Free Downloadable RAM for Linux? Help!

[Jared_Hulbert]

I asked the mods if I could do this but it’s hard to say this stuff without coming off as corporate. And software to improve RAM usage has been a scammer game. But we’re just fellow Linux hackers that choose to go all in on our cool project and the tech comes from decades of research at Intel and Micron labs (turns out Micron didn’t want to sell less RAM, long story…)

We at https://bitflux.ai could really use your help getting our product out there. Help us now as a beta user and we’ll give you free premium features when we start charging. And cool stickers, yeah, confirmed installs will get stickers! I wasn’t planning on that but that’d be cool.

Think of us as a service like Datadog, only instead of just recording metrics, we actively improve them for you. We’re starting with memory and moving on to storage next.

Here’s what you can do:

1. Sign up using this link so we can know who to send stickers to. LevelOneTechs Promo
2. Follow Quick Start Guide | BITFLUX.AI Wiki to install to your vm/hw running Linux.
3. Let us know how it works (or doesn’t) for you [email protected]
4. Receive cool sticker!

Ever notice how no matter how much RAM you add, your memory ends up full of… something? How your desktop is snappier after a reboot? Strange lags when nothing of interest is going on? Do you reboot your servers nightly or weekly to avoid a service from crashing? Do you wonder how much RAM are you really using or how much more you actually could use? Have you run a job confident you have more than enough memory for, but it’s not enough? Feel like your system just isn’t very smart at managing memory? You’re right, it isn’t very smart and that’s where BitFlux can help. Works especially good keeping your desktop clean and lets you cram more services into your homelab set up. Take a look at BitFlux Uses Cases

Thanks for your support!

,Jared Hulbert

CEO/co-founder

BitFlux, Inc.

[get_off_my_lawn]

“Ever notice how no matter how much RAM you add, your memory ends up full of… something?” - uh nope can’t say I notice this…given my mobo is maxed out. ;p

Probably the wrong crowd to be “marketing to” for problems only someone who bought a garbage Dell in the 1990’s would have. Ever notice how Bonzai Buddy uses all your RAM? You need Windows NoobdySeven!

[Jared_Hulbert]

Ha. I like the name. Obviously the problem is you don’t run enough crappy software then.

Still would help us to see how little RAM you could have gotten away with. Stickers…!?

[Trooper_ish]

Is the general premise, like QOS for ram? as in, you have a pre-determined priority for memory, and scale what is being used by which app?

Or kinda try to cut down on what is in memory? as in, kill caches and unload databases / apps?

Any telemetry in the app / it’s installer?

[cowphrase]

Reading this post I couldn’t help but think of this episode of computer chronicles - The Computer Chronicles - Memory Problems (1994) - YouTube - I’m glad that some problems are no longer so severe.

[Trooper_ish]

Okay, checking the FAQ, the system does decide which apps / caches to unload, and it does allow for telemetry, so one can log / monitor with grafana.

Looks pretty cool, given the tendency to over-provision stuff “just in case” which I guess could add up to actual money wasted

[wertigon]

This reminds me of a freeware product I used back in the days of Windows 98. It defragmented RAM. Just like harddrives, RAM gets fragmented over time which impacts performance, not to mention programs were terrible in releasing allocated RAM. They still are pretty terrible about that actually, but given I now have order of a magnitude more RAM (from 32 MB to 32 GB) than I had back then, and fragmentation chunks in memory pages are usually in the kB range, weeell… I no longer really need it. Atleast not on Linux.

Not to say this product is bad or anything though, I’m sure it is a lot better than that crappy old freeware program. This product sounds especially nice when you run RAM-intensive workloads, so while not for me, I can see a place for it.

[Novasty]

@wertigon

It defragmented RAM. Just like harddrives,

iobit software, I had the same shit installed during the XP~Vista era

[Exard3k]

Optimizing limited resources is always a good thing. The more you can get out of your memory, the better.

Just to remind you: We only have “storage” because we’re too poor to get more RAM.

[gee_one]

Can you post a picture of the stickers?? I want to see what I’m getting into.

[Carlo]

Maybe you should put off storage and work on connections next.

[Jared_Hulbert]

Yeah. We’re basically just taking a longer, deeper AI assisted look at the memory compared to the OS itself, then telling the OS which pages to act on. Acting meaning attempting a purge or swap, Lots of details here Bitflux Tech Talk - YouTube

We’re planning on extending telemetry to include other useful factors too. What data would be most useful?

