Analysis

posted on 02 August 2008 15:54

This is the third part of the Blocks & Files tour around the XIV product. Part one is an introduction to the basic architecture and function of the current Release 1 system. Part two explains how XIV storage capacity is organized, presented, used and managed. This part examines the coming Release 2 XIV Storage System and then looks at overall XIV positining and competition. It finishes up by having a brief discussion of the near-term road map.

Release 2 XIV Storage System

In Q3 of his year a revised Release 2 and IBM-branded box will come out with changed HW components and a major software revision to v10.0. It's aim, apart from increasing the manufacturability of the box, is to increase capacity and both external and internal bandwidth:-

- The IMs and DMs will share the same basic 2U base system. This will then be given an IM or a DM personality with the HW elements added to it.

- There will be 8 x 4Gbit/s FC host ports per IM and 2 1GbE iSCSI ports so FC host port capacity will triple overall from 6 to 24. iSCSI connectivity stays at 6 x 1GbE overall.

- There will be 2 x 48-port GigE switches for increased any-to-any internal connectivity.

- There will now be 15 x 2U DMs with 12 1TB SATA drives making a total of 180 HDDs.

- Each DM has double the memory at 8GB. Thus there is 120GB cache overall.

- There is 180TB of raw capacity and 82.5TB of usable capacity.

IBM now talks of single frame systems. This is a temporary restriction as IBM will add multi-frame R2 XIV configurations.

The v10.0 software release adds the capability to restrict users to only accessing their own data with application-level control. Users of one appication set cannot see the data belonging to users of another application set. Users can be restricted to operate only with the LUNS they are authorized to use.

Version 10 of the software adds LDAP support.

V10 could also extend the XIV system's ability to work with host-based multi-pathing. DMP support has been mentioned.

XIV Benefits

IBM says that XIV benefits centre on lower acquisition costs, potentially large disk capacity savings and much reduced management overhead.

Thin provisioning can save 20 - 50 percent capacity. The avoidance of orphaned capacity in the system can save 10 - 20 percent of the capacity. Differential snapshots can save 15 - 30 percent of capacity. This suggests that an alternative array might need 100GB of capacity to do what a 45GB XIV array does. However, the XIV mirrors everything and a full comparison of capacity utilization between an XIV system and, say, a Symmetrix is just not possible with these IBM figures.

Positioning

The XIV is targeted at applications and customers needing very scalable storage who have rejected IBM's DS8000 high-end storage array. The markets are:-

- Online archival data storage
- Web 2.0 with massive scale and fast response and affordability (no Fibre Channel SANs)
- Cloud storage with fast and efficient snapshot-based backup
- Online media delivery with fast streaming
- High Performance Computing (HPC) with clustered servers. However XIV has no InfiniBand connectivity.
- Very large scale development and test.

The XIV is not positioned by IBM as a DS8000 alternative but as an augmentation to it. This seems to me an artificial distinction as the next section will explain.

An implication is that, since it has no file-level access it is not a replacement for IBM's DCS9550-based Scale-Out File Services (SOFS) product bundle. If and when it gets file-level interfaces then it would render the SOFS redundant.

The XIV Nextra name was dropped because it looked and sounded too much like Sun's Netra product brand.

Competition

The main identified competition is 3PAR and its InServ arrays. These have a good base in service providers and excellent scalability - 3PAR(ameters) is the name after all. Then the competition includes mid-range and high-end SAN storage array products from the usual suspects:-

1. High-end arrays such as EMC's Symmetrix, Hitachi Data Systems' USP and the OEM'd Sun 9990 and HP XP versions. (It seems obvious here that IBM's own DS8000 'must' be included in this list since these systems compete with it. IBM's exclusion of the DS8000 as internal XIV competition is arbitrary and probably intended to protect DS8000 revenues.)

2. Because the XIV is so scalable, competition includes mid-range storage arrays such as EMC's CLARiiON, HDS' AMS and HP's EVA. (Again the internal IBM competition 'has' to include the DS6000 on the same grounds as those above and with the same implications.) This category also includes Compellent and Pillar and I would add in Isilon although its market reach is a little limited at present due to its recovering from management changes and financial difficulties.

Another competitor has to be DataDirect.

No NetApp storage arrays are presented as competition, neither FAS6000s nor FAS3000s. This may be good manners towards a partner, IBM reselling these boxes as its N-Series, or it may be because it simply doesn't class NetApp products as mainstream competition in the high-end and mid-range block-level storage array space. That's an intriguing idea and one which NetApp would vehemently disagree with.

It seems intuitively obvious that the XIV product will also compete with EMC's coming Hulk (Infinflex 10000) hardware and Maui software combination.

Futures

The box could be given file-level access through CIFS and/or NFS protocols in 2009. It could also get asynchronous mirroring.

The XIV could have concurrent code load added to enhance availability.

The XIV Manager software could be integrated into IBM's overall TPC management software.

It could have something equivalent to RAID 5 and 6 support so as to free up more usable space. An issue here is how to do that without extending disk rebuild times too much from the current 60 minutes for a 1TB drive.

It also needs more scalability so as to justify the unique scale out architecture tag. This we could expect the XIV frame count to go back up towards the R1 XIV's eight. That implies that multiple XIV frames might present themselves a a single logical XIV storage pool. In fact if a new frame is added to an existing XIV then the intent is to have the existing data redistributed across all the drives in all the frames.

This implies that the distribution map has a frame'ness aspect added to it. It might also imply a concept of primary and secondary frames.

The XIV product could also have System z (mainframe) support added plus System i support as well. The mainframe support suggests an ESCON link.

- - - -

Once again we note that the informaton in the three part XIV product tour is indirect and not from IBM. Therefore, in the absence of IBM cinfirmation it must be regarded as speculative.

^{[Chris Mellor.]}