SQL Server Cluster - AX100/SATA vs. P220/DAS performance?

Question

Hi,

I'm trying to determine the best hardware configuration for a clustered database configuration w/ SQL Server 2000 and Windows Server 2003.

We need to run a 2-node cluster, with 2 separate databases on shared storage. We don't have a huge budget, and we're trying to find the most cost-effective way of supporting this configuration.
- 2 fairly-capable systems (probably 2850s with 4Gb of memory)
- 5,000 sustained users, peaks at 10,000 concurrent client connections
- 2-3 transactions per min per user (about 700/sec total peak).
- 10Mb/sec written data, about triple that for read data (OLTP primarily).
- ~ 200 Gb database, expected to double over one year, then level out.

For cost reasons, we're considering one of 2 storage solutions
- PowerVault 220 with multiple drives, directly attached to each node, or
- AX100, iSCSI to each node.

I think both meet our budget/functional requirements (though the SAN propvides more room for growth). My concerns relate to performance of each of these solutions in a clustered database configuration
- the PV 220 sounds ideal, except that write-caching needs to be diabled for each node, since it's clustered. What kind of impact does this have on performance?
- the AX100 sounds scalable, and the price is excellent, but I'm concerned with the ability of SATA to keep up with the IO volume (I've read newsgroup posting suggesting that these drives can't handle more than 30 IO ops/sec per drive!).

Does anyone have any insight into this? I've already read EricB's post about SATA performance, which leads me to believe that SATA won't work but I don't think we can afford a CX-solution either :/

(See http://forums.us.dell.com/supportforums/board/message?board.id=pv_fibchan&message.id=1716&query.id=226760#M1716)

Any insight would be greatly appreciated!

Thanks,

Jeff

tommo666 · Answer

In your scenario, the PV220 would perform best but the upgrade path is limited. An AX100 has a max transfer rate of 150mbs and 30,000 i/o per sec for a dual sp system and has some scaleability. While a 220 can acheive up to 320mbs.

Using fibre in a direct attach cluster for sql i would recommend a CX300, this can achieve 50,000 i/o per sec and 400mbs plus is scaleable for a bit of future proofing.

Dev Mgr · Answer

I'd highly recommend reading Eric's post here.

It goes into SATA vs performance (SATA being AX100). The AX100i will be slower (but cheaper to implement) than the AX100 (fibre channel version) due to the 1 Gbit/s per iSCSI port as opposed to the possible 2 x 2 Gbit/s bandwidth (would need 2 fibre cards per server and 2 switches or a carefully configured single switch that would act like it was 2 switches). From what I remember a salesrep told me the AX100 is supposed to be able to handle a 300-400 user database. Personally I think that's a little on the high side, but maybe I just expect more in the form of performance.

jeff_avery · Answer

Thanks for the advice everyone. I've read Eric's post (that's the same one I referenced in my original post), which is what made me suspect the SATA/AX100 solution might not work for me. I think I'm back to the P220 then, I'll just have to cluster with capable enough machines to handle the load.

All other advice welcome, I have a day or two before I have to move on this :D

Regards,

Jeff

wooglins · Answer

You need to be carefull with the PV220 in a clustered config.    I am currently going thru the same type of project, and we ordered ax100's for two clusters.    As you can guess the performance was very very poor for random/IO which is what SQL and Exchange do.   With the PV220 your write performance will be very low.     I just finished a conference call with Dell working to rectify the situation with the AX100's and a PV220 was ruled out by Dell.   Write performance is in the 10mb per second range fully loaded with 15k drives on the PV220.   You need the CX300,  it can be configured in a more basic way, and I would guess they can get close to your budget point.    Direct connect, no switch, single HBA per server.

Dev Mgr · Answer

One thing about setting up a direct connect cluster with a single HBA per server, make sure both servers connect to SPA or both to SPB. In a cluster Clariion SAN environment, all nodes need access to the same SP or SPs, so one server going to SPA only and the other to SPB only isn't a valid setup. It's easy to fix, but also easy to do wrong when setting it up.

PowerVault

SQL Server Cluster - AX100/SATA vs. P220/DAS performance?

Was this post helpful?