Hacker News Reader

Comment by callmeed | Hacker News Reader

It would be interesting to go back through the Everpix post mortem docs[1] and see if a big price swing like this could have put them in the black (or close to it).

[1] https://github.com/everpix/Everpix-Intelligence

View on HN

That's no different than startups of the past having to pay thousands to millions of dollars to setup servers. These days those same servers can be setup in minutes for a fraction of the cost.

Timing is everything. Sometimes the current market pricing for a commodity is too expensive to make your business viable today but in the future that will not be the case. Just depends how long that will take.

Fully agree. Most of what makes any startup work in the first place is that time + money + technology have finally aligned for them to start their idea.

If S3 storage pricing was the issue for this startup they could always have gone with their own dedicated servers for storage for a fraction of the cost. For that reason, I doubt a lower S3 cost would have prevented them from shutting down.

Fraction of the cost long-term. Not short term, where they found themselves.

Renting dedicated servers with lots of disk has typically been vastly cheaper than S3 "forever" on a monthly contract basis, so there are plenty of options before going the route of bare metal in your own rack.

But even renting space by the 1U and leasing servers directly adds surprisingly little overhead.

http://blog.backblaze.com/2014/03/19/backblaze-storage-pod-4...

I'm not sure where you get that data, but it doesn't match anything I've found in the 27 dates centers we do business in. Storage, versioning, redundancy, durability and geographic availability. That is the core of S3, not just storage.

The data centres makes little difference in this case. You can fit far more storage than what Everpix used comfortably in a half rack (e.g for density I can buy 5U servers storage servers with up to 48 x 4TB drives). In London, that runs me at most about $1k/month per site. Three sites for $3K/month. 400TB worth of off the shelf storage servers would run us ~$1600/month without even shopping around, in leasing cost per location, for ~$8000/month for triple redundancy on top of raid. Then there's bandwidth.

400TB even at the Reduced Redundancy Storage pricing is ~$20k/month up until the price drop. With the price drop it looks like it'll be more like $9300 with RRS, which is actually getting cheap enough that it's worth it in some cases, to avoid having anyone spend their time on managing anything. The gap will increase again once you factor in request pricing and bandwidth, though, as even my standard price list prices for bandwidth would add up to less than the cheapest published bandwidth tier at Amazon with just a tiny little bit of hassle (BGP setup to multihome each sites with 2-3 different transit providers with different coverage - lowers cost by letting you pick a transit deal that does not give full access to the internet at very low cost coupled with a full price full coverage one)

Or I could spend about $12k/month at Hetzner for the same storage + an aggregate stated "guaranteed" bandwidth of about 2Gbps per location at no extra cost. Again, without shopping around. The moment you do more than about 20-30TB transfer/month, Hetzner becomes cheaper than S3 with that setup, and probably at about 50TB+/month it'd take effort to beat them when buying my own transit too.

You're leaving out the part about the software and operational support to handle failover, data integrity checks and correction, etc. You're not just tossing disks in a rack but managing a highly-available, highly-durable storage system on top of them.

You can definitely beat those rates at large enough scale but for a startup you're talking about dedicating a non-trivial chunk of your skilled technical staff's time in the hope of breaking even. That's an obvious move for Backblaze or Crashplan but for other companies it might be a dubious investment versus nailing the core product.

Not sure how you can compare your own servers to s3 - s3 has redundancy built in that can survive the loss of 2 data centers. I'm pretty sure it would be more expensive to roll your own version of that...

See my comment elsewhere with some numbers. You can beat S3 even with the new pricing in many cases. With the old pricing its trivial.

With replication to 3 data centres.

There is no way to replicate S3s full service level at a fraction of the cost. You can store the data pretty cheaply, but you can't replicate the full service level anywhere near as inexpensively. Especially the redundancy, vesioning and durability.

It might have been enough for them to keep investor interest [new round of funding] and continue. It wouldn't have put them in the black.

https://github.com/everpix/Everpix-Intelligence/blob/master/...

From that, it looks like their loss over 26 months was ~$300,000. Their total AWS costs was less than $400,000 over the same period.

Downloading 1 GB = 4x the cost of the S3 storage at the reduced prices. If everything is viewed once [doubtful], that is 4x the price of the S3 storage. Given that ti would be impossible to reduce the total cost by even 50%, let alone the 75% they'd need to break approximately even.

I didn't see S3 broken out specifically anywhere which is why I'm just making an educated guess. :/

very cool repository, found the S3 data here:

https://github.com/everpix/Everpix-Intelligence/blob/master/...

So basically at their peak, they had ~174TB S3 and ~47TB S3 RRS. That month cost them ~$16.2k, whereas after April 1 they could have put everything in S3 for ~$6.7k. That's a big difference, but they were still getting taken to the cleaners over some other services like RDS.

Cool. So they had roughly (with rounding):

RRS: 2+3+6+8+9+18+19+22+27+34+42+48+7=~245 TB/months=~

Normal: 7+9+13+19+24+46+56+65+76+95+123+153+174+25=~885 TB/months

Difference:

$0.085-$0.0300=$0.055

$0.068-$0.0240=$0.044

End:

0.055 * 885 * 1024=$49,843 0.044 * 245 * 1024=$11,039 $49,843+$11,039=~$60k saved which is 20% of what they were short, if they had started with April 1st's pricing.

If you can reduce the AWS $400k by 40% due to the reduced pricing for EC2, you'd still fall short.

Why wouldn't they put all of that in a Backblaze Storage Pod? 2 months of AWS costs could've built more than 2 pods and stored everything with plenty of space to spare.

Price out the backup solution for that back blaze. Then make the backblaze geograohiclly redundant. Then write the system to keep them in sync. Then add versioning. Then pay for someone to manage it. Now you are getting towards the real cost of that backblaze to replace S3.

>Then write the system to keep them in sync. Then add versioning.

Are there any ready-to-deploy solutions for this?

I had the same thought. Their last Amazon bill was $35k. I don't think saving $10-20k/month would have made a significant difference. They had many freemium users, and were not adding customers at a sufficiently high growth rate over the life of the company. They were trying to raise $5m.

The importance of this depends on unit economics. For businesses with historically tight margins (e.g. retail), small swings in input prices (labor, cost of goods) can make or break profitability.

Whereas software, if you're doing it right, should have much higher margins.

Sidenote: I'd be very curious to know what the "typical SaaS business" has as its gross margin, factoring in all sales against "cost of goods": servers, bandwidth, operations, etc.

80% seems to be the common 'educated guess' number:

http://www.quora.com/What-should-be-the-typical-gross-margin...

http://www.forentrepreneurs.com/saas-metrics-2-definitions/

http://www.forbes.com/sites/jjcolao/2012/10/02/running-a-saa...

Its one of the factor, but it's not the only thing which can put them in the black.

A miss is as good as a mile.

Reply on news.ycombinator.com