It's time once again to look back on the storage predictions I made a year ago and look ahead at what might be in our storage future for 2008 and beyond.
We'll start with a review of how I did last year, which for me is always an interesting exercise because I get to see how the market measured up to my perceptions. We'll break it down into hardware and software, then look at where the storage market might be headed next year and beyond.
Last Year's Software Predictions
Last year I wrote that "Object storage will be available from the host side in 2007, and to storage devices, including object managers that replace RAID controllers, but not until 2008. This will be the first major change in storage technology end-to-end for many decades. This will have impacts on areas such as HSM, write reconstruction and virtualization, just to name a few."
Boy was I off on that one. There is no host-side object storage file system or device drivers, nor will object storage managers replace any RAID devices in 2008, mostly because there is no host-side object storage environment except for Panasas. This is a chicken and egg problem. Without vendors' development in the data path, this will not change.
Last year I said that POSIX file system semantics "will be changed to allow for things like lazy metadata updates for shared file systems in 2008 (see Solving the I/O Problem). This is a pretty easy prediction, but it still needs ratification. The standards are over 25 years old at this point and we need to update these standards."
This process is currently moving forward, and there is a reference implementation. The group is moving more slowly than expected, but the movement is still in the right direction. I have not gotten this prediction wrong yet since it was for 2008, but I will likely be wrong given the current slow pace.
I said that metadata allocation and management for file systems "will begin to be seen as a problem in 2007 and will be addressed in 2008. Areas such as databases and other additions for file systems to address the problems of ls–l and find and the impact on file system performance."
A number of vendors have started to address this problem and are using databases for metadata management, so I'll put this one in the win column.
"Storage management will not get any major standards, and managing thousands of disks and RAID controllers is not going to be easy, much less standardized," I wrote. "The vendors cannot get together to support all of the layers of management and reporting. The global tools that exist today do not show all the tunable parameters or all the details various error conditions."
Yes, SNIA has released SMI-S, but many vendors are still not completely compliant, and more importantly, information such as per disk SMART (Self-Monitoring, Analysis and Reporting Technology, which provides a common framework to collect disk error information) data cannot be collected. Some vendors disagree and claim they provide everything, but SMI-S is not rich enough to provide everything. Remember that SNIA is a group of vendors that have to agree and play nicely together in the SAN box. It is hard enough sometimes to get interoperability (this is better than it was five years ago but still a problem for new technologies), much less agreement of management and error statistics. I got this one right.
I got one very wrong, another one will likely be wrong in 2008, and two correct. Not all that good this year at 50 percent.
Last Year's Hardware Predictions
My prediction from last year: "Disk drive densities will continue to slow both in 2007 and 2008. This is an easy prediction. Drive vendors will push SAS drives as an alternative to both Fibre Channel and SATA, given the SAS characteristics, in late 2007 and 2008. Given the cost and the reliability, SAS 2.5 inch drives will make their way into midrange and enterprise RAID controllers in 2008. Database index file and file system metadata need high numbers of IOPS, and 2.5 drives are the answer."
I think this is all true and everything is on track for 2008. One right.
I said that vendors "will announce support for tapes over 1TB in late 2007 and release in 2008. Tape densities and performance will continue to grow at a rate faster than disk for at least a few years."
Close but no cigar. IBM's TS1120 is now 700GB and LTO-4 is 800GB, so I was wrong on this one. One of these days...
I said that RAID controllers from high-end enterprise and midrange vendors will support 10Gb Fibre channel, InfiniBand and 10Gb Ethernet in 2008. "The first two parts of the prediction are easy, but I am going out on a limb for 10Gb Ethernet in 2008," I wrote. "I believe that RDMA technology will become standard in the enterprise."
The introduction of 10Gb Ethernet in 2008 is a stretch, but still likely from some vendors. The introductions of FCoE (Fibre Channel over Ethernet) from T11 will make this happen much faster than for the broad market after 2008, and this might be the final blow that prevents iSCSI from market penetration. This one looks like it will be wrong for 2008 but not 2009.
