- casino online slots

On the Street G2 about IBM XIV

by Administrator on July 1, 2008

A reader just pinged me with this information.  So concise and thoughtful it was, I think it merits a post here:

I recently attended a meeting put on by IBM about their new XIV disk system. I thought you might be interested in the notes I took. This was a public meeting, so IBM is not trying to keep any of this secret.

0) IBM will keep XIV as a separate company with it’s own sales force.

1) Provides thin provisioning with the ability to reclaim storage that is no longer used by files. There was no explanation of how reclaim would work other than that it would reclaim “zero” blocks.

2) Provides snapshots which will not bog down as writes are performed. No explanation was given as to how snapshots work – whether they use COW, log type writes (like Netapp), or some other method.

3) A system comprises a full 120 SATA drives. It’s not sold in anything smaller. That is 120x1tb drives, giving 51tb usable. The 120 drives are arranged in 8 DataModules of 15 drives each. You can see that the usable space (mirrored) leaves 18 drives unused. This is the equivalent of one full DM plus 3 more drives. Multiple system can be interconnected via 10GE, presenting a single system image/view.

4) Microcode upgrades to Data Modules can be accomplished by a full outage (all at once), or, a slow rolling process where data is migrated off a DM, the DM is upgraded/restarted, then data is redistributed back.

5) The host view of the XIV is a full active/active array (a host I/O can be sent to any IF (interface module) at any time. This is because the cache is in the DM. IF modules are basically routers. I did not think to ask how cached writes are protected on a DM, which are commodity PC servers. It will make use of the native MPIO features of the host operating system.

6) Positioning – this was strange. IBM has been telling the media that the XIV is for Web 2.0 applications, but in the presentation this was never mentioned. Instead, all the talk was about TIER 1 applications. They specifically mentioned that this is targeted at DB environments, Mail/Exchange, and VTL. This system is being squarely aimed at T1 level transaction processing systems.

I asked about this specifically and was told that the Web 2.0 positioning was to keep the system “flying under the radar”. The DS8000 is going to be relegated to mainframes. The N series (Netapp) will still be around. The DS4000 line will still be around, but positioned for less than 25tb while the XIV would be for greater than 25tb. (DS6000 was not mentioned).

7) It will support sync and async remote replication.

8) Performance. This was the strangest thing. Performance was mentioned over and over – It’s a high performance system!!! But, there was no data shown to back this up. All questions asking about performance specifics were talked around and not answered. The only “hard” data was a couple of graphs of one companies home grown benchmark. One graph was labeled “write miss” and showed 25k writes at 3ms response time. The other graph was labeled “read miss” and showed 25k reads at 2ms response time. Several people asked if they would publishing standardized benchmarks. They danced around this question giving no direct answer. The distinct impression was that really did not want to do this.

Summary: I would summarize the meeting as follows:

The XIV is a cluster architecture, built out of commodity components, glued together by gigabit ethernet,uses SATA drives, with claimed performance of a DMX or DS8000. The discussions I heard sounded like people thought the XIV is a neat array, but the performance claims are hard to swallow without hard data.

Excellent summary and thanks to the reader for this intel.  

{ 4 comments… read them below or add one }

elopescardozo July 1, 2008 at 3:30 am

Thin Provisioning with reclaim of “zero” blocks.

This sounds a lot like the schema I proposed earlier. It starts with the idea that in the interest of data security, newly allocated storage blocks would have to be wiped in order to prevent leakage of information from previous use. Wiping is a expensive operation, so can we avoid it? Yes, if we keep track of which blocks have been written to by the current owner of the space. New blocks would be flaged as virgin. If a host tries to read a virgin block, we don’t read the block but create a block of zero bytes in cache and hand that over to the host. When the host writes to a block, we clear the virgin flag.

