Devon H. O'Dell

We [discussed this a while ago](https://discord.com/channels/500028886025895936/670880858630258689/789273616498819130) in Discord and then I never got the time to prioritize doing this that I thought I'd have. My apologies. Copying over the suggestion...

N.B. if it does get to the point of analyzing different caches, LRU may not be the one to implement for a server cache since it is not scan-resistant. That...

> I would expect that in the "no resumption" case, utilization would be very high, meaning efficient use of the machine. In the "yes resumption" case I would expect utilization...

I hadn't heard of this, but I just read the blog post. It reminded me quite a lot of 2Q, and I like the combination of a frequency sketch. There...

Thanks for investigating that, I was surprised the throughput was this low — can’t really explain that with lock contention alone. I haven’t been able to look at the source...

I realized this morning why I was bothered by frequency count in S3-FIFO for this workload. The combination of **G** retaining data with the fact that counters only go to...

Sorry for the long delay! I only have time for a drive-by response at the moment, will plan to come back to this again later in the week with more...

Oh, please don't take my suggestions as acceptance criteria. I have some experience the domain of caching, and with ticket use in this crate. I agree with @jsha's suggestion for...

TLS 1.3 doesn't reuse tickets, which means the 2-bit frequency sketch can't be used. The ghost cache doesn't work because we can't reconstruct objects that get evicted. The two properties...

No matter what caching scheme we use, we have to optimize for worst-case performance. This is because TLS 1.3 doesn't reuse tickets, and any mechanism that allocates more than 1...