[Jared_Hulbert]

Error 503, crap. On it. Thanks for the patience as we figure this out.

[Jared_Hulbert]

Stickers would be like this. Equivalent to this only it’d be a bulk order so it’s not stupid expensive.
BITFLUX.AI Logo Bubble-free stickers - Shop.BitFlux.AI

[cotton]

I’m trying to understand what this does.

Does the product attempt to propose a better eviction algorithm than the standard LRU Eviction policy the linux kernel uses?

Does it find and free memory that’s part of a memory leak? AKA does it find “wasted memory” in heap that is no longer removable by the application?

If a crucial working set must move into memory does it attempt to swap important pages from memory until the working set completes and then reload them?

Is there a increased demand for swap and furthermore in disk?

For me it would be helpful to be able to determine if this product is worthwhile with a better description of the underpinnings of how it makes your machine run better. Typically, “Downloading moar RAM” is a really bad idea, and may be a play at that?

Perhaps… Our tool is a kernel space running application which through telemetry uses statistics and modeling to determine the best way to manage your RAM. In fact, our management of RAM beats the standard linux kernel by {{ insert value here }} by forgoing the typical “dumb LRU” and smartly through machine learning.

Ever run out of ram from a pesky compiled program, or JVM app leaking memory? We’ve developed magic to somehow identify heap pages no longer referenced by pointers that your developer missed or the JVM overlooked.

Ever have a report that’s too large of a working set? We can priorities your current memory map, migrate it to swap, process the working set, and swap it back in for faster of business processing, but at some cost to other programs running on the box.

Or, we just watch top and muck with the niceness and swapiness of your apps and have automated linux sys admining.

…

Some sophisticated technical details would be warranted if you’re asking a tech site to beta test your app for some stickers… we are a community and we look out for each other.

An easy way to answer a lot these questions would be to open up the “Source Code” section on your site so we can read what it does.

cotton

[Jared_Hulbert]

@cotton

Does the product attempt to propose a better eviction algorithm than the standard LRU Eviction policy the linux kernel uses?

This is basically it, smart proactive sanitation.

@cotton

If a crucial working set must move into memory does it attempt to swap important pages from memory until the working set completes and then reload them?

If I understand your question correctly then I would say it’s more that we make sure you have free memory ready for the crucial working set ahead of time.

1. ‘available memory’ is not all actually available
2. free memory is much faster to allocate than ‘available memory’
3. we attempt to maximize free without impacting performance

@cotton

Is there a increased demand for swap and furthermore in disk?

As always, it depends. How close to full are you running, how aggressive are you optimizing (documentation is in progress but we have tuning factors obviously) etc. IF you swap to zram than none at all.

In many cases we actually a lot do less IO per GB in swap. I’m sure you could design cornercases where we do more. Throttle settings are planned to limit swap IO.

The big thing is though by being proactive the swap IO doesn’t impact you because you aren’t doing it when you have memory pressure already. We help you avoid the swap storms, we’re not trying to cause them.

@cotton

For me it would be helpful to be able to determine if this product is worthwhile with a better description of the underpinnings of how it makes your machine run better.

We’ll think on this stuff, thanks. I appreciate the feedback.

@cotton

Typically, “Downloading moar RAM” is a really bad idea, and may be a play at that?

Actually we do own a couple downloadable RAM domains, we were literally thinking about doing a play on that!!

@cotton

Some sophisticated technical details would be warranted if you’re asking a tech site to beta test your app

I encourage you to watch Bitflux Tech Talk - YouTube I stupidly forgot to mention that in the first post.

@cotton

we are a community and we look out for each other.

I posted on this site for a reason

@cotton

An easy way to answer a lot these questions would be to open up the “Source Code” section on your site so we can read what it does.

Yeah oversight on my part I thought we had the link up on there. Our kernel mod and build scripts are here. Build scripts make it a mess of complexity and it doesn’t actually answer as many of the questions as you might think. GitHub - resurgentech/bitflux_patch: kernel patches and tools

The kernel changes are as uninteresting as we can make them. We want them to go away obviously. They boil down to a /proc interface where you write physical page numbers and the kernel calls the underlying code it normally does when evicting.

Again thanks for the feedback and engagement!

[Four0Four]

@Jared_Hulbert

@cotton

Does the product attempt to propose a better eviction algorithm than the standard LRU Eviction policy the linux kernel uses?

