When fundamental particles are heavier or lighter than expected, physicists’ understanding of the universe can tip into the unknown. A particle that is just beyond its predicted mass can unravel scientists’ assumptions about the forces that make up all of matter and space. But now, a new precision measurement has reset the balance and confirmed scientists’ theories, at least for one of the universe’s core building blocks.

In a paper appearing today in the journal Nature, an international team including MIT physicists reports a new, ultraprecise measurement of the mass of the W boson.

The W boson is one of two elementary particles that embody the weak force, which is one of the four fundamental forces of nature. The weak force enables certain particles to change identities, such as from protons to neutrons and vice versa. This morphing is what drives radioactive decay, as well as nuclear fusion, which powers the sun.

Now, scientists have determined the mass of the W boson by analyzing more than 1 billion proton-colliding events produced by the Large Hadron Collider (LHC) at CERN (the European Organization for Nuclear Research) in Switzerland. The LHC accelerates protons toward each other at close to the speed of light. When they collide, two protons can produce a W boson, among a shower of other particles.

Catching a W boson is nearly impossible, as it decays almost immediately into two types of particles, one of which, a neutrino, is so elusive that it cannot be detected. Scientists are left to measure the other particle, known as a muon, and model how it might add up to the total mass of its parent, the W boson. In the new study, scientists used the Compact Muon Solenoid (CMS) experiment, a particle detector at the LHC that precisely tracks muons and other particles produced in the aftermath of proton collisions.

From billions of proton-proton collisions, the team identified 100 million events that produced a W boson decaying to a muon and a neutrino. For each of these events, they carried out detailed analyses to narrow in on a precise mass measurement. In the end, they determined that the W boson has a mass of 80360.2 ± 9.9 megaelectron volts (MeV). This new mass is in line with predictions of the Standard Model, which is physicists’ best rulebook for describing the fundamental particles and forces of nature.

The precision of the new measurement is on par with a previous measurement made in 2022 by the Collider Detector at Fermilab (CDF). That measurement took physicists by surprise, as it was significantly heavier than what the Standard Model predicted, and therefore raised the possibility of “new physics,” such as particles and forces that have yet to be discovered.

Because the new CMS measurement is just as precise as the CDF result and agrees with the Standard Model along with a number of other experiments, it is more likely that physicists are on solid ground in terms of how they understand the W boson.

“It’s just a huge relief, to be honest,” says Kenneth Long, a lead author of the study, who is a senior postdoc in MIT’s Laboratory for Nuclear Science. “This new measurement is a strong confirmation that we can trust the Standard Model.”

The study is authored by more than 3,000 members of CERN’s CMS Collaboration. The core group who worked on the new measurement includes about 30 scientists from 10 institutions, led by a team at MIT that includes Long; Tianyu Justin Yang PhD ’24; David Walter and Jan Eysermans, who are both MIT postdocs in physics; Guillelmo Gomez-Ceballos, a principal research scientist in the Particle Physics Collaboration; Josh Bendavid, a former research scientist; and Christoph Paus, a professor of physics at MIT and principal investigator with the Particle Physics Collaboration.

Piecing together

The W boson was first discovered in 1983 and is predicted to be the fourth heaviest among all the fundamental particles. Multiple experiments have aimed to narrow in on the particle’s mass, with varying degrees of precision. For the most part, these experiments have produced measurements that agree with the Standard Model’s predictions. The 2022 measurement by Fermilab’s CDF experiment is the one significant outlier. It also happens to be the most precise experiment to date.

“If you take the CDF measurement at face value, you would say there must be physics beyond the Standard Model,” says co-author Christoph Paus. “And of course that was the big mystery.”

Paus and his colleagues sought to either support or refute the CDF’s findings by making an independent measurement, with an experiment that matches CDF’s precision. Their new W boson mass measurement is a product of 10 years’ worth of work, both to analyze actual particle collision events and to simulate all the scenarios that could produce those events.

For their new study, the physicists analyzed proton collision events that were produced at the LHC in 2016. When it is running, the particle collider generates proton collisions at a furious rate of about one every 25 nanoseconds. The team analyzed a portion of the LHC’s 2016 dataset that encompasses billions of proton-proton collisions. Among these, they identified about 100 million events that produced a very short-lived W boson.

“A particle like the W boson exists for a teeny tiny moment — something like 10-24 seconds — before decaying to two particles, one of which is a neutrino that can’t be measured directly,” Long explains. “That’s the tricky part: You have to measure the other particle — a muon — really well, and be able to piece things together with only one piece of the puzzle.”

Gathering momentum

When a muon is produced from the decay of a W boson, it carries half of the W boson’s mass, which is converted into momentum that carries the muon away from the original collision. Due to the strong magnetic field inside the CMS detector, the electrically charged muon follows a path whose curvature is a function of its momentum. Scientists’ challenge is to track the muon’s path and every interaction it may have with other particles and its surroundings, in order to estimate its initial momentum.