The next step is that realization that when a block of zero bytes is written, we can avoid the write by simply setting the virgin flag for that block. This is completely transparent for the host. We would have to check each block for all-zero’s but in most cases we’ll find a non-zero in one of the first few bytes, so that operation is fairly cheap, even in software.

The final step is the realization that in the context of thin provisioning system, we can de-allocate a block when we set its virgin flag. So now we have a simple mechanism to let the host tell the storage system that it does not need a block: just write a zero block to it. Note that this operation a fairly quick, since the storage system never executes a true write, it just flips the virgin bit and if appropriate, de-allocates the block.

All you need to automatically de-allocate unused space is a simple utility running in the background on the host that writes zero’s to unused blocks in the file system.

I don’t know if this is what IBM does in their XIV system, but I’m sure more people have thought of this schema, so expect this to pop up in other storage systems as well.

joshuasargent July 1, 2008 at 1:28 pm

I’d be interested to hear an update from actual IBMers in Tucson about the market positioning, rather than the XIV sales people (mostly ex-EMC sales people). I spoke with Tony Pearson a couple of months ago and this report contradicts everything he said about its position in the IBM storage portfolio. I seriously doubt that the DS8000 will be relegated to mainframes only. Of course I could be completely wrong, but I wouldn’t put it past a sales guy to pull something like that out of thin air. Just because XIV’s original intent was to target Tier 1 applications doesn’t mean that IBM has the same vision. I don’t have a hard time accepting most of the performance claims (although hard data would be nice)…I’m more concerned about availability, particularly if I’m thinking about putting my Tier 1 applications on it.

Good grid-based systems are designed to take major hits and keep running. For instance, the Data Modules are not a single point of failure in the XIV system…you could lose an entire DM with 15 disks and not lose access to your data. Everything still runs because the data chunks are scattered redundantly across all the DMs in the system. But what happens if a SINGLE disk drive fails in one of those other DMs while the failed DM is rebuilding? What’s the performance impact of a failed DM? Questions like that are the kinds of questions you would be asking if you were thinking about using this for Tier 1 applications. Maybe there are acceptable answers to those…it’s too early to tell. Very little in the way of technical detail has been published about the system to date.

Administrator July 1, 2008 at 2:22 pm

This just in from a reader who wants to remain anonymous…


I have a comment for your DrunkenData blog that I would prefer to remain anonymous, since I work for an IBM partner…

There is considerable consternation among the IBM Business Partner community regarding these sales “sessions” being put on by the XIV sales reps. Some of my customers have attended and told me some of the ridiculous claims being made that completely contradict IBM’s messaging to the partners and to the public.


1. It will replace the DS8000.
2. IBM is targeting Tier 1 applications with it.
3. It’s got the functionality of the SVC built in.

Of course this conflicting information confuses customers and causes undue disruption of normal sales cycles, pissing all the partners off. I know you probably don’t care about resellers being pissed off, but the aggressive (misleading?) sales tactics of the XIV team are hurting customers because there is so little in the way of technical detail available to help the customer decide for themselves. I would caution customers not to give those guys the time of day without full up-front technical disclosure and a consistent, clear direction from IBM. Don’t get me wrong….I think the XIV gear is going to be impressive – IBM just needs to be clear and stop playing both sides of the fence.

Again, if you decide to mention this, please keep my name confidential.  

Done. Done. And Done.

az990tony July 11, 2008 at 5:18 pm

Since my name was mentioned above, I thought I would provide a response. I was not in the room when the overzealous IBM sales rep spoke, but for the most part the current XIV product is sold only in 120-drive frame increments with 51TB usable capacity. Snapshots work like our N series using a log-structured-array approach. Our DS8000 and N series will continue to be our “wall-to-wall” general purpose machines that can be tuned for a variety of different workloads, and IBM XIV will provide incredible price/performance for specific “area rug” workloads, including Web 2.0 and digital archives, databases with small block random I/O and a few others. IBM will publish performance numbers once they are ready.
— Tony Pearson (IBM)

Previous post:

Next post: