Community

Virtual Storage Tiering changes the game again with Flash Pool

There’s no doubt about it: Flash technology is changing the way that IT thinks about storage efficiency, application performance and, yes, life in general. You may have heard the “Flash changes everything” statement from your storage vendor, your peers and even your tech savvy kids but you may still be wondering how to best implement it for optimal impact to your business. At NetApp, we think that question is best answered through an intelligent approach to adding agility to your IT infrastructure and that’s what we had in mind when we introduced the Virtual Storage Tier(VST). And since we did that, our customers have deployed over 15 petabytes of Flash to accelerate over 1.2 Exabytes of their data.  That’s a whole lot of Flash but it’s only a little more than 1% of the total storage so the business impact has been huge based on VST’s ability to combine the inherent performance boost of Flash with the unmatched storage efficiency features which are built into Data ONTAP.  And now, with Data ONTAP 8.1.1, the Virtual Storage Tier has a new family member… Flash Pool

 

Flash Pool, like its Flash Cache sibling, implements VST through intelligent caching of hot data thus delivering real time performance improvement with the right mix of Flash and HDD.  Flash Pool, is actually a specific type of Data ONTAP aggregate which is composed of both HDD and SSD devices. While Flash Cache (implemented as a PCI-e attached controller cache) accelerates random read operations for all data on the storage system, Flash Pool is enabled for specific data sets and provides caching of random write operations as well as random reads. And since Flash Pool is physically implemented at the disk shelf level, it provides a cache that follows data persistently, even in the event of HA failover events. You can deploy Flash Cache and Flash Pool independently or combine them on the same storage system giving you the ability to tune VST to the requirements of specific workloads-you can use Flash Pool for workloads with lots of random writes and Flash Cache will cover the rest. 

 

So you may be asking: Is Flash Pool mostly a performance enhancer or a way to save money? Great question and here’s the good news: It’s both.  We recently put Flash Pool through its paces on a dual controller FAS6210 system running an OLTP workload. We wanted to find out what would happen if we moved the entire workload from a configuration with performance optimized SAS HDDs to one with capacity optimized SATA HDDs.  When we added Flash Pool to the SATA configuration, we saw the following:

 

·         Significantly improved response time

·         50% more storage capacity

·         46.3% lower cost per terabyte

·         18% lower $/IOPS

·         26.5% lower power consumption

 

 

So it’s kind of a no brainer.  With VST, you can meet the majority of your application requirements by combining Flash technology with capacity optimized disk drives- you get the best value without compromise. The best way to use Flash-no doubt about it.

 

Oh, and one more thing, Flash Pool isn’t just for you high end guys. You may have noticed that Flash Pool was mentioned in our recent announcement of the FAS2220 making NetApp the only vendor combining Flash with this unique intelligent caching approach in the entry-level storage market. 

 

For more background on the NetApp Virtual Storage Tier, here are a few other posts which will shed more light on the topic:

·         Why VST Makes Sense

·         The Next Step in Virtual Storage Tiering

·         The FIX is in

 

Comments
wdillaway on ‎2012-06-23 05:39 PM

Could you post the details of the SAS versus FlashPool comparison?  (How may SAS drives, how many SATA drives, how many SSDs, what kind of application running, etc.

         Thanks!

on ‎2012-06-25 12:39 PM

For this comparison, we started with an OLTP(SPC)workload on a FAS6210 HA Pair with 144TB of SAS. There were 240 600GB SAS drives @10K RPM each. The entire workload was then moved to a FAS6210 HA Pair with 216TB of SATA. In this case, 216 1TB SATA drives @ 7,200 RPM. There were also 24 100GB SSD drives which were Flash Pool enabled.

on ‎2012-06-25 01:11 PM

Thanks!  Very helpful.

on ‎2012-06-26 03:47 AM

I have a two questions related to this.

1. If I have a systems with two aggr, and say one is SAS agg, and the other is SATA aggr, and I have a flash-cache to improve the overall filer read performance. Then if I add a small RAID group of SSDs to the SATA aggr, then is it possible to use the flash-cache for the SAS aggr only, if not than will it be doubled cached for the SAT aggr, since the hot blocks will be cached in SSD, as well as flash-cache, I mean how it will treat the hot blocks.

2. As far as I understand WAFL with the help of CP, and NVRAM commit logs, always convert any writes into sequential writes, so in my view all writes in Netapp are sequential write, irrespective of whether the top level writes are random or sequential. And for sequential writes, SATA/SAS/SSD donot have much differences in write performance.

on ‎2012-06-27 06:44 AM

1. For aggregates that are configured as a Flash Pool, Flash Cache will ignore those volumes, so there is no double caching. There would actually be no benefit to such double caching in the Flash layer anyway. So, for those systems with Flash Cache installed, Flash Pool can be used as a more granular level of service level for those data sets that can benefit from write caching and cache persistence across HA takeover-giveback events.

2. Yes, since Data oNTAP is fundamentally optimized for writes to begin with, the write caching capability of Flash Pool isn't really for improvement in write latency, per se. But write caching will help with back end storage throughput since the write load on the hard disk drives is less. If you are trying to avoid back tio back CP's, this can be helpful. Additionally, write caching will slow down the rate of disk fragmentation which will make the HDDs more efficient as well.

on ‎2012-06-27 07:09 AM

Thanks Paul.

Your reply is helpful.

Also I think that flash-pool being available even for 2220 system, so lower end filers which cannot afford to have flash-cache(because of CPU limitations I believe) can take advantage of flash for read caching, which is a good news. Even I remember that 3210 is not suppose to be used with flash-cache.

on ‎2012-07-05 09:39 AM

Hello Guy's

After reading TR-4070 i've interrogation about what really is a "random overwrite", and in which workload (OLTP seems to be one) i can find this kind of disk access?

more generaly : for which kind of workload type (VDI, File Server, Server Virtualization, ...) FlashPool technology should be smart to use?

Thanks in advance for your feed back.

Best Regards,

Yannick,

on ‎2012-07-06 12:43 AM

Yes, but the 3210 can be configured with FlashCache again in Data ONTAP 8.1.1. It will only use 128 GB of the 256 GB card then due to memory constraints.