Years ago Red Auerbach was asked which player he would draft first, Bill Russell or Larry Bird. Red thought it over and decided that Russell was the better choice because “he would get you the ball”. Then, after thinking a bit more about this decision, Red added “then I’d trade for Bird”.
Tonight is the NFL draft. There has been a lot of data analysis over the last few months, even years, regarding the selections that will be made tonight. For teams on top of the NFL, tonight is their night to make an effort to pick players that help them to stay on top. For the teams on the bottom (hello Cleveland!) tonight is a night for them to rebuild everything.
And that brings me to the thought I had today regarding drafts, and rebuilding.
If you were given the chance to build the ultimate server, what would you draft first? I don’t care if you want to run Linux, Unix, or Windows, every server and O/S has the same constraints: network, memory, disk, and CPU.
So, which one would you draft first?
I know lots of data experts who make a living off of configuring disk and storage networks. They will tell you that the secret to performance is getting the fastest disks possible. These people are the Chip Kelly of server performance. Every Eagles fan is nodding their head, and sobbing, right now.
I know folks that will throw RAM at every performance issue. They go out of their way to say that the secret to good performance is to load everything into memory as quickly as possible so that queries will run faster. These people are the Jerry Jones of server performance, people who throw money at a problem and never seem to buy their way to a proper solution.
Then there are those that insist everything depends upon the best and newest CPUs available. These people are the Dan Snyder of server performance, always moving on to something they think is better than where they were but ultimately not going anywhere.
If I was tasked with drafting the parts for building the best server possible, my first pick would be used for the network. None of the above matter if you can’t move the data through the pipe. This is why the New England Patriots with Bill Belichick are the best team in the NFL (#FreeBrady). They know how to assemble the pieces, and in the correct order.
Here’s my server draft ranking, assuming money is not a factor:
1. Network
2. Disk
3. Memory
4. CPU
For a database server I may swap disk and memory but I’d want to know more about the specifics of the workload. And I haven’t even brought into this discussion the financial aspects here. Same as with the NFL, you may find yourself going after hardware based as much on finance as performance.
What would be your server draft order? I’m thinking I should revisit this post in a few months and break it down a bit more, adding in specific hardware components and prices.
Am I not allowed to buy the software providers who will eventually turn whatever lovingly-crafted server I end up with into a festering POS with their POC software? I’ll obviously only do this to get them to change their shoddy programming practices and Do It Right, rather than anything more nefarious…
*cough*
If you have the budget for it, why not?
Ha, clearly Belichick would first try to refurb an older Orange server, cobbling together new with old parts, and learn that doesn’t work well. He’d then move to a new data center, luck into a SuperDome and then build up the network to prevent any outages.
…and be the G.O.A.T. for doing so…
I am a NOOB in the SSAS world and have been tasked with figuring out the specs for a DEV and PROD server. From what I am reading in whitepapers and other guides, I should be looking at 1.Memory, 2.CPU, 3.Disk. I hadn’t seen mention of network before. I understand why, it is the what do I need that I don’t get. Are we talking multiple 10gb NICs, some sort of fibre controller… It looks like I have some more searching to do with that parameter. Considering this is Tabular and not Multidimensional it was said that Disk speed wasn’t as important and memory and CPU. I’m looking forward to your break down. Right now I figured I would look for comparable specs to our QlikView servers since they both work off calculating measures off in memory tables.
Physical Memory = 261928Mb
40 Core Intel Xeon CPU E5-269 v2@3.00 Ghz
I don’t know about this disks outside of them being part of a SAN.
Yes, for some reason network is overlooked when it comes to specs. perhaps that’s because it is the most difficult thing to change. I don’t know. What I do know is that if you have a small network pipe, it doesn’t matter how fast everything else might be.