2p4) and it is on literature from slide presentations to educational documentation. It looks like it was recently shortened to "Question Everything"(TM).

RE: "Netezza Underground" - NPS at your bookstore

shawn_ramsey — Wed, 17 Dec 2008 21:26:38 GMT

I was exited to order the book after reading your blog. I have been looking for a while for some 3rd party book that I could pass onto my colleagues when they wanted to know more about the Netezza. Unfortunately, I guess I am still looking for that book. After spending a week of commutes trying read the book and work through my issues with the ranting and author's humor I have given up on it. There are good and valid points and observations sprinkled through the book but they are they are lost in the other "stuff".

RE: Setting the Record Straight

jlowNZ — Wed, 17 Dec 2008 21:16:12 GMT

Speaking of things to question....For someone who claims to have done "investigative research" on Netezza, I would imagine a good, first place to go might have been Netezza.com. However, it's clear that didn't happen. Then Greg would have quickly seen our slogan is not "The Power to Question Everything," as he claims.