Andrew Fitzgerald
@apfitzge
Followers
697
Following
199
Media
5
Statuses
81
Explore trending content on Musk Viewer
Claudia Sheinbaum
• 501506 Tweets
Fauci
• 251220 Tweets
BB X LISA
• 232366 Tweets
$GME
• 113156 Tweets
NORAWIT and KND
• 104830 Tweets
Venom
• 75493 Tweets
God Kabir Prakat Diwas 22June
• 54852 Tweets
エミリエ
• 50345 Tweets
First Take
• 44895 Tweets
音楽の日
• 39666 Tweets
#キュウゴー
• 36912 Tweets
#すとぷり8周年
• 34306 Tweets
Jadue
• 33896 Tweets
#アンメット
• 29040 Tweets
Emilie
• 28955 Tweets
ベンチマーク
• 26316 Tweets
더보이즈
• 23819 Tweets
Alpine
• 23705 Tweets
Başın
• 23624 Tweets
Ocon
• 20495 Tweets
Justin Jefferson
• 17934 Tweets
ヴェノム
• 17466 Tweets
Barriga
• 15217 Tweets
Welcome to Chelsea
• 13639 Tweets
STELL IS COMING
• 13275 Tweets
Last Seen Profiles
Backported the (reward/cost) prioritization to the new scheduler in version 1.18.
Devs & users get higher priority if you request fewer CUs!
This also begins to consider base fees, so simple transfers, even without priority fees, have **some** priority in queue!
7
8
51
Currently priority-order leader buffer is done by the cu_price in the compute budget ixs. This totally ignores the non-priority rewards leader gets for including a transaction.
#34888
will instead prioritize by total_fee / cost, hoping to land in 1.18 w/ new scheduler.
6
6
30
mean fee brought up by high priority fees on jup action; median fees staying low. love to see it. we can still do better
2
2
21
This was fun to do, looking forward to seeing other devs present on other parts of the code!
There's so much to Solana that help make it fast and efficient, but not everyone knows how Solana works
I spent time with
@apfitzge
learning how the current scheduler works and how 1.18's scheduler further improves Solana
Learn about the scheduler 👇
14
41
152
2
2
22
Happy to announce I'll be moving to Anza. I'm so pumped for everything going on in the network/ecosystem rn!
1/ We're thrilled to announce Anza — a next-gen,
@Solana
focused dev shop.
Founded by execs & engineers from
@SolanaLabs
, Anza will build a forked validator client called Agave & contribute to major Solana protocols.
61
174
664
1
0
14
I just had to sign a document online, but it made me re-sign using mouse for every box. 27 signatures. This is hell.
0
0
6
An example of why we want this:
2 txs each with 1 sig, one requests 5000 cus, the other requests 200000 cus both with cu_price=1 lamport/cu.
currently these have same priority. But Tx2 uses much more of block limits.
Reward/cost is almost double for Tx1, it should be prioritized!
1
0
5
Without user adjustment, lower-cu txs will naturally have higher reward/cost and be prioritized.
By making accurate requests for compute_unit_limit, you naturally get higher priority since the denominator is lower!
Also taking unneccessary write-locks gives you lower priority!
1
0
4
Trying to give some notice on this change since it will likely require user/dev adjustment. This is outside of consensus and any operator is free to use similar logic if they so choose.
0
0
4
@aeyakovenko
@7LayerMagik
@solflare_wallet
credit where credit's due, the vote changes were from Ashwin not me! my scheduler changes are coming in 1.18
1
0
4
yesterday I woke up at 3:30am, just randomly couldn't sleep. coulda used that energy today, damn off by one error
0
0
3
@7LayerMagik
Just a reminder, in 1.18 the scheduler will still be opt-in via cli - there will be a note in there when the call to upgrade goes out.
1
0
3
@0xShitTrader
Until SIMD-0096 gets implemented they are essentially the same just off by a factor of 2. We want the reward to leader as numerator
1
0
3
@0xMert_
@Cypherj_sol
I want it too man, but its' a big changeset that is very unlikely to make it through the backport reviews.
At least 1500 loc added, ignoring any changes AND ignoring all the stuff outside the transaction_scheduler mod for hooking up and enabling it.
0
0
3
@7LayerMagik
@aeyakovenko
yeah, but then just do what I said. Shuffle tx to move your hot account and get the non-elevated fees for your users. As long as this is done in a predictable pattern (maybe w/ some cut-over period) then seems the apps can get around it
1
0
2
This app is making me tweet this to unlock my account.
@hey_wallet
bonk me 0xSOL9DE21F4B0-B4E5-43F5-8544-E4F3F3C991D6
0
0
1
@bqbrady
we've got a unique leader for each slot, so iirc there's only gonna be 1 block(hash) for each valid block.
could have malicious leader produce multiple blocks, which we have some special handling for. potentially leads to forking w/ multiple blockhashes per slot, not sure.
0
0
2
@CantelopePeel
It’s a loose term, so depends what you mean. But there are definitely things that are essentially effortless to parallelize. Monte Carlo simulation is a classic example
1
0
1
@CantelopePeel
@ceterispar1bus
@0x_Osprey
@markodayan
@7LayerMagik
Not as familiar with the purpose the tick boundary serves, it doesn’t seem necessary to me either though.
Not sure I agree with structuring entries as DAGs. I go the other way and feel should remove restrictions on entries rather than further restricting the format.
2
0
1
@0xMert_
@PierreArowana
1.18 will not 100% fix this - nothing will. If there are a million higher-paying that conflict with your tx, it will likely expire. No amount of optimization in the scheduler will let us get around the current account/block limits.
1
0
1
@fd_ripatel
@markodayan
@7LayerMagik
Current labs client uses pessimistic concurrency (locks held from start of execution to end of commit). This is a consequence of synchronous execution - and there’s a few protocol reasons we do this to ensure valid block creation. These could be relaxed in the future
0
0
1
@aeyakovenko
@7LayerMagik
@solflare_wallet
need to think about it more, but I think if we go towards async we'll need to separate vote-processing from non-vote txs. That processing should also be optimized... At that point, might make sense to also explicitly separate vote/non-vote block limits?
1
0
1
@CantelopePeel
@ceterispar1bus
@0x_Osprey
@markodayan
@7LayerMagik
I suppose my thought is that I already don’t trust their transactions, by some DAG ordering now I have to check they did that correctly as well.
I’m not sure that’s less expensive than me just building a DAG myself
0
0
1
@CantelopePeel
@ceterispar1bus
@0x_Osprey
@markodayan
@7LayerMagik
Think the validating node should receive and accept minimal set of info from leader in order to check their block is valid
0
0
1
this is not at time of launch, just when i happen to grab the screenshot off of
@SolanaCompass
0
0
1
@mfrager
@7LayerMagik
@thehasheddude
@dubbel06
@0xMert_
@cryptomiester
Separating state would help to a point - too much and you increase cost/execution time from loading a bunch of accounts. And will only help if they are written to separately. If they all write to some common account then that'd still be a bottleneck
0
0
1
@0xMert_
@PierreArowana
yeah sounds right. it's still not gonna be totally deterministic, and you want to get to leader as early as possible. if tx has highest priority when scheduling starts it will still get in before a higher-priority tx that shows up near end of block. we don't re-order after record
0
0
1
@aeyakovenko
546 just started recently, I'm guessing this spike will go down as the epoch continues. I'd guess seeing congestion early on that makes this look exaggerated.
1
0
1
@0xMert_
@PierreArowana
1.18 scheduler will greatly increase the changes you get included if you are close to top-priority for whatever state you're accessing.
1
0
1