"With NFSv4 and pNFS, you will see the convergence of NAS and SAN devices because of high-performance Ethernet connections such as 10Gb Ethernet," I predicted. "The complexity of management of large SANs is not easy, nor is the management of huge numbers of NAS devices. Given the performance limitations with NFS, NAS was not scaling to the performance of SANs and many NAS file systems were designed around NFS. If the NAS vendors rewrite and/or update their file systems and use NFSv4 and pNFS, they might achieve the performance of SANs. Combine this with object storage, which is a bit farther out, and the world we know might change."
The NFSv4.1 standard will integrate pNFS. The NFSv4.1 standard is expected to be published sometime in early 2008 and be put out for ratification. This is on track, along with 10Gb Ethernet, to change the world, so another one in the win column.
I predicted that big switches for both FC and IB running at rate to all ports full duplex would be announced in late 2007 and shipped in 2008. "Right now, large switches are limited to less than around 300 ports," I said. "I think this number will at least triple given the need for consolidation."
Cisco has a 528 FC switch, but it is not full rate full duplex any port to any port. On the InfiniBand side, Cisco, QLogic and Voltaire are all are stuck at 288 ports, so I got that one wrong.
I said that enterprise drive vendors "will begin discussion of other interfaces besides FC. This will not happen till 2008. The cost of FC interface per drive compared to commodity parts such as Ethernet might force drive vendors to consider other technology."
I got this one right. Enterprise drive vendors are going to SAS, at least for now. With FCoE, this might change. I am not yet ready to predict when, but it is coming.
I predicted that PCIe bus interfaces performance will grow faster than memory bandwidth. Right now, it is reasonably standard to have 8 Lane PCIe buses, which is about 2,000 MB/sec, I noted; given that memory bandwidth on most single boards is 6,400 MB/sec, the ratio was 3.2 to 1. "With 16 Lane buses becoming available, this might be an easy prediction," I wrote. "It is easier to build a bus that can move data at high rates than to be able to read or write it to and from memory."
I got this one correct. PCIe 2.0 is now out with double the performance. Memory bandwidths have not doubled in the last year.
I got four right, two wrong and one likely wrong. I'll give myself 57 percent — great for a baseball player or point guard, but pretty bad for a storage pro.
Predictions for 2008 and 2009
And now it's time to look ahead. Some of these are easy calls and some are a stretch.
- By 2009, at least one additional vendor will support T10 OSD file systems (see Let's Bid Adieu to Block Devices and SCSI). Having OSD file systems will likely allow better scalability over most current block-based file systems.
- Multiple implementations of NFSv4.1 will be available by early 2009 with pNFS support. This is a standard to watch, since it has the potential to change a significant portion of what happens in large environments.
- No significant changes will be available for error management in large system configurations. Though we need a big change in error management, there will be none forthcoming. This is an easy prediction, unfortunately.
- Undetectable errors will dictate more software changes in the data path beyond T10 DIF and Sun ZFS. People are starting to realize that undetectable errors are a bigger issue than they initially thought (see When Bits Go Bad).
- PCIe 2.0 (5GB/sec) will be come available in servers with AMD and Intel CPUs.
- By the end of the year, SAS drive shipments will exceed FC drive shipments for new systems.
- SAS 2.5 drives will become the standard for enterprise drives by early 2009, with shipments exceeding 3.5 inch drives (this is a bit of a stretch, so we'll see how it turns out).
- 8Gbit FC will enter the market for both HBAs and switches late in 2008. This will have a limited impact in 2008 and might have an effect in 2009.
- Tape density will finally hit 1TB uncompressed in 2008. If you predict it every year, you'll eventually be right.
- Disk drive density will continue to grow, but the growth will continue to slow. This is an easy prediction unless there is an unforeseen technology breakthrough.
- Flash technology will begin to be integrated into enterprise environments to address the IOPS per watt issue. I'll write more on flash drives in an article next year.
Those are my predictions for 2008 and 2009. I didn't do much better than a coin toss last year, but hopefully the industry won't let me down this year. Happy holidays, and best wishes for a prosperous 2008.
Henry Newman, a regular Enterprise Storage Forum contributor, is an industry consultant with 27 years experience in high-performance computing and storage.
See more articles by Henry Newman.