Late last and most of this year, Oracle has promoted (or is it rebooted?) the FS1 and there is always the comparison to XtremIO. So how do they compare? I started reading the FS1 administration manual and was quickly engulfed in a range of features that in my opinion were overly complex and didn’t apply to all-flash configurations. In the end I was relieved as comparing the FS1 to any storage array would be exhausting. Thus no comparisons in this blog but rather my impressions of a few key differences.
Most all-flash arrays promote two key benefits: sub-millisecond performance and simplicity of management. XtremIO has been designed from the ground-up as an all-flash array: no legacy storage features. Provisioning storage in XtremIO is an easy three step process:
- Create new volume
- Create Initial Group(s)
- Map the new volumes to the initiator group
Here is one of my favorite XtremIO YouTube videos showing volume management and tagging (about 2min 30sec). These three steps provide the consistent sub-millisecond performance DBAs need for their databases. The power of XtremIO all-flash transforms storage management into a simple lightweight process.
In reading the FS1 manual the (part number E52309-07) my impressions were:
- Lots of storage options and many of them legacy features
- High storage management costs because the complexity of all the options
Please keep in mind the context of this discussion is all-flash arrays. In the banner picture you can see my attempt at organizing the FS1 QoS features. Many of the features are legacy and don’t apply very well to all-flash storage. The choice of using RAID levels (5, 6, and 10) is an example of how the FS1 uses legacy RAID versus RAID designed for all-flash.
Figure 2: XtremIO XDP Compared to FS1 Use of Traditional RAID for All-Flash
Figure 2 summarizes the differences between using traditional RAID and XtremIO Data Protection (XDP). With the FS1 the DBA has the classic complexity of choosing RAID and number of drives for performance and protection of their databases. RAID 10 has the lowest read updates and two writes per strip update but has a 50% capacity overhead. So let’s stop there a moment and answer the question, “Why does Oracle always compare the FS1 to XtremIO using raw space?” Here is the respective specification sheets for each: XtremIO and FS1. The 8 X-Brick XtremIO Cluster has a raw capacity of 320TB and a usable capacity of 268TB. The FS1 has a maximum raw capacity of 912TB but does not list usable capacity. Why not? I suspect the FS1 data sheet doesn’t list usable capacity for two reasons: it depends on the configuration (again with the complexity) and the usable capacity would be significantly less than raw capacity.
Using the capacity overheads in figure one let’s show what the usable capacity is for the maximum FS1 configuration of 912TB:
- RAID 10: 50% overhead means the FS1 would have a useable capacity of 456TB
- RAID 5: 25% overhead means the FS1 useable capacity would be 684TB
- RAID 6: 25% overhead means the FS1 useable capacity would be 684TB
These numbers illustrate why Oracle likes to compare raw capacity as there is no loss of disk space to RAID. Usable capacity also has an impact on price. I’ll reference Wikibon’s article Flash will soon be cheaper than disk, Wikibon research confirms as they show the difference in technologies designed for all-flash and those that are not. From the Wikikbon article:
- Standard all-flash array costs up to $16,500 per usable terabyte. The FS1 is in this category as it doesn’t offer full data reduction and sharing.
- With full data reduction and sharing enabled the cost per usable terabyte drops to just $1,875. Seems inline deduplication significantly lowers the cost per terabyte.
- However, flash is close to the crossover point, with fully compressed and shared array coming in at $225 per terabyte. XtremIO is in category as it offers inline deduplication and compression.
Now this is an important point: I’m comparing storage technologies as they relate to usable per terabyte costs. Oracle’s database does have Hybrid Columnar Compression (HCC) but it is a database feature and not a storage feature. The FS1 does not have compression, data reduction or data deduplication capabilities so the only way to remain competitive is to significantly lower (including software for free) the price of the array and perhaps attach unrelated database licensing savings to the purchase of the storage array. Oracle will sometimes call into question the effectiveness of compression and deduplication but our proven published papers by EMC and third party show a consistent 2X compression savings and very high deduplication savings. Here are some examples:
- Principled Technologies 3rd party test report Scaling Oracle 12c Database Performance with EMC XtremIO Storage in a Database-as-a-Service Environment has these proven metrics –
- Compression of 2.2 to 1 for Oracle databases
- Overall storage efficiency of 51 to 1 with deduplication, compression, and thin provisioning
- EMC paper Oracle 11g and 12c Database Consolidation and Workload Scalability with EMC XtremIO 3.0 has these proven metrics -
- Compression of 3.4 to 1 for Oracle databases
- Overall storage efficiency of 130 to 1. This was achieved by creating 192 copies of a database which took no initial capacity!
Achieving 2X compression for Oracle databases on XtremIO is easy but how does it impact usable capacity? An XtremIO array fully utilized with Oracle databases will have a usable capacity of 536TB. Include inline deduplication of 2 to 1 (very conservative) the useable capacity increases to 1,072TB. In the EMC paper referenced above three Oracle databases were each copied 64 times (total of 192) and the physical space used was 13TB and the logical space used was 1,690TB. The more copies of Oracle databases on XtremIO the greater your capacity savings.
Summarizing the points of today’s post:
- With XtremIO provisioning storage is a simple three step process
- XtremIO was designed from the ground-up as an all-flash array
- The XtremIO specification sheet show both raw 320TB (8 X-Brick w/40TB per brick) and usable 268.7TB (about a 16% difference between raw and usable)
- XtremIO has inline compression, deduplication, and thin provision (applies to all data on the array)
- XtremIO compression, deduplication, and thin provision effectively increase the usable space. This is proven and published across many papers.
- XtremIO compression, deduplication, and thin provisioning lower the cost per terabyte
FS1 all-flash array
- Provisioning storage can be complex with all the QoS Plus options
- The FS1 has many traditional storage array features that do not optimize using all-flash
- The FS1 data sheet does not publish usable capacity
- The FS1 does not have storage features like inline compression and deduplication
- Oracle messages HCC but that is a database feature
- · Without storage based inline compression and deduplication the cost per terabyte is higher
- Oracle must lower the cost to cover the price difference
- Oracle might discount other licensing as a package approach to be more competitive
Looks like I’m going to have to make this a two or three part series as there isn’t the time to explore other areas. My next blog will be an overview of Oracle OpenWorld as I hope to learn more about the direction Oracle is moving.
The opinions expressed here are my personal opinions. Content published here is not read or approved in advance by EMC and does not necessarily reflect the views and opinions of EMC nor does it constitute any official communication of EMC.