The muon’s momentum is also influenced by the momentum of the W boson before it decays. Decoding the impact of the W boson’s motion from the effects of its mass presented a major challenge. To infer the W boson mass, the team first carried out simulations of every scenario they could think of that a muon might experience after a proton-proton collision in the chaotic environment of the particle collider. In all, the team produced 4 billion such simulated events described by state-of-the-art theoretical calculations. The simulations encoded diverse hypotheses about how the muon momentum is affected by the physical features of the CMS detector, as well as uncertainties in the predictions that govern W boson production in LHC collisions.

The researchers compared their simulations with data from the 2016 LHC run. For every proton-proton collision event that occurs in the collider, scientists can use the CMS detector at CERN’s LHC to precisely measure the energy and momentum of resulting particles such as muons. The team analyzed CMS measurements of muons that were produced from over 100 million W boson events. They then overlaid this data onto their simulations of the muon momentum, which they then converted to a new mass for the W boson.

That mass — 80360.2 ± 9.9 megaelectron volts — is significantly lighter than the CDF experiment’s measurement. What’s more, the new estimate is within the range of what the Standard Model predicts for the W boson’s mass, bolstering physicists’ confidence in the Standard Model and its descriptions of the major particles and forces of nature.

“With the combination of our really precise result and other experiments that line up with the Standard Model’s predictions, I think that most people would place their bets on the Standard Model,” Long says. “Though I do think people should continue doing this measurement. We are not done.”

“We want to add more data, make our analysis techniques more precise, and basically squeeze the lemon a little harder. There is always some juice left,” Paus adds. “With a better look, then we can say for certain whether we truly understand this one fundamental building block.”

This work was supported, in part, by multiple funding agencies, including the U.S. Department of Energy, and the SubMIT computing facility, sponsored by the MIT Department of Physics. 

MIT.nano has announced that 16 startups became active participants in its START.nano program in 2025, more than doubling the number of new companies from the previous year. Aimed at speeding the transition of hard-tech innovation to market, START.nano supports new ventures through the discounted use of MIT.nano shared facilities and a guided access to the MIT innovation ecosystem. The newly engaged startups are developing solutions for some of the world’s greatest challenges in health, climate, energy, semiconductors, novel materials, and quantum computing.

“The unique resources of MIT.nano enable not just the foundational research of academia, but the translation of that research into commercial innovations through startups,” says START.nano Program Manager Joyce Wu SM ’00, PhD ’07. “The START.nano accelerator supports early-stage companies from MIT and beyond with the tools and network they need for success.”

Launched in 2021, START.nano aims to increase the survival rate of hard-tech startups by easing their journey from the lab to the real world. In addition to receiving access to MIT.nano’s laboratories, program participants are invited to present at startup exhibits at MIT conferences, and in exclusive events including the newly launched PITCH.nano competition.

“For an early-stage startup working at the frontier of superconductor discovery, the combination of infrastructure and community has been irreplaceable,” says Jason Gibson, CEO and co-founder of Quantum Formatics. “START.nano isn’t just a resource,” adds Cynthia Liao MBA ’24, CEO and co-founder of Vertical Semiconductor. “It’s a strategic advantage that accelerates our roadmap, allowing us to iterate quickly to meet customer needs and strengthen our competitive edge.”

Although an MIT affiliation is not required, five of the 16 companies in the new cohort are led by MIT alumni, and an additional three have MIT affiliation. In total, 49 percent of the startups in START.nano are founded by MIT graduates.

Here are the intended impacts of the 16 new START.nano companies:

Acorn Genetics is developing a "smartphone of sequencing," launching the power of genetic analysis out of slow, centralized labs and into the hands of consumers for fast, portable, and affordable sequencing.

Addis Energy leverages oil, gas, and geothermal drilling technologies to unlock the chemical potential of iron-rich rocks. By injecting engineered fluids, they harness the earth’s natural energy to produce ammonia that is both abundant and cost-effective.

Augmend Health uses virtual reality and AI to deliver clinical data intelligence services for specialty care that turns incomplete documentation into revenue, compliance, and better treatment decisions.

Brightlight Photonics is building high-performance laser infrastructure at chip scale, integrating Titanium:Sapphire gain to deliver broadband, high-power, low-noise optical sources for advanced photonic systems.

Cahira Technologies is creating the new paradigm of brain-computer symbiosis for treating intractable diseases and human augmentation through autonomous, nonsurgical neural implants.

Copernic Catalysts is leveraging computational modeling to develop and commercialize transformational catalysts for low-cost and sustainable production of bulk chemicals and e-fuels.

Daqus Energy is unlocking high-energy lithium-ion batteries using critical metal-free organic cathodes.

Electrified Thermal Solutions is reinventing the firebrick to electrify industrial heat.

Guardion is making analytical instruments, chemical detectors, and radiation detectors more sensitive, portable, and easier to scale with nanomaterial-based ion detectors.

