Disqus - Latest Comments for krobertson

Re: Dell XPS 13 9350 on Linux — In Valid Logic

krobertson — Wed, 25 Nov 2015 19:44:47 -0000

Sorry took so long to reply!

AHCI/sata are essentially where you have your flash SSD drive emulating a spinning disk. It is using drivers and protocols that are a bit dated in some senses, but also have some design aspects around the limitations of that hardware. Command queuing, needing to do seeking, etc.

NVMe and PCIe are essentially the same thing. NVMe is a newer drive interface built specifically for flash storage, it doesn't have the drive emulating some existing device handling. The drive is a straight PCIe device, with its own local controller, it is not plugging in to an existing controller, like the SATA card.

I saw a graphic a while ago about the software layers that AHCI/SATA SSDs go through in Linux compared to NVMe and it was crazy. Multiple layers of hardware controllers (SSD's controller, SATA's controller), in the kernel how SATA is built on SCSI, and all the layers. NVMe shortens the path removing like 30 years of handling.

Re: Linux Desktop Coming to an End — In Valid Logic

krobertson — Fri, 13 Feb 2015 13:25:00 -0000

The distro wars are old news. IMO, it doesn't matter what distro you run. It doesn't matter which is more GNU, more Linux, or more Unix. I run arch now. I've used Ubuntu, Debian, Slackware, I compiled compiled Gentoo from scratch. Distro is entirely a personal choice.

The "not so with open source" is false. You disprove it in the next paragraph. Why can't nvidia get fixed on the fly with open source? But nouveau isn't any way near the closed source drivers. Blaming firmware/closed drivers is still the hardware. So Broadcom sucks. So my new MBP which has a PCIe webcam with zero Linux support, but works just fine in OS X.

Open source as a end all solution is false, instead it more often leads to passing the puck. Take the root cause of my Chrome crashes:

https://code.google.com/p/c...

"Oh, it is gtk's bug, talk to them". And they link to 4 other bugs related to the issue, and those link to even other ones. Yet the gtk bug still ruins the Chrome experience, but no one is jumping in to fix it on the fly. It has been open since 2008! Think if that type of bug was happening on Windows or OS X it would be around for 6+ years?

But whatever. I ended up back on Linux anyway because workaround the issues was better than being impaired full time by the overall experience of OS X.

Ultimately the grass is always greener. You chose the one that has the pains you can bear, while looking at the other side jealous that you can't have both.

Re: Archlinux on a Lenovo Thinkpad X240

krobertson — Mon, 28 Jul 2014 03:04:21 -0000

I actually got it working with gummiboot just fine, but couldn't get refind working again. I just kept gummiboot and removed refind, whatever works. :)

Part of it may have been refind's installer... it was getting hung up on part of the boot disk's setup, since I had it at "/boot" and part of the drive had a ".../bootmnt" and it did some simplistic "cat /etc/mtab | grep /boot". Gummiboot worked out of the box, once I wrote a config entry for arch.

Re: Archlinux on a Lenovo Thinkpad X240

krobertson — Sat, 26 Jul 2014 16:13:08 -0000

Did installing refind work right out of the box for you? I'm setting up Arch on a new X240 and initially reused the windows EFI partition and that worked fine, but redoing it and this time setup a fresh EFI partition and now the system won't boot at all.

Re: Deploying Ubuntu 12.04 on XenServer Made Easy — In Valid Logic

krobertson — Fri, 22 Mar 2013 13:31:08 -0000

As pepoluan said, but also specify the root in case I attach multiple volumes, just for specificity... on many of the VMs, if I knew I'd need more storage for that machine's role, I'd create a separate volume.

'ro' is used along with specifying the root where it will do a fsck on boot, it tells it to mount it read-only until the bootup processes remounts it as read-write.

'quiet' simply shortens the output in the boot process... not really required by any means.

Re: Deploying Ubuntu 12.04 on XenServer Made Easy — In Valid Logic

krobertson — Fri, 22 Mar 2013 13:25:40 -0000

Creating your own mirror is pretty easy with apt-mirror... it has been a while since I've done it, but basically something like this: http://www.danbishop.org/20...

Then in the kickstart script, change any reference to "http://us.archive.ubuntu.co..." to point to your local apt mirror and you're pretty much set.

Re: This is how we know PaaS is winning

krobertson — Fri, 07 Sep 2012 18:48:41 -0000

"If your Platform is installed software – it’s not PaaS."

So by your own definition, Cloud Foundry is not a PaaS. AppFog is a PaaS, however since Cloud Foundry itself needs to be installed, it isn't a PaaS. Picking apart definitions leads to a slippery slope.

Re: This is how we know PaaS is winning

krobertson — Fri, 07 Sep 2012 18:45:49 -0000

By what source?

Re: Wells Fargo BillPay = FAIL — In Valid Logic

krobertson — Wed, 08 Aug 2012 16:57:27 -0000

Apparently wasn't paying much attention 8 months ago. Post was over 3 years old... you may need to get more too if you feel the need to attack a 3 year old post on some random blog.

Re: Wells Fargo BillPay = FAIL — In Valid Logic

krobertson — Wed, 08 Aug 2012 16:55:49 -0000

You also realize you're replying to a 4 year old post with your ever insightful comment?

Re: Deploying Ubuntu 12.04 on XenServer Made Easy — In Valid Logic

krobertson — Wed, 30 May 2012 02:26:41 -0000

I'll be accepting those pull requests, and will update the post about naming and localhost in /etc/hosts. Thanks!

Re: Deploying Ubuntu 12.04 on XenServer Made Easy — In Valid Logic

krobertson — Wed, 30 May 2012 02:26:06 -0000

Sorry for the delay, but at which point did you get the XenguestHelper error?

