Likey buying - 800k for a 800sqft - sounds about right for major Canadian metros like Toronto/Vancouver

Yep, that's right. Translates to rent one way or another.

I don’t follow - what do you mean by this?

“The American Telephone and Telegraph Company (AT&T) was listed on the NYSE on Sept. 4, 1901. Only 11 other companies have been listed longer than AT&T.“

Yeah, impressively wrong gp! 1984 was when they were broken up by Justice Department.

FWIW to avoid the plugging in (which I hated) Calibre can be configured to send emails as well, which works well with the email to kindle feature, and with a little fiddling you can bridge the gap to have Calibre auto-email you. (https://www.mobileread.com/forums/showthread.php?t=314401)

Is it? The line preceding the bullet list on that page seems to state otherwise:

“”

  Each storage volume can deliver the following performance (all measured using 4 KiB blocks):

  * Up to 8000 MB/second of sequential reads

“”

Just tested a i4i.32xlarge:

  $ lsblk
  NAME         MAJ:MIN RM   SIZE RO TYPE MOUNTPOINTS
  loop0          7:0    0  24.9M  1 loop /snap/amazon-ssm-agent/7628
  loop1          7:1    0  55.7M  1 loop /snap/core18/2812
  loop2          7:2    0  63.5M  1 loop /snap/core20/2015
  loop3          7:3    0 111.9M  1 loop /snap/lxd/24322
  loop4          7:4    0  40.9M  1 loop /snap/snapd/20290
  nvme0n1      259:0    0     8G  0 disk 
  ├─nvme0n1p1  259:1    0   7.9G  0 part /
  ├─nvme0n1p14 259:2    0     4M  0 part 
  └─nvme0n1p15 259:3    0   106M  0 part /boot/efi
  nvme2n1      259:4    0   3.4T  0 disk 
  nvme4n1      259:5    0   3.4T  0 disk 
  nvme1n1      259:6    0   3.4T  0 disk 
  nvme5n1      259:7    0   3.4T  0 disk 
  nvme7n1      259:8    0   3.4T  0 disk 
  nvme6n1      259:9    0   3.4T  0 disk 
  nvme3n1      259:10   0   3.4T  0 disk 
  nvme8n1      259:11   0   3.4T  0 disk

Since nvme0n1 is the EBS boot volume, we have 8 SSDs. And here's the read bandwidth for one of them:

  $ sudo fio --name=bla --filename=/dev/nvme2n1 --rw=read --iodepth=128 --ioengine=libaio --direct=1 --blocksize=16m
  bla: (g=0): rw=read, bs=(R) 16.0MiB-16.0MiB, (W) 16.0MiB-16.0MiB, (T) 16.0MiB-16.0MiB, ioengine=libaio, iodepth=128
  fio-3.28
  Starting 1 process
  ^Cbs: 1 (f=1): [R(1)][0.5%][r=2704MiB/s][r=169 IOPS][eta 20m:17s]

So we should have a total bandwidth of 2.7*8=21 GB/s. Not that great for 2024.

So if I'm reading it right, the quote from the original article that started this thread was ballpark correct?

> we are still stuck with 2 GB/s per SSD

Versus the ~2.7 GiB/s your benchmark shows (bit hard to know where to look on mobile with all that line-wrapped output, and when not familiar with the fio tool; not your fault but that's why I'm double checking my conclusion)

If you still have this machine, I wonder if you can get this bandwidth in parallel across all SSDs? There could be some hypervisor-level or host-level bottleneck that means while any SSD in isolation will give you the observed bandwidth, you can't actually reach that if you try to access them all in parallel?

The aggregate throughput matches the advertised number of 22,400 MB/s: https://aws.amazon.com/blogs/aws/new-storage-optimized-amazo...

Can you addjust --blocksize to correspond to the block size on the device? And with/without --direct=1

I wonder if there is some tuning that needs to be done here, it seems suprising that the advertised rate would be this much off otherwise.

I would start with the LBA format, which is likely to be suboptimal for compatibility.

somehow I4g drives don't like to get formatted

    # nvme format /dev/nvme1 -n1 -f
    NVMe status: INVALID_OPCODE: The associated command opcode field is not valid(0x2001)
    # nvme id-ctrl /dev/nvme1 | grep oacs
    oacs      : 0

but the LBA format indeed is sus:

    LBA Format  0 : Metadata Size: 0   bytes - Data Size: 512 bytes - Relative Performance: 0 Best (in use)

It's a shame. The recent "datacenter nvme" standards involving fb, goog, et al mandate 4K LBA support.

it'd be great if you'd manage to throw together quick blogpost about i4g io perf, there obviously something funny going on and I imagine you guys could figure it out much easier than anybody else, especially if you are already having some figures in the marketing.

that's 16m blocks, not 4k

Last I checked, Linux splits up massive IO requests like that before sending them to the disk. But there's no benefit to splitting a sequential IO request all the way down to 4kB.

Don't work for Google, but been to the Toronto office (it is on the smaller side).

It is in the heart of the Toronto downtown, near Richmond & Spadina, next to the old & new City Hall. Definitely disagree with GP, I'm not sure what better area you would pick (but I love downtown). Similarly in Taipei, the Google office is right in Taipei 101 (like having an office in CN Tower - very cool.)

To be fair, Amazon's office in Toronto is next to the CN Tower and has a great view of it - so maybe Amazon takes the cake here. You have to pay for the cake though.

> It is in the heart of the Toronto downtown, near Richmond & Spadina

No, it's Richmond and University or Richmond and Bay. It's basically in financial district which is as boring and corporate as it comes. My point was to the GP saying "Google always seems to lease the best/coolest office real estate"

I'll agree wholeheartedly that the analogy needs some work. Tools are different - we have our literal physical tools that we don't generally dive into (keyboards, mice), we have tools that are maybe more battle tested and rarely examined (cat, grep, find).

We have do have tools like the hammer - there is one design, everyone more or less agrees on it. There is still high quality and low quality, but it has one job. We have tools like a bulldozer - complex, numerous parts, requires constant maintenance, closed source.

As the parent said - it is not uncommon to have to maintain old equipment, as well as design new tools as new requirements pop up.

Sure, our rust is a little bit different - time wears on software in a different way. Use wears on software differently. (Changing product requirements leading to a new tool is probably common.)

The maintenance may be trickier - but I'm sure changing components on a tool when a certain component is no longer available is not easy, thats where shim layer comes from!

+1 - Finished reading it this year, very fun read and much in the same vein as this article.

In my experience, the missing something is "have at least one computer that is a Windows PC" - at which point you are limited to a website. (Not the end of the world.)

Even better - if you use this Secure Note feature you simply can't get the note on a Windows PC.

It is actually the exact thing described in the 2nd section of the main article!

Wild that AMEX covers it! (while on a trip booked through them though, it looks like)