News for hackers. Stuff that matters.
Re:Alphabet Third Quarter 2021 Results [pdf] by easton on Tuesday October 26, 2021 @04:34PM (#29005899)
> In January 2021, we completed an assessment of the useful lives of our servers and network equipment and adjusted the estimated useful life of our servers from three years to four years

Does anyone know what the observation was here that made this possible? Have Google's internal workloads not grown in size as quickly as they thought, or is this a "Moore's law is slowing down" thing?

Re:Alphabet Third Quarter 2021 Results [pdf] by kyrra on Tuesday October 26, 2021 @04:37PM (#29005930)
Googler, opinions are my own. I know nothing about this.

I wonder if this was Google's response to the chip shortage? Maybe the cost/benefit (and just sheer inability to get certain parts) made them hold on to hardware for longer.

Re:Alphabet Third Quarter 2021 Results [pdf] by bduerst on Tuesday October 26, 2021 @05:14PM (#29006333)
Could also be the fact that Google has built a ton of new data centers in the last two years. Newer hardware, infra, utilization, etc. could lead to better quality/lifespans.
Re:Alphabet Third Quarter 2021 Results [pdf] by DennisAleynikov on Tuesday October 26, 2021 @04:39PM (#29005957)
That would be an interesting adjustment
Re:Alphabet Third Quarter 2021 Results [pdf] by slo_news on Tuesday October 26, 2021 @04:57PM (#29006152)
While I'm unsure of the the reason the computers are lasting longer, I would bet this observation was originally triggered from the public company accounting firm that audits Alphabet's financials. The accounting firm likely asked Alphabet to prove out why they used a 3 year estimate and found out 4 years was more accurate (FYI 3 years is a standard useful life for computers for most companies - sometimes even shorter).
Re:Alphabet Third Quarter 2021 Results [pdf] by hogFeast on Tuesday October 26, 2021 @05:18PM (#29006376)
It doesn't work this way.

If Alphabet wanted to prove they lasted three years, they would have found a way to prove it. Auditors don't go around challenging on things like this. In particular, they don't go around challenging companies on depreciation because that is a key lever that management can pull if they need to hit numbers.

It is very difficult to discern exactly why they will have done this but the functional answer is: they have done this because they needed earnings to increase. In this case, it looks like the difference between a 15% and a 20% beat (I believe the change was made in January, I think they had quite a big beat in that quarter after some flattish momentum, middle of the year very strong anyway).

Re:Alphabet Third Quarter 2021 Results [pdf] by slownews45 on Tuesday October 26, 2021 @06:14PM (#29007034)
Umm, intel flatlined basically on chip development. Chip shortages also hit the sector. And the long tail has gotten to point where folks keep systems in workloads even if "better" solutions exist - because existing systems are working fine. 4 years is not unreasonable (now). Once chip shortages ease and development picks up again at intel useful life may decrease again.
Re:Alphabet Third Quarter 2021 Results [pdf] by slo_news on Thursday October 28, 2021 @04:19PM (#29031361)
But sometimes it does.

I worked in audit for several years and the auditors will challenge any estimate made by management that the auditors deem material to the financial statement readers. And the useful life of how long computers last is an estimate by management. While at most companies useful lives might not be an estimate worth looking at, Alphabet has enough computers for this to be a material estimate (likely due to GCP) that is subject to management bias. The public auditors are there to ensure (to a reasonable extent) that management can't manage earnings by changing estimates like this.

Re:Alphabet Third Quarter 2021 Results [pdf] by ChuckMcM on Tuesday October 26, 2021 @05:23PM (#29006456)
No, computers typically have a service life of 5 years, but depreciating them faster gives you more loss to to write down against your income.

What you are seeing though is the trick they used to make 'net income' pop. Let's say they had a million servers and a server costs $5K each. When depreciated at 3 years that is $1.6 billion dollars a year of depreciation. But you stretch that out to 4 years and now its only $1.25 billion dollars in appreciation. Since depreciation is subtracted from revenue you just "bumped" up your revenue by $350M and you didn't have to do anything at all (except change how you treated your assets).

That they had to resort to this level of shenanigan to get their revenue numbers up is interesting to me.

Re:Alphabet Third Quarter 2021 Results [pdf] by gradys on Tuesday October 26, 2021 @05:53PM (#29006801)
> Since depreciation is subtracted from revenue you just "bumped" up your revenue by ...

No, you bumped your net income. Revenue is unaffected by the depreciation change. Revenue is up 61%.

Also, nobody would be fooled by an accounting trick like this. Analysts routinely compute EBITA, earnings before interest, taxes, and depreciation, exactly for this reason - filtering out the more purely financial/virtual expenses that are less informative for understanding the core business.

They likely did it because you're required to report things like depreciation in a way that reflects reality. There could be tax implications for instance, since you can count depreciation expenses against your taxable earnings (though often companies maintain separate depreciation accounting for financial reporting vs taxes due to the different rules for each).

Re:Alphabet Third Quarter 2021 Results [pdf] by ChuckMcM on Tuesday October 26, 2021 @06:36PM (#29007259)
Fair, it bumps net income not revenue.

When I was younger, I took an accounting class from Santa Clara University (it was part of their Executive MBA program although I did not get an MBA) because I knew that I'd like to start my own company some day and needed to know how accounting worked.

> Also, nobody would be fooled by an accounting trick like this.

