SNW Spring 2006 San Diego

I returned from the Spring Storage Networking World a few weeks ago. There are a few topics that have been on my brain ever since….

1. Is ILM worth it?
I realize ILM has means different things to different people. When I mention ILM, I am referring to the automatic categorization and migration of data between data tiers. ILM and HSM continue to be written about more then they are deployed. While ILM like solutions appear to be great in theory, actual deployments are few and far between. In talking with customers who have deployed solutions, there are several challenges:

1. Categorizing data
2. Complex management and deployment
3. Product maturity.
4. Support.

At this point, I am taking a wait and see approach. As an infrastructure Architect I design systems scale over hundreds of TB's. We have demo'd and tested a few ILM/HSM like solutions. Currently it is cheaper and much easier to manage without ILM.

In the mean time I am keeping my eye on a few up and coming vendors in this space,
1. Arkivio http://www.arkivio.com/2/splash2.asp they deal with mostly unstructured data such as media and document repositories.
2. Local Alpharetta Georgia company Scentric has recently released an ILM product for structured data (http://www.scentric.com/ ).

2. NAS Benchmarking

Spec SFS is highly regarded as the premier NAS benchmark in the industry. The spec organization sells their benchmark to customers and assists them in running the benchmark. When searching for a NAS solution to store large amounts of Rich Media storage, I ran across several vendors who tout their spec SFS benchmark numbers. Since our usage pattern may not match a particular benchmark, I decided to research the spec SFS benchmark before taking these results into account. One of the big things that I noticed is the file size distribution:

TABLE 2. File size distribution

Percentage	Filesize
33%	1KB
21%	2KB
13%	4KB
10%	8KB
8%	16KB
5%	32KB
4%	64KB
3%	128KB
2%	256KB
1%	1MB

This distribution is nothing like load that my projects will put on a system, and honestly what does this workload match? The file size distribution is much too small for general file serving. I decided that this benchmark is out of date, and not applicable for my environment.

I have used several other benchmarking tools in the past, and for one reason or another they all seemed to be flawed. This led me to email the spec committee to express my thoughts on the subject (like they care). Surprisingly, I received several emails back from spec committee members. There seems to be a consensus that the benchmark is outdated, and not of much use for real world scenarios. Ok…what next? I can’t get a straight answer. From what I can tell the spec committee is made up of vendor representatives that all have a stake in the game. I don’t see a consensus on this matter in the near future.

There are a few benchmarks for NAS systems that show promise. 1. IOZONE (www.iozone.org). Iozone appears to be very configurable and scalable. It is also more complex then several other benchmarking tools, and it does not have a GUI interface.

2. NSPLABS NASBENCH (www.nasbench.com ) NSPLabs is creating what appears to be a great product. One unique feature is a workload recorder. You will be able to record your real workload, and replay it in various scenarios.

For now, I will continue to use IOMETER and I will hopefully ramp up on IOZONE. I am ultimately waiting for the NASBENCH product to come out with a beta soon.