One thing I'm hoping for in our third generation fileservers

September 23, 2015

If all goes according to my vague schedule, we should be at least starting to plan our third generation of fileservers in 2018, when our second generation fileservers are four years old. 2018 is not all that far off now, so every so often I think a bit about what interesting things might come up from the evolution of technology over the next few years.

Some things are obvious. I certainly hope our entire core network is reliable 10G (copper) Ethernet by 2018, for example, and I optimistically hope for at least doubling and ideally quadrupling the memory in fileservers (from 64 GB to 128 GB or 256 GB). And it's possible that we'll be completely blindsided by some technology shift that's currently invisible (eg a large scale switch from x86 to ARM).

(I call a substantial increase in RAM optimistic because RAM prices have been remarkably sticky for several years now.)

One potential change I'm really looking forward to is moving to all-SSD storage. Running entirely on SSDs would likely make a clear difference to how responsive our fileservers are (especially if we go to 10G Ethernet too), and with the current rate of SSD evolution it doesn't seem out of the bounds of reality. Certainly one part of this is that the SSD price per GB of storage keeps falling, but even by 2018 I'll be surprised if it's as cheap as relatively large HDs. Instead one reason I think it might be feasible for us is that the local demand for our storage just hasn't been growing all that fast (or at least people's willingness to pay for more storage seems moderate).

So let me put some relatively concrete numbers on that. Right now we're using 2 TB HDs and we have only one fileserver that's got more than half its space allocated. If space growth stays modest through 2018, we could likely replace the 2 TB HDs with, say, 3 TB SSDs and still have growth margin left over for the next four or five years. And in 2018, continued SSD price drops could easily make such SSDs cost about as much as what we've been paying for good 2TB 7200 RPM HDs. Even if they cost somewhat more, the responsiveness benefits of an all-SSD setup are very attractive.

(At a casual check, decent 2TB SSDs are currently somewhere around 4x to 5x more expensive than what we paid for our 2 TB HDs. Today to the start of 2018 gives them two years and a bit to cover that price ground, which may be a bit aggressive.)

Written on 23 September 2015.
« The Go 'rolling errors' pattern, in function call form
We're probably going to need new Linux iSCSI target software »

Page tools: View Source, Add Comment.
Login: Password:
Atom Syndication: Recent Comments.

Last modified: Wed Sep 23 02:04:19 2015
This dinky wiki is brought to you by the Insane Hackers Guild, Python sub-branch.