The course specifically covered "accounting tricks" that would create the most favorable impression in the eyes of the public and in the eyes of the analyst community. Playing around with depreciation rates was one of them. The assertion in class (which I've never had a chance to test) was that the IRS rarely, if ever, cared about your depreciation tables. That was damn close to 20 years ago so it most certainly could be different now, and I am not a tax accountant. If it has changed since then I am sure someone who has taken the class will correct me.

I don't think it "fools" analysts, but I suspect they might trade on an understanding that retail investors might have a different take on what was reported than what the analysts read.

Re:Alphabet Third Quarter 2021 Results [pdf] by remus on Tuesday October 26, 2021 @05:57PM (#29006857)
Looking at it from a naive point of view, could it just be an actual change in what they estimate to be the useful life of their servers? Accounting shenanigans for a (relatively) small bump in revenue seems not really worthwhile.
Re:Alphabet Third Quarter 2021 Results [pdf] by jdavis703 on Tuesday October 26, 2021 @04:38PM (#29005948)
My assumption is the hardware was more reliable than expected. If it was about slowing workload growth one might expect them to switch to a usage-based (instead of time-based) depreciation schedule.
Re:Alphabet Third Quarter 2021 Results [pdf] by jeffbee on Tuesday October 26, 2021 @05:12PM (#29006313)
Uh, both 3 and 4 are radical understatements of the economic lives of servers. They are TCO+ for 10 years in reality. For a company like Google or any other company designing their own equipment the usual reason to retire equipment (end of warranty) does not exist.

There are still Sandy Bridge CPUs available in GCE. Sandy Bridge came out in Q2 2011.

Re:Alphabet Third Quarter 2021 Results [pdf] by sidibe on Tuesday October 26, 2021 @04:37PM (#29005939)
I don't think the workloads would have much to do with this, probably saw that they were having less failures than anticipated or memory/chip shortages have them keeping equipment around longer than they planned.
Re:Alphabet Third Quarter 2021 Results [pdf] by greenail on Tuesday October 26, 2021 @05:05PM (#29006253)
AFAIK it comes down to some degree to a tradeoff between the projected density and power efficiency gains of newer gear vs the cost savings of keeping older gear around and in the racks. Some datacenters simply can't support higher density due to power constraints and if the newer gear isn't offsetting the capital costs by pulling in more revenue then you might as well keep the old gear around longer.
Re:Alphabet Third Quarter 2021 Results [pdf] by jeffbee on Tuesday October 26, 2021 @05:26PM (#29006488)
I don't know why people always mention density. Google isn't constrained on floor space. If they need another 100 acres of Iowa for their servers to stand on, they'll just buy that for a comparative $0 cost.

Density is an issue for people who are still renting from Equinix.

Re:Alphabet Third Quarter 2021 Results [pdf] by greenail on Tuesday October 26, 2021 @05:28PM (#29006514)
the substation has a limit, so do all of the power distribution systems.
Re:Alphabet Third Quarter 2021 Results [pdf] by jeffbee on Tuesday October 26, 2021 @05:56PM (#29006841)
Sure but that's the opposite trend from density, is it not? A dual 64-core EPYC Rome machine - and we know they are using these for web search, from published research papers - has 400W of CPU TDP alone, which is a more than double the per-host TDP of a Westmere EP (as an example of a retired platform). So if anything there is more empty space in the racks in 2021 than there would have been in 2011, based on the design power load of existing buildings.
Re:Alphabet Third Quarter 2021 Results [pdf] by greenail on Tuesday October 26, 2021 @06:05PM (#29006955)
The cycle for servers is 3 years, the cycle for the building is 30 years. What I'm trying to say is that if you can deliver more widgets per watt per sq ft by replacing the server than you go for that. If you can't why not keep the already depreciated server operating for another year. The incremental costs are mostly the power (and ac) plus the opportunity cost of putting a more efficient server in a year later. If it fails you just put in the newer server. If it doesn't you may save a bit.
Re:Alphabet Third Quarter 2021 Results [pdf] by mattlondon on Tuesday October 26, 2021 @05:07PM (#29006272)
Probably because intel chips feel like they have stagnated somewhat performance-wise while they milked the market IMHO.

Why upgrade to the latest intel CPU when you get a paltry 100Mhz clock increase for the bother? May as well wait another year if 3 year old CPUs are pretty much the same performance as current models (give it take a few %)

Re:Alphabet Third Quarter 2021 Results [pdf] by cj on Tuesday October 26, 2021 @04:56PM (#29006143)
I wouldn't be surprised if the decision was driven by Google's army of accountants.

Extending the life of an asset would affect the depreciation schedule, which would affect when depreciation is recognized / recorded on their books in a given tax year.

Re:Alphabet Third Quarter 2021 Results [pdf] by green-eclipse on Tuesday October 26, 2021 @05:57PM (#29006853)
Re:Alphabet Third Quarter 2021 Results [pdf] by bpodgursky on Tuesday October 26, 2021 @05:00PM (#29006194)
This is probably less exciting than you imagine. It's hard to estimate HDD and CPU failure rates 5 years into the future, especially on new-generation hardware. 3 years was an informed guess.
Re:Alphabet Third Quarter 2021 Results [pdf] by DennisAleynikov on Tuesday October 26, 2021 @04:38PM (#29005952)
Seemingly the latter. Less benefit between generations and more stress resistance in hardware compared to earlier data center practice