SAP on NetApp

[FERDIE_GOJO]

Hi,

I have a question regarding SAP on NetApp.Basically I was told that each node has it's own set of disk drives and volumes that is controls. For failover, it takes 30 seconds or more for the disk drivers to move to the other node, so this will affect the OSes running on the failed controllers, since the Luns/Disk Drives would be unresponsive for over 30 seconds. I'd like to get confirmation from if this is really true or not, plus what alternative solution can NetApp provide?

We are looking at two storage solutions for our company. Below is a throughput & DB size info for your reference.

Current Disk Usage All		Current Requirement	3 Year Growth	Total After 3 Years	Deduplication Percentage	Netapp Requrirements with Deduplication	Netapp Flexclone Requirements (30%)	XIV SNAPShot Requirements (40%)	Netapp at Shaw with DR	Netapp at Q9 with DR	Netapp at Shaw with no DR	Netapp at Q9 with no DR	XIV at Shaw Court with DR	XIV at Q9 with DR	XIV at Shaw Court with no DR	XIV at Q9 with no DR
SAP	SAP Production	9,709	2,690	12,399	0.00%	12399	16119	17359	16119	16119	16119		17359	17359	17359
	SAP Non-Production	20,177	5,380	25,557	0.00%	25557	33224	35780		33224	33224			35780	35780
	SAP TSM	1,457		1,457	0.00%	1457	1894	2040	1894	1894	1894		2040	2040	2040
	Future SAP Projects	1,028		1,028	0.00%	1028	1336	1439	668	668	1336		1439	1439	1439
Technical Services	File Services	2340	702	3,042	30.00%	2129	2768	4259	2768	2768		2768	4259	4259		4259
	Database	810	243	1,053	0.00%	1053	1369	1474	1369	1369		1369	1474	1474		1474
	ESX	17324	5197.2	22,521	60.00%	9008	11711	31530	11711	11711		11711	31530	31530		31530
	TSM	4397	1319.1	5,716	0.00%	5716	7431	8003	7431	7431		7431	8003	8003		8003
	Email (Include this number with exchange for recovery)	1135	340.5	1,476	0.00%	1476	1918	2066	1918	1918		1918	2066	2066		2066
	Exchange (number from Netapp)	8865	2659.5	11,525	0.00%	11525	14982	16134	14982	14982		14982	16134	16134		16134
	Cardlock (Guess)	2000	600	2,600	0.00%	2600	3380	3640	3380	3380		3380	3640	3640		3640
	Intranet Re-Launch (Guess)	2000	600	2,600	0.00%	2600	3380	3640	3380	3380		3380	3640	3640		3640
	Office 2010 Project (Guess)	2000	600	2,600	0.00%	2600	3380	3640	3380	3380		3380	3640	3640		3640
Total		73,242	20,332	93,574		79148	102893	131003	69000	102225	52574	50319	95223	131003	56618	74385

[aborzenkov]

Yes, each controller in HA pair has own set of disks which are taken over by partner if controller fails.

Takeover time depends on many factors, but 30 seconds could be considered as rule of thumb average. Hosts are expected to be configured so that they won’t fail (by settings appropriate timeouts, number of retries etc).

Please understand that any vendor that offers failure tolerance against path failure will face exactly the same situation – there will be some timeout before host finally gives up on failed path and continues (retries) over remaining ones. Such timeouts are usually in order of 60 seconds. So there is nothing unusual in how NetApp behaves.

[pascalduk]

I have good experience with running SAP, its Oracle databases and virtual ESX servers over NFS on NetApp storage. When the timeout has been configured properly on your (virtual) hosts, storage failovers do not cause problems.

[paulstringfellow]

As other people have said on here, configured correctly it shouldn’t have impact on your database…

It also maybe a matter of what you are trying to achieve as well… for example technology such as metro cluster would allow you to spread the controllers across datacenters with automated failover of disks…and things like local syncmirror allows you to protect at a disk shelf level…

Of course you also have to look at your server builds for config purposes etc…

Then finally gets to the major value of NetApp in this environment, things like dedupe for space efficiency, the protection technologies for SAP to protect workloads with snapshot backups and snapmirror/snapvault to get this data to an alternate location.

Flexclone another excellent feature in a SAP environment, allowing for zero sized clones of production data to provision up to your dev, test, QA environments are hugely efficient in terms of space and speed of delivery…

So without giving to much of a netapp sales pitch… the NetApp value is a much bigger proposition than how quickly cluster failover works…and as has been said…cluster failovers work pretty much the same for all vendors…which is why I thought mentioning the rest of the value proposition maybe helpful…

Good luck with your decision…

[FERDIE_GOJO]

Thanks for your reply.  I did a search about metro cluster and it seems that metro cluster for SAP does not really work, you need your app servers close to the Database. So 3-4 Hours to spin up at the alternate site meets the business requirement, so I would keep it simple. Also would need hot standby servers for Database and Application servers (high cost for something that mey never be used). I think Metro Cluster for email is good but maybe not on our SAP environment.

[paulstringfellow]

I think you’re right…metrocluster is a specialist technology but needed for people who don’t want downtime…it needs to work with a server environment capable of taking advantage of it…so for example we have some installations where we have vmware stretched between DC’s with metrocluster…so vm’s can be spun up immediately in the DR site if needed…no long delay…

In your case if you are going to mirror to another site, then flexclone will have massive value as you can clone you DR instances, without having to break the mirror relationship, to do full DR test.

Obviously flexclone great for SAP environments in terms of Dev, Test, QA etc…zero sized copies of production, created quickly and efficiently…

Anyway all I’d say look at what NetApp can bring, much more function that anyone else on the market…but then look at if that technology is relevant and has business value…if it does NetApp is your answer…if not…then maybe other platforms have validity…

Good luck on your search!

Feel free to drop me a question if you need to…