As for kernel updates, I never update the kernel in machines I've already provisioned. I will update and version the template, do some validation tests, and reprovision everything on the new template.

Re: Deploying Ubuntu 12.04 on XenServer Made Easy — In Valid Logic

krobertson — Wed, 16 May 2012 14:46:13 -0000

There aren't a whole lot of changes, and I believe the generic kernel includes the paravirtualization support (since it uses it by default). However the virtual kernel is slimmed down... it is intended for guest systems, so doesn't have a lot of hardware drivers or other things.

Re: Deploying Ubuntu 12.04 on XenServer Made Easy — In Valid Logic

krobertson — Tue, 15 May 2012 13:24:03 -0000

You aren't supposed to use the ISO or use "other media" at all. Read through my instructions and it should help you.

Re: Computational Journalism Server – The Way Forward

krobertson — Wed, 02 May 2012 20:27:44 -0000

Just to clarify, AppFog isn't for PHP and PaaS.io isn't for Haskell, just those two respective companies have added support for those to Cloud Foundry. I think your reason for not using either of them is more that they are public services rather than available software.

I also think point #1 for not choosing Cloud Foundry isn't valid either. It has support for those languages, but that doesn't mean you have to learn them, set them up, use them, or anything. If you don't want to use them, just comment a few lines out of a config file.

Documentation does suck though.

With CF/OS, one point to look at is which distro you wish to run on. CF is heavily Ubuntu/Debian focused, while OS is very RHEL/CentOS focused, and likely better for OpenSUSE as well.

Re: Configuring MongoDB Replica Sets With Keepalived — In Valid Logic

krobertson — Sun, 01 Apr 2012 01:08:54 -0000

That would be very useful! I currently use it in conjunction with HAProxy in front of the routers, as well as here with MongoDB. It has been working quite well, however does require the customized MongoDB service... I will be posting that though, I'm still working on merging in the latest code drop.

Re: Runtime Upgrades — PaaS.io Blog

krobertson — Wed, 07 Mar 2012 13:17:53 -0000

Do you mean to deploy to your own datacenter? No, we currently don't offer private installs, however that is on our longer term roadmap. We do have some options for doing custom deployments though, where it is in our datacenter, with hardware to your specifications, and the PaaS running on top of it.

Re: BlueScripts - Software Engineering & Development | Why the cloud isn't for your startup

krobertson — Tue, 06 Mar 2012 00:04:02 -0000

You are assuming your current setup would carry 1:1 to another platform. You've tuned your stack to your current setup. Each platform should be evaluated independently. 24 unicorn workers doesn't mean you need 24 dynos. You've probably made other design decisions that might not be necessary. Ever looked at async instead of simply using unicorn?

Also echo the other point about the most expensive thing in your company is you. Another angle... looks like you currently use just a single server, especially since you nail Heroku on a read-only file system. So you currently aren't scaling horizontally, and your entire service is a SPOF. How much does downtime cost you?

Re: Follow Up On Our Downtime Last Week

krobertson — Thu, 12 Jan 2012 18:32:42 -0000

Why not instead use a local "store and forward" approach? Now, if your syslog server goes down, you lose all data during that time. With store and forward, it would be buffered locally.

Re: Introducing Papercut Web Edition — In Valid Logic

krobertson — Tue, 04 Oct 2011 12:09:35 -0000

No, I haven't made the source for DummySmtp available. I may consider open sourcing it, but haven't really looked at it at all.

Re: Why I don’t share on Google+ much

krobertson — Mon, 29 Aug 2011 14:47:30 -0000

Same as me. My domain is a Google Apps for Domains account and I can't use G+ with it. Creating unnecessary roadblocks will stunt adoption.

Re: Serving Static Content Via POST From Nginx — In Valid Logic

krobertson — Wed, 27 Jul 2011 10:25:54 -0000

In our case, that wouldn't work and was less than ideal. First, some of the requests referred to the file name, so would need ugly hacks like file.html.php and more options in try_files calls. Second, we don't use PHP at all and for something this trivial, not worth introducing. And even then, the files then need to be served by a fastcgi process rather than nginx and would likely have greater throughput limitations.

Re: Skynet, EC2, and Zencoder

krobertson — Fri, 22 Apr 2011 16:32:33 -0000

Thanks for the updates on your status dashboard and the post-mortem. Glad to see you guys are working on the improvements.

Re: Our pain points with EC2 and how our moved solved them — In Valid Logic

krobertson — Fri, 11 Mar 2011 16:18:03 -0000

Our main virtualization chassis are Dell R810s, dual 6-core systems with hyperthreading, capable of quad CPU (but none of them are), 128gb ram. Just some basic storage on them, all VMs stored on an EqualLogic SAN, running XenServer Enterprise.

Re: Our pain points with EC2 and how our moved solved them — In Valid Logic

krobertson — Sun, 20 Feb 2011 18:55:47 -0000

I purposely left hard numbers out because some of them can be considered confidentially... especially things like budget.

Additionally, where the threshold is can vary a lot from case to case. It is easier to identify for someone already on AWS (or any service for that matter).

Estimating it before you are on the service is incredibly difficult. Just moving from AWS to our new setup, we tried estimating CPU performance/utilization and were way off. Having raw CPUs was far faster than what we thought were comparable EC2 machines, and our combined CPU usage is lower than we thought it'd be.

You should look at your type of workload and try to see what you're bound by. If it is IO, forget EC2, unless you are able to easily spread across more systems. If you are still pretty small, then EC2 is great. You'd be within its bounds and it gives a lot of value-added things that are excellent for a small service.