Malevolent Cartography

Write C

Drink your coffee black

Sleep on the floor

Burn your rolling luggage

Contact

github.com/qpfiffer

qpfiffer+qpweb@gmail.com

Resume

Links

MulletCache

2020-04-03 by Quinlan Pfiffersoftware

Introduction

Herein a novel idea for multi-nodal caching is proposed. The main problem solved by this idea is thus: Home storage is cheap, cloud storage is expensive. The project’s name is MulletCache.

What is it?

MulletCache consist of at least one remote node (the business node) and one local node (the party node). The party node has a maximum cache size allowance, which should be roughly the size of your on-disk availability. The business node has no cache size, it just serves any file requested from it.

       Home                 Cloud
+----------------+      +------------+                         +--------------+
|                |      |            |    Request for Asset    |              |    HTTP Request
|   Party Node   |      |  Bzns Node | <---------------------+ | HTTP Service | <----------------+
|                |      |            |                         |              |
+----------------+      +------------+                         +--------------+

The party node will handle any incoming request for a file, check it’s internal LRU cache (and by proxy the local filesystem), and request any file it doesn’t have from the business node.

       Home                 Cloud
+----------------+  RX  +------------+                         +--------------+
|                <------+            |    Request for Asset    |              |    HTTP Request
|   Party Node   |      |  Bzns Node | <---------------------+ | HTTP Service | <----------------+
|                +------>            |                         |              |
+----------------+  TX  +------------+                         +--------------+

The Party node can be configured in either write-back or write-through configurations, depending on speed vs. safety requirements. Heres example pseudo-code for the party node’s asset handler routine:

def fetch_file(requester, filename):
    if file_present_locally(filename):
        requester.send_file(file)
    else:
        new_file = await request_file_from_business_node()
        if cache_is_full():
            cache.evict_file()

        if config.write_through:
            new_file.write_to_disk()

        requester.send_file(file)

        if config.write_back and not config.write_through:
            new_file.write_to_disk()

Caveats

  • All communications channels between nodes must be encrypted.
  • The business node must connect to the party node: The party is in the cloud, so the business must connect up to it. This keeps us from having to do any port shenanigans on the home router.
  • The business node must be configurable to connect to multiple parties.
  • Protocol between the whole mullet TBD: Probably checksum, filename and file contents.
  • The interface to the party node must be through FUSE.