This is basically it, smart proactive sanitation.

Are there any benefits to smart proactive sanitation other than free memory being faster to allocate than available memory? And by proactively freeing available memory are applications able to allocate more memory than if it was left only to the linux kernel’s memory management?

@Jared_Hulbert

If I understand your question correctly then I would say it’s more that we make sure you have free memory ready for the crucial working set ahead of time.

…

2. free memory is much faster to allocate than ‘available memory’

Are you able to share some of your testing data with us? I wasn’t able to find any meaningful statistics on latency in relation to reallocation of memory when freed by the linux kernel. Curious how many milliseconds improvement should we expect when launching a process with Bitflux vs relying on the linux kernel to reallocate the memory?

@Jared_Hulbert

I encourage you to watch Bitflux Tech Talk - YouTube I stupidly forgot to mention that in the first post.

So maybe a video where you’re comparing apples to apples would be helpful? I’m really confused why there is a 2GB RAM /w Bitflux vs 4GB RAM (without Bitflux) server comparison. It would had made more sense to see 2GB /w Bitflux vs 2GB to compare latency and if your application could provide better memory management than the linux kernel on its own. But instead it feels similar to an infomercial trying to sell a dietary supplement using carefully chosen words. Sorry, I know that sounds a bit harsh but that that’s just how it came across when I watched it.

As for your wiki, I think it would be helpful to include more technical details and provide the source code if this is an open source project(?) I’m sorry to sound skeptical but there are a lot of performance claims and no benchmarks to quantify them. I really hope my gut feeling is wrong and I would love the opportunity to apologize to you later if that’s the case.

[Jared_Hulbert]

@Four0Four

Are there any benefits to smart proactive sanitation other than

Yeah, knowing what you are actually using. Left to it’s own devices the kernel is very lazy about reclaim, so you don’t know how much memory actually have available. ‘available’ is way to optimistic, and ‘free’ is way too small. Say you have 4GB available and 1GB free according to a stock kernel. Can you add an app or service that uses 2GB? or 3GB? How much is left over in reserve for spikes? BitFlux will help answer that by showing how much is actually ‘used’ over time.

@Four0Four

Are you able to share some of your testing data with us?

As we produce it sure. I spent a lot of my career doing benchmarking so I tend to be a perfectionist with it documenting it which makes it more work to release. We’ll try to figure out a middle ground on that.

The best proof point of ‘much faster’ is in the Tech Talk video.

@Four0Four

Curious how many milliseconds improvement should we expect

The trouble with answering this is like a lot of things it depends… In what pools of memory was the ‘available’, how much CPU is available to work on it. How many processes memory maps need to be walked? How much contention will there be for the various resource locks?

So how do you design an experiment to benchmark that? That’s a PhD thesis right there.

@Four0Four

It would had made more sense to see 2GB /w Bitflux vs 2GB

Hmm, I think we did that and like one of the apps would get shut down maybe?

But we chose the 2 vs 4 on purpose because we’re not really selling on faster. We’re selling do more with less. Does that make sense.

@Four0Four

infomercial trying to sell a dietary supplement using carefully chosen words

I mean… that’s marketing. I’d be even lamer for me to show a bunch of corner cases where there was barely noticeable differences, right?

BTW Our informerical - BitFlux Infomercial - YouTube

@Four0Four

there are a lot of performance claims and no benchmarks to quantify them

Hopefully we’re not projecting a lot of performance claims, we haven’t been trying to sell this as speeding up your RAM. We’re trying sell on getting more workloads to fit in a machine with better predictability.

See the other comments for the open source components.

As far as benchmarks what’s interesting to y’all?

[Four0Four]

@Jared_Hulbert

Hopefully we’re not projecting a lot of performance claims, we haven’t been trying to sell this as speeding up your RAM. We’re trying sell on getting more workloads to fit in a machine with better predictability.

Yeah, sorry that message was lost on me when I watched your other YouTube video titled BitFlux Opening App that shows an over 100% application performance increase with BitFlux on a 2GB machine vs. a 4GB without.

2GB /w Bitflux: 32 seconds
4GB /wo: 1 min 21 seconds

[Jared_Hulbert]

@Carlo

This 503 error was just on the website not our app right? If so we just host that on webflow… not much to fix there if it happens alot we’ll just have to find a different hosting company I guess.