February 21, 2008 at 8:30 pm
Hi,
A network engineer from my company wants to implement a solution he calls "SAN" which will be one large NetApp storage with single set of controllers and it will host most of the environment: database servers (both OLTP and Reporting/Datawarehouse), file storage, SAS both dev and production (very I/O hungry app), file servers, exchange and everything else.
My concerns are:
1. This is not a SAN, this is just direct attached storage as SAN being Storage Area Network means at least two storages are required to form a network.
2. NetApp only supports RAID4 and RAID6 and although they claim that due to some fantastic architecture their RAID4 outperforms RAID10 and all other vendors storage solutions, I will doubt it until I see some independent tests.
3. Putting all eggs in one basket is risky, especially when you place OLTP database server together with very I/O heavy reporting tools. The vendor argues this is ok though as they have some priority settings that address such issue.
4. Write cache will be full most of the time because of the reporting stuff and this will remove very big advantage of having cache on the OLTP servers
5. We are going to move six SQL servers into two node active/active cluster three instances on each node, which is already a squiz from six separate physical servers with their own controllers and internal arrays.
6. The network administrator perception is to estimate the size of the databases and fill up the storage with biggest possible drives (300GB). He plans to increase the space as needed. I can't convince him that for OLTP database servers you don't measure storage in terabytes, the most important criteria is the number of spindles.
7. From the first look NetApp looks to me as a very simple and budget brand, which lacks some of the features compared to other vendors like EMC and HP but this may be not relevant as there are more important issues I struggle to address.
Am I overreacting? In this company it is considered that hardware is none of the DBA business. What arguments would you use to convince the management that DBA should be involved and nothing should be done on database servers, including hardware without DBA review?
Thanks.
June 6, 2008 at 8:41 am
Firstly as a DBA hardware is very much your concern. Poor performing hardware can lead to poor DB performance so it is very important that you at least have your say on what your database needs to perform at its best, your the first person the CEO will want to talk when the db starts running like a dog!
From previous expereince of Storage and Network guys they seem to be more impressed by how much the squeeze onto the storage and not the performance they can get out of it for their databases... they don't always appreciate how db's work and treat the db LUNS as file shares. just personal experience, not saying they are all the same.
A good argument i have used previously when confronted with a similar situation, is to explain that using the storage design suggested by the network guys, the system will perform at x% less than if you used the technology or configured the hardware the way you suggest...basically the business will 'produce' at x% less capacity (assuming the OLTP drives your business) it won't take a good bean counter long to understand that a small investment upfront can be rewarded with greater business effieciency...point out the bottom line and they'll get the gist.
For what its worth i have worked with both NetApp and EMC SANS previously and you get what you pay for, I would take the EMC SAN over the net app san any day.
Hope this helps...sorry if it sound like a rant...been here before.
Gethyn Elliswww.gethynellis.com
June 6, 2008 at 10:33 am
I've heard good things about NetApp products, but haven't really worked with them, so no firsthand experience.
The vendor wants to sell the thing. They'll always say performance isn't a problem, they can handle multiple servers, etc. However my wife works in sales engineering, I hear about sales reps all the time. Most don't have a clue what they're selling. They're selling!
Document your concerns. Explain reasonably what you've written here and get it to managers and directors. Let them decide. I'd leave out the "low budget" parts since it sounds like a grudge against NetApp, but I'd point out that your two types of servers have fundamentally different access patterns and if their LUNs share physical disks, it's an issue. Also file servers, Exchange, access disks differently.
If they can give you separate physical disks from other servers, not a big deal. R4, R6, R10? It depends. Enough spindles, fast disks, depends on your load, can be faster than DASD. Don't say it won't work, but express concerns.
Big disks are issues. Same thing though, depending on the load, they may or may not work. If NetApp can support different disk sizes, meaning you can have 100GB and the Exchange guys get 300GB, that's ok. Alleviates some concerns.
What's a SAN? Our first one was 1 large drive box (+ controllers), 2 switches for redundancy, servers attached to switches. Don't argue that point. SAN is getting to be strange with iSCSI and other technologies.
If it works, it's OK. Just voice your concerns in a logical manner, allow them to be argued, and get someone else to sign off. That email might help later if things go south.
June 6, 2008 at 10:43 am
Also - there's no compelling reason that everything needs to be rolled onto the SAN on the same day. As a matter of fact - there would be lots of good reasons why you wouldn't want to dump everything on it. So - let them dump a bunch of things on it, and let them figure out it works out.
Next thought would be - just test the crap out of it. You ought to be able to set up a test playground, with essentially the same or similar data. Find the most intense thing you do - and run it on both - compare the differences. Do it at similar times (hint - don't wait for things to be quiet; test this thing when things would be busy - you want to find out how it behaves at the worse time, not the best time), and compare the performance.
And like Steve said - let the vendor "prove" that this won't wreck or hurt your performance. Gut feelings are great - but enough testing should tell you whether it works for you, rather than the "it SHOULD work for you" you will get out of the marketing folks from the company.
Finally - if management wants to take storage concerns out of your hands, then raise the issue, but don't fight them. Just be ready to show them why their decision was stupid. I'd personally start capturing metrics to show how things perform now: if things take a nose-dive - show them why.
----------------------------------------------------------------------------------
Your lack of planning does not constitute an emergency on my part...unless you're my manager...or a director and above...or a really loud-spoken end-user..All right - what was my emergency again?
March 15, 2009 at 10:57 am
NetApp uses a technique called wide striping. Basically, they stripe all the disk in Raid-4 DP (dual parity) and create a pool that you allocate to logical units that you then map back to the server as drives. I have not used their Raid-6.
Because all your disks are utilized they get good performance. However, having all disks in a big stripe puts you at more risk of data loss from a multi-disk failure. To appease people who worry about redundancy and data loss (e.g. DBA's) NetApp is normally set up as a mirrored cluster. This unfortunately hurts their price advantage.
You did not mention interface. Fiber Channel or iSCSI?
March 15, 2009 at 11:20 am
You won't be able to prove which is better without hard figures, look into SQLio and SQLIOSIM to get stats on performance. Benchmark your current config, and then run tests on the new config before it goes live.
---------------------------------------------------------------------
March 16, 2009 at 5:48 am
This sounds like a relatively high-risk approach to your company's storage requirements. Document your concerns. Try to give required read and write I-Os per second, average and peak - this is what you need the 'SAN' to do in order to meet your SLAs. Make sure your manager acknowleges receipt. CYA.
At the end of the day, if it works then all is OK. If it does not, then make sure you can show you are blameless.
Original author: https://github.com/SQL-FineBuild/Common/wiki/ 1-click install and best practice configuration of SQL Server 2019, 2017 2016, 2014, 2012, 2008 R2, 2008 and 2005.
When I give food to the poor they call me a saint. When I ask why they are poor they call me a communist - Archbishop Hélder Câmara
Viewing 7 posts - 1 through 6 (of 6 total)
You must be logged in to reply to this topic. Login to reply