Symbolic Butt posted:

bobbilljim posted:

its a sign, though, that no sane person should use sourceforge

quote:

Let's talk about a hypothetical public-facing service that offers tools for collaboration, revision control, and software publishing. For purposes of this discussion, let's imagine the service was developed in the early 00s and their last major hardware purchase was done in the late 00s, so many of the contemporary infrastructure technology we take for granted nowadays (virtualization, cloud, software-defined networking, logical RAID) were on the horizon but not necessarily mature enough to rely on.

You work on the systems team for this service, and you have a need for terabytes of disk space to store user data.

Because of the nature of your system that isn't virtualized (or is barely virtualized), you look at the two options for accessing disks from many servers at once: Shared-access block storage (Fiber/iSCSI, requiring some sort of shared filesystem like GFS) or shared-access file storage (NFS, SMB, etc. well-known protocols with which you have a lot of experience. Despite all the drawbacks with NFS (most of which are minimized with NFSv4), this seems like the best approach at the time, but you don't want to manage a fleet of NFS storage heads in your lights-out datacenter a thousand miles away, so your team looks at off-the-shelf storage systems that include an NFS filer head.

After picking apart several leading candidate products, your team picks one and submits it to the nebulous powers-that-be for approval. Unfortunately, the powers-that-be have their own horse in the race: Board members have affiliations with this product's leading competitor. They influence the purchasing decision and you end up with a completely different product than you initially specified. The one that arrives made by an Extremely Massive Corporation who until now you've generally respected.

The product arrives and you start using it. It does pretty much everything you expect: Combines a bunch of spinning hard disks into a bigger, redundanter volumes and lets you slice those into filesystems. It also lets you export those filesystems over the network with varying degrees of access permissions.

You decide (because it's still the 00s and thinking was done differently back then) to use standard UNIX permissions for all your user data: Each user will have a user ID and each software project will have a group ID. These IDs will be distributed to your servers through industry-standard protocols like LDAP. This has worked for thousands of other places, it should work here!

Except, you learn very quickly when you start storing user data to your big, new, shiny storage array, that there is a fundamental flaw in the way it keeps track of permissions. It writes a full 32 bits of numeric user ID to its filesystem, but to save a few bytes it only stores 16 bits of group IDs. Some engineer probably thought that'd be enough for anybody. The device happily accepts 32 bits of group ID, but can only remember the first 65k that it encounters (which it stores in a neat little table at the beginning of the disk mapping the 32-bit IDs from the wire to the 16-bit IDs on disk). For added fun, the engineer designing the system must have thought that anything that creates 65537 group IDs must be running out of control so the failure state for the ID mapping table would be to make the whole filesystem permanently read-only.

You discover this flaw and work around it in some time-honored ways: Sharding! Instead of a big filesystem for your users' data, you make 27 filesystems. One each for names starting with A through Z and one more for anything else. You could easily choose another hash function that would distribute the data more evenly, but that's irrelevant to this situation so far.

This works! It meets requirements! Your data is suddenly fast, redundant, and you can easily back it up using RAID-side snapshots and block copying without your front-side servers being impacted. (You have a moment where you envision the future of virtualized storage and think about how great it will be when storage is nearly free and outsourceable and you can stop buying disks from Amazon every few months)

Unfortunately, all things must end. The executive-mandated storage system has a three-year manufacturer warranty. The disks in it, from another major manufacturer, have a five year warranty. You purchase all the service plans and support contracts for these things, but the actuaries who determined the warranties were intimately familiar with the lifespan of these devices.

But then your entire team gets forced out of their job and all the knowledge about how to handle this moody piece of hardware is lost, like tears in the rain. Every once in a while you wonder who is maintaining it nowadays, and how they're handling some of its quirks, and whether it's been replaced yet or not.

Of course it has. It'd be eight years old by now. Surely whoever is responsible for it understands a bathtub curve.

Surely.

A Wheezy Steampunk posted:

A Wheezy Steampunk posted:

OSI bean dip posted:

sourceforget

Silver Alicorn posted:

I actually had to download unetbootin today and was forced to resort to a mirror site

MALE SHOEGAZE posted:

there are plenty of other websites to visit if you want some malware installed on your pc

Maximum Leader posted:

rufus is better anyway

SYSV Fanfic posted:

Guys, after anciously hitting f5 all day, I'm pleased to announce its back up!