@continuousimprv.bsky.social
Start from where you are. Use what you have. Do what you can.
December 23, 2025 at 7:43 AM
December 23, 2025 at 4:36 AM
December 17, 2025 at 7:11 AM
Gutter Pool: address the failure of a small number of hosts
are inaccessible due to a network or server failure
pdos.csail.mit.edu/6.824/papers...
are inaccessible due to a network or server failure
pdos.csail.mit.edu/6.824/papers...
December 17, 2025 at 2:25 AM
Gutter Pool: address the failure of a small number of hosts
are inaccessible due to a network or server failure
pdos.csail.mit.edu/6.824/papers...
are inaccessible due to a network or server failure
pdos.csail.mit.edu/6.824/papers...
Thundering Herd Problem:
A SINGLE write triggers MANY frontends read from the database when that key is POPULAR that many frontends need to get that key.
pdos.csail.mit.edu/6.824/papers...
A SINGLE write triggers MANY frontends read from the database when that key is POPULAR that many frontends need to get that key.
pdos.csail.mit.edu/6.824/papers...
December 17, 2025 at 1:44 AM
Thundering Herd Problem:
A SINGLE write triggers MANY frontends read from the database when that key is POPULAR that many frontends need to get that key.
pdos.csail.mit.edu/6.824/papers...
A SINGLE write triggers MANY frontends read from the database when that key is POPULAR that many frontends need to get that key.
pdos.csail.mit.edu/6.824/papers...
December 16, 2025 at 9:44 AM
Why favour replication in this instance over further dividing the key space?
pdos.csail.mit.edu/6.824/papers...
pdos.csail.mit.edu/6.824/papers...
December 16, 2025 at 7:34 AM
Why favour replication in this instance over further dividing the key space?
pdos.csail.mit.edu/6.824/papers...
pdos.csail.mit.edu/6.824/papers...
The memcache problems that FB(Meta) aims to solve for reducing the latency of memcache responses
pdos.csail.mit.edu/6.824/papers...
pdos.csail.mit.edu/6.824/papers...
December 6, 2025 at 3:27 AM
The memcache problems that FB(Meta) aims to solve for reducing the latency of memcache responses
pdos.csail.mit.edu/6.824/papers...
pdos.csail.mit.edu/6.824/papers...
What if RO TX 1 read X before the X=10 @10 is replicated to the client local replica?
www.youtube.com/watch?v=ZulD...
www.youtube.com/watch?v=ZulD...
December 1, 2025 at 9:56 AM
What if RO TX 1 read X before the X=10 @10 is replicated to the client local replica?
www.youtube.com/watch?v=ZulD...
www.youtube.com/watch?v=ZulD...
OSDI12 - Spanner: Google’s Globally-Distributed Database Presentation by Wilson Hsieh
www.youtube.com/watch?v=C75k...
www.youtube.com/watch?v=C75k...
December 1, 2025 at 3:28 AM
OSDI12 - Spanner: Google’s Globally-Distributed Database Presentation by Wilson Hsieh
www.youtube.com/watch?v=C75k...
www.youtube.com/watch?v=C75k...
November 29, 2025 at 4:10 AM
November 21, 2025 at 8:57 AM
November 21, 2025 at 3:36 AM
November 17, 2025 at 4:25 PM
Coordination among concurrent actions can be considered to be correct if **every result** is guaranteed to be one that could have been obtained by some **purely serial application** of those same actions.
November 16, 2025 at 8:39 AM
Coordination among concurrent actions can be considered to be correct if **every result** is guaranteed to be one that could have been obtained by some **purely serial application** of those same actions.
BABE's leader election:
1. VRF elects primary ledaer => 0 or 2+ # of leaders can be elected
2. Round robin elects secondary leader => 1 leader will be elected for each slot; Secondary blocks are ignored if the same slot also produces a primary block
Why use VRF + RR? No slots are ever skipped
1. VRF elects primary ledaer => 0 or 2+ # of leaders can be elected
2. Round robin elects secondary leader => 1 leader will be elected for each slot; Secondary blocks are ignored if the same slot also produces a primary block
Why use VRF + RR? No slots are ever skipped
November 15, 2025 at 7:35 AM
BABE's leader election:
1. VRF elects primary ledaer => 0 or 2+ # of leaders can be elected
2. Round robin elects secondary leader => 1 leader will be elected for each slot; Secondary blocks are ignored if the same slot also produces a primary block
Why use VRF + RR? No slots are ever skipped
1. VRF elects primary ledaer => 0 or 2+ # of leaders can be elected
2. Round robin elects secondary leader => 1 leader will be elected for each slot; Secondary blocks are ignored if the same slot also produces a primary block
Why use VRF + RR? No slots are ever skipped
Distributed Counter without using lock
November 12, 2025 at 7:50 AM
Distributed Counter without using lock
Usage of notification:
November 12, 2025 at 7:16 AM
Usage of notification:
Read operation guarantee:
* observe the last write of the same client
* observe the prefix of the full log
How?
* write operation always go to leader
* leader will return the zxid to the client (the index of the committed write operation)
* client call read operation with zxid
* observe the last write of the same client
* observe the prefix of the full log
How?
* write operation always go to leader
* leader will return the zxid to the client (the index of the committed write operation)
* client call read operation with zxid
November 11, 2025 at 7:23 PM
Read operation guarantee:
* observe the last write of the same client
* observe the prefix of the full log
How?
* write operation always go to leader
* leader will return the zxid to the client (the index of the committed write operation)
* client call read operation with zxid
* observe the last write of the same client
* observe the prefix of the full log
How?
* write operation always go to leader
* leader will return the zxid to the client (the index of the committed write operation)
* client call read operation with zxid
Why this trace is not linearizable?
November 11, 2025 at 3:57 PM
Why this trace is not linearizable?
November 11, 2025 at 3:43 PM
Simple Locks without Herd Effect
November 11, 2025 at 9:01 AM
Simple Locks without Herd Effect
A problem can arise when clients have their own communication channels in addition to ZooKeeper:
pdos.csail.mit.edu/6.824/papers...
pdos.csail.mit.edu/6.824/papers...
November 11, 2025 at 6:52 AM
A problem can arise when clients have their own communication channels in addition to ZooKeeper:
pdos.csail.mit.edu/6.824/papers...
pdos.csail.mit.edu/6.824/papers...
* Why is ideal consistency hard? concurrency & failures
* Why is client cache important? minimise the communication with the SINGLE master
www.youtube.com/watch?v=6ETF...
* Why is client cache important? minimise the communication with the SINGLE master
www.youtube.com/watch?v=6ETF...
Lecture 3: GFS
YouTube video by MIT 6.824: Distributed Systems
www.youtube.com
November 9, 2025 at 9:54 AM
* Why is ideal consistency hard? concurrency & failures
* Why is client cache important? minimise the communication with the SINGLE master
www.youtube.com/watch?v=6ETF...
* Why is client cache important? minimise the communication with the SINGLE master
www.youtube.com/watch?v=6ETF...
Lease Mechanism and Data Flow
November 8, 2025 at 7:51 AM
Lease Mechanism and Data Flow