TR Forums

Waco wrote:

just brew it! wrote:

MileageMayVary wrote:

The storage team at my work is setting up a 1 PB all flash NetApp cluster this year.

FWIW one of the managers at work mentioned today that one of our larger customers has a data "ingest rate" in excess of 1 PB per day.

(And no, the array isn't all flash in this case...)

One of my in-house designed systems averaged 1 PiB per day of real data for 3 weeks straight when a particular user unleashed hell on us. Singular user with a mission.

Another user who shall remain unnamed managed to nearly fill a pair of 40 PiB filesystems in a workday. Things move quickly at 1 TB/s+.

HOLYCRAP

MileageMayVary wrote:

HOLYCRAP

Big data means different things to different people.

Waco wrote:

MileageMayVary wrote:

HOLYCRAP

Big data means different things to different people.

Yeah, 80 PB in one day is pretty darned big by most people's standards!

Just take a moment to think about the kind of traffic and data capacity that the big Cloud service providers need to deal with, and you realize that there have to be an awful lot of datacenters with really fat pipes to the internet, and millions of SSDs and HDDs out there, all behind the scenes making our modern interconnected world function.

Furthermore, because of the crazy way copyright laws work, any service that provides "Cloud DVR" functionality (like Comcast and other cable providers do) must make a separate physical copy of all the digital bits for each user who records a program, in order to stay legal. Reading from a single shared copy to stream to multiple users, or using any kind of de-dup tech is not allowed. Essentially, they're relying on the same decades-old court rulings which carved out the copyright exception for time-shifting using home VCRs, but doing it "in the Cloud". So somewhere out there, scattered all around the world, there are entire datacenters dedicated to storing thousands of PB worth of redundant copies of TV programs...

just brew it! wrote:

Furthermore, because of the crazy way copyright laws work, any service that provides "Cloud DVR" functionality (like Comcast and other cable providers do) must make a separate physical copy of all the digital bits for each user who records a program, in order to stay legal. Reading from a single shared copy to stream to multiple users, or using any kind of de-dup tech is not allowed. Essentially, they're relying on the same decades-old court rulings which carved out the copyright exception for time-shifting using home VCRs, but doing it "in the Cloud". So somewhere out there, scattered all around the world, there are entire datacenters dedicated to storing thousands of PB worth of redundant copies of TV programs...

Exhibit 1 for why lawyers and politicians should generally stay away from anything relating to technology (that they don't understand the implications of).

If only that could actually become true.

Waco wrote:

Exhibit 1 for why lawyers and politicians should generally stay away from anything relating to technology (that they don't understand the implications of).

If only that could actually become true.

Yup. The notion that -- since individual VCRs are explicitly a loophole -- they need to exactly simulate the functionality of a per-user VCR in a remote server somewhere in order to remain legal is just retarded. What a waste of equipment and electricity. I wonder how much of a factor "Cloud DVR" has been in keeping Seagate and Western Digital in business as consumer HDD sales have tanked...

I imagine as far as the cable operators are concerned it is still a net win versus providing in-home DVR boxes, since they don't need to deal with swapping out a physical DVR at someone's home when the HDD goes bad. They're just replacing a drive in a server in their datacenter.

Waco wrote:

Another user who shall remain unnamed managed to nearly fill a pair of 40 PiB filesystems in a workday. Things move quickly at 1 TB/s+.

Let me guess. Your network fabric is built on vibranium.

MileageMayVary wrote:

HOLYCRAP

Check his reported location and think about what's been going on there since 1943 or so.

Captain Ned wrote:

MileageMayVary wrote:

HOLYCRAP

Check his reported location and think about what's been going on there since 1943 or so.

Social Security Administration? I don't work for them.

I do love seeing the scale of things that people work on.

MileageMayVary wrote:

Captain Ned wrote:

MileageMayVary wrote:

HOLYCRAP

Check his reported location and think about what's been going on there since 1943 or so.

Social Security Administration? I don't work for them.

SSA is older than that. Think "things that explode quickly" as opposed to "things that will implode slowly".

just brew it! wrote:

MileageMayVary wrote:

Captain Ned wrote:

Check his reported location and think about what's been going on there since 1943 or so.

Social Security Administration? I don't work for them.

SSA is older than that. Think "things that explode quickly" as opposed to "things that will implode slowly".

The department of fast energy.

holy crap is indeed appropriate here..

i'm seeing PiB and TB/sec... please wait while i wipe the drool from my desk..

Captain Ned wrote:

Waco wrote:

Another user who shall remain unnamed managed to nearly fill a pair of 40 PiB filesystems in a workday. Things move quickly at 1 TB/s+.

Let me guess. Your network fabric is built on vibranium.

Infiniband and OPA can push that much data around with some prodding. You're looking at supercomputer-class hardware there... but that does make sense for an environment with at least two 40-PiB filesystems.

No need for unobtainium or equivalent.

Infiniband, generally. We have some OPA and custom interconnects (Cray), but the majority of the storage networking is IB.