Why are we constantly told, "There's a disk space issue"?

Question

This morning, I was helping a BA with a particularly complicated report. I notice that in the reporting Development environment, we were only seeing a single day of data. Not only that, but the day is from a period of time where the data, itself, isn't particularly useful or indicative of what Production looks like.

I dared to ask, "Why isn't there more up-to-date information in the Development environment?" The BA responded, "We can't update Dev because there's a space issue."

Throughout my big boy career, at a handful of different firms, I've constantly heard that we don't have enough disk space. We can't run that report; not enough disk. We can't archive for more than a quarter; not enough disk. Everyone, drop what you're doing and clear your workspace on the network; we're almost out of disk space. Etc, etc.

What gives? I can get a stack of 1TB drives for about a grand, and even my personal computer at home has a rudimentary RAID. Is all NAS that expensive, or am I simply on a career path where I'm hitting a cluster of misers or PHBs?

What am I not understanding, here?

Thanks,
IVR Avenger

Answer 1

What gives? I can get a stack of 1TB drives for about a grand, and even my personal computer at home has a rudimentary RAID. Is all NAS that expensive,

I run into this exact question quite a lot internally when we tell users how much it costs to add more space to their allocation. It is eye-bleedingly expensive on one of our arrays, and this is a major, major problem. However, it is a forklift upgrade (6 digit replacement cost) so won't happen in the next 12 months. This is a 'miser' failure-mode.

I did run the cost-per-GB numbers on a pair of our arrays on a blog post of mine from two years ago (link if you want to read the whole thing). On one array, the cost-per-GB was a whopping $16.22, on another array it was $3.03. Once you factor in backup and disaster recovery coverage costs into the deal the numbers rose to $26 and $7.50 respectively. This is for Fibre Channel attached enterprise storage designed to service a lot of I/O requests at the same time.

I just finished writing the spec for another new array system that is targeted at a cost-per-GB of about $2/GB (not including backup/DR coverage). This is for an iSCSI/NAS device. The lack of fibre-channel makes it significantly cheaper, but we pay for it in the form of less resilience in the actual array. It can pump out the I/O operations, but it simply won't scale as far as the FC-based ones.

When figuring out the cost of storage per GB, you have to keep in mind that data is duplicated any number of times. RAID level impacts this. How deep your Backup infrastructure runs is a significant cost-factor (if you have to keep 7 years of data, and you keep one full copy each week, your data is replicated potentially 365 times. That's a lot of tape, and is why Deduplication vendors charge what they do). The storage infrastructure directly supporting the actual hard drives needs to be amortized across all of the storage contained in that infrastructure. Plus any support contracts needed to make it all supportable.

What further adds to the misery of big-storage needing departments is that the storage vendors are the LAST to roll out the big drives. Those 2TB SATA drives you've been seeing for quite a while now are only now beginning to roll out from the likes of HP, EMC, and Dell. This is due in large part to those brand new storage tiers not having enough usage history behind them to bet the farm on (or put another way, accurately price the support contract costs). That cheap array I recently built is based on 750GB 3Gb SAS drives, not 2TB SATA drives.

Answer 2

What are you not understanding?

1. Storage capacity has increased at a pretty fast exponential rate, a bit faster than CPU speed, memory density, etc... (doubling every 15 months instead of 18)
2. Storage speed has lagged, with latencies slowly creeping down. Storage for a single user system is cheap, but for a multi-user system (server), you're likely to need just as many drives as you did a few years ago. RAID parity calculation for the more space-efficient forms of RAID gives you a second bottleneck for writes.
3. That storage needs to be managed, which is a human time problem
4. That storage needs to be backed up, etc. This multiplies the cost of enterprise storage significantly above the cost of a single drive, and also adds human cost.
5. They probably bought all that storage more than a year ago, so the cost per TB was double or more the current cost.
6. They're probably buying the higher end drives (for many good reasons and some questionable reasons), which cost more than the cheap 1TB drives you're talking about.

In my experience, we run out of I/O to storage before running out of free blocks on disk. It's easier to answer in simple terms like "we're out of space" than to explain something like "we're at the 50% mark of I/O capacity to the storage we have and don't want to go past that in order to maintain adequate headroom for usage spikes".

Answer 3

Both. First, enterprise data is not about a RAID. It is aboutr a RAID and backups further down, and there it gets expensive.

Second, most companies do not understand slow and large, for their own stupidity. Heck, I can get a special disc cage (5 rack units high) that fits 48 drives. I put in larg 3.5" drives and I am safe. But most larger companies talk WAN, fast, 2.5" ONLY.

Third, it mostly is laziness and saving moeny on teh wrong end, too.