Regarding OVH, their wide and unrealistic price fluctuations makes me think that it's being run by a madman. I do not care what a madman does.
But realistically, the price will most likely remain static throughout the life of the product, but what fits into the "slot" of that product would change as hardware prices evolutionize. Since CPUs (especially in the low-power sector) and hard drives and such are somewhat slow to develop and price-drop, I expect that even a yearly "refresh" of our offering may be too short.
At that point, if we do refresh our offering on a yearly basis, then that gives 12 months to pay off the 11 month hardware. If we discount the "old" stuff down to, say, $20/month, then that's fine too because we're no longer needing to amortize the hardware; it's just gravy.
As we're not going to swap drives between servers, if customers want to spend time and effort and money to buy and set up another server and move their stuff over to the new hardware offerings, then that's fine too.
1. The 11 month payoff sounds really optimistic. Even OVH with their $100M's of capitalization got themselves in a bad spot trying that. The hw depreciates too fast. I think 4 months or so is more normal for the budget sector, and less than 4 months for the higher end.
2. $150 for 256GB SSD's sounds like cheap consumer drives, maybe even 3-level flash. That also sounds shaky. The only reason I'd go for 256GB instead of 3TB at the same price is to run database workloads, which will pound the crap out of those SSD's and destroy them with write wear. I'd frankly rather have 120GB enterprise SSD's (Intel S3500) than 256GB consumer ones for the $150 if I'm going the SSD route. With only 4GB of ram I couldn't really run too large a database anyway. Are ram upgrades possible?
3. I like the suggestion of 2x1TB instead of 1x3TB as the default or as another option. In my case I'd want 2x3TB though. Actually why not 4TB drives for a few $ more? I see them for $170 on newegg right now.
4. If it's up to me, I'd rather not pay extra for 5TB bw instead of 1TB. I'll use far less than 1TB most months. There's a one-time download of around 3TB that I want to do that I mentioned, but I could spread it over several months, or I'd be ok with paying a reasonable overage fee if I had a very busy month. Even 5TB sounds way overcommitted at the price level.
5. How many drives physically fit in the server, i.e. is 2 the maximum? 4 would be nice.
6. Would you be willing to plug a small USB device into the server for me if I send it to you? It would be cig pack sized, plug in with a cable, and you'd just leave it plugged in. Or if necessary it could be like a flash stick, just plugged into the socket with no cable. Small extra cost is ok either way.
You have good input; keep it coming.
I'll itemize my responses on the same numbers as your questions
1. We really aren't in this business to see how poorly we can do at it... there keeps being repeated remarks of "I don't see how you'll make a profit!"; believe me, we run the numbers and have quite a few discussions about new products/ideas far before they ever get to a public forum.
2. The ~256gb SSDs I was looking at are "Intel 335 Series Jay Crest SSDSC2CT240A4K5 2.5" 240GB SATA III MLC Internal Solid State Drive (SSD)" which are indeed a consumer-level drive.
http://www.anandtech.com/show/6462/intel-explains-20nm-nand-endurance-concerns-on-the-ssd-335 has more information on the endurance, specifically:
"Kristian's review showed that to be roughly 250TB of writes, which means the actual value is aroung 500TB of actual NAND writes (incompressible). Doing the math on the 240GB capacity gives us 2083 full drive writes over the life of the drive, or about 5.7 years of useful life if you write 240GB of data to the NAND every day. Even if your workload has a write amplification factor of 10x, you're still talking about 24GB of writes per day for nearly 6 years."
I'm still waiting to hear from Intel if they'll replace an in-warranty drive when the MWI gets to zero.
The price spike for an "enterprise" SSD would probably mean that we would sell very few, if any. One of the pivots of this service is to keep hardware on-hand for when things break; adding additional drives into the mix arbitrarily increases costs, as hard drives (and other things) that we keep on hand for variation are hard drives that aren't being used for revenue. To use an automotive analogy, think of Toyota vs BMW vs Aston Martin: Toyota provides you a narrow range of interior and exterior choices but at a low cost, BMW expands the range but also has a wider profit margin, and Aston Martin will paint your new car in any PPG color that you want, But It'll Cost You.
As this service will not be a hand-holding type of dedicated server offer and that the user is fully aware of the service's requirements for proper usage, such as installing the OS yourself, it's hoped that if a user is choosing an SSD over a mechanical drive, that they have a full understanding of the implications of their choice.
For what it's worth, we hammer the bejeezus out of SSDs in various CacheCade arrays on LSI controllers, and @Wintereise hammers the bejeezus out of SSDs for database things, and neither of us are seeing catastrophic systemic failures. That's not to say, of course, that we'll have more failures versus mechanical drives, but it's already factored into our end of things.
2 (again). 4gb seems to be the limit of current-generation Atom boards ("boards" being the operative here.. the CPUs themselves can address more) , and there's not much of a price diff for 4gb of ram versus lesser amounts, so there's not much benefit to us or to our users to sell less-than the max.
3. It would indeed be 2x3TB or whatever the drive size ends up being. If 4TB drives end up being ~$150 when we move ahead, then we'll use those. Once again, too much variation will end up costing an excessive amount in the future.
4. The price increase was actually due to using a different form factor. I intentionally under-offered the original offer at 1 TB to see what kind of reaction was going to be received; 5 TB is a much more reasonable value anyway. Regarding overcommitment, there's a fine line between "we've sold too much and everything is going to shit" and "we've sold too much but we're adequately managing it"; proper management is paramount. Most of us have cell phone plans that provide us with xxx minutes per month but we only use yyy minutes per month. I can provide more information on actual bandwidth usage versus what we've sold on our existing services privately, if you have any interest.
5. 2 is indeed the maximum.
6. Sure, we'd have to charge remote hands though. We'd just charge whatever the datacenter's cost is for remote hands; there's no real need for us to arbitrarily mark it up.
So when can we purchase this server?
Sorry, had a bit of a crisis last week so I didn't really do much further research on this. I'm thinking it'll probably be January or February 2014 at earliest.
Would defintiely be interested. Have clients who would be interested. Unfortunately, no IPMI would be a deal-breaker. Absolutely no reason not to have it on a server. Accessing it via a VPN would be just fine.
Yeah we'll be able to offer IPMI. It's too much effort for us to manage OS (re)installs otherwise, so i'd rather have the infrastructure in place to do IPMI. There's an updated listing on the second page of this thread.