Mantel Capture is designing carbon capture materials to operate at the high temperatures found inside boilers, kilns, and furnaces — enabling highly efficient carbon capture that has not been possible until now.

nOhm Devices is developing highly-efficient cryogenic electronics for quantum computers and sensors.

Quantum Formatics is speeding discovery of the world’s next superconductors using proprietary AI.

Qunett is building the foundational hardware stack for deployable quantum networks to power the next era of global connectivity.

Rheyo is developing new ways to make dental care more effective, efficient, and easy through advanced materials and technology.

Vertical Semiconductor is commercializing high-voltage, high-density, high-efficiency vertical GaN (gallium nitride) to power the next era of compute.

VioNano Innovations is developing specialty material solutions that reduce variability and improve precision in semiconductor manufacturing, allowing chipmakers to build even smaller, faster, and more cost-effective chips.

START.nano now comprises over 32 companies and 11 graduates — ventures that have moved beyond the prototyping stages, and some into commercialization. See the full list here.

A significant challenge for researchers in materials science and drug discovery is that even the most minor change to a molecule’s structure can completely alter its function. Historically, making these adjustments meant researchers had to re-synthesize the target molecule from scratch — a time-consuming and expensive bottleneck akin to tearing down a house just to move a lamp.

In an exciting discovery recently published in Nature, MIT chemists led by Professor Alison Wendlandt have developed a precision technique that allows scientists to seamlessly relocate alcohol functional groups from one spot on a molecule to a neighboring site. This process bypasses the need to rebuild the entire structure and is the result of a multi-year collaboration with Bristol Myers Squibb.

Functional group repositioning

Using a special light-sensitive molecule called decatungstate as a catalyst, the reaction triggers a highly controlled “migration” of the alcohol group. The process is remarkably predictable, ensuring the molecule retains its precise 3D shape and orientation throughout the move.

The ability to implement subtle structural tweaks without the waste of “from-scratch” synthesis eliminates a primary hurdle that has long plagued the field. Furthermore, because the reaction is gentle enough to work on complex, nearly finished structures, it serves as a powerful fine-tuning tool for late-stage drug candidates.

Precision editing to unlock new chemical designs

When combined with existing chemical methods, this tool provides new pathways to create challenging molecular architectures and oxygenation patterns that were previously out of reach.

“This alcohol migration strategy allows for precise, molecular-level tuning of oxygen atom positions,” says Qian Xu, the co-first author of the paper and a postdoc in the Wendlandt Group. “With predictable stereo- and regioselectivity and late-stage operability, it presents an enticing chance to modify natural products and drug molecules through ‘editing.’”

Ultimately, this precision editing tool holds the potential to dramatically improve the efficiency of molecular design campaigns, accelerating the development of new pharmaceuticals, materials, and agrochemicals.

In addition to Wendlandt and Xu, MIT contributors include co-lead author and graduate student Yichen Nie, recent postdoc Ronghua Zhang, and professor of chemistry Jeremiah A. Johnson. Other authors include Jacob-Jan Haaksma of the University of Groningen in The Netherlands; Natalie Holmberg-Douglas, Farid van der Mei, and Chloe Williams of of Bristol Myers Squibb; and Paul M. Scola of Actithera.

Cells rely on tiny molecules called microRNAs to tune which genes are active and when. Cells must carefully control the lifespan of microRNAs to prevent widespread disruption to gene regulation.

A new study led by researchers at MIT’s Whitehead Institute for Biomedical Research and Germany’s Max Planck Institute of Biochemistry reveals how cells selectively eliminate certain microRNAs through an unexpectedly intricate molecular recognition system. The open-access work, published on March 18 in Nature, shows that the process requires two separate RNA signals, similar to how many digital systems require two forms of identity verification before granting access.

The findings explain how cells use this “two-factor authentication” system to ensure that only intended microRNAs are destroyed, leaving the rest of the gene regulation machinery in operation.

MicroRNAs are short strands of RNA that help control gene expression. Working together with a protein called Argonaute, they bind to specific messenger RNAs — the molecules that carry genetic instructions from DNA to the cell’s protein-making machinery — and trigger their destruction. In this way, microRNAs can reduce the production of specific proteins.

While scientists recognized that microRNAs could be destroyed through a pathway known as target-directed microRNA degradation, or TDMD, the details of how cells recognized which microRNAs to eliminate remained unclear.

“We knew there was a pathway that could target microRNAs for degradation, but the biochemical mechanism behind it wasn’t understood,” says MIT Professor David Bartel, a Whitehead Institute member and co-senior author of the study.

Earlier work from Bartel’s lab and others had identified a key player in this pathway: the ZSWIM8 E3 ubiquitin ligase. E3 ubiquitin ligases are involved in the cell’s recycling system and attach a small molecular tag called ubiquitin to other proteins, marking them for destruction.

The researchers first showed that the ZSWIM8 E3 ligase specifically binds and tags Argonaute, the protein that holds microRNAs and helps regulate genes. The researchers’ next challenge was to understand how this machinery recognized only Argonaute complexes carrying specific microRNAs that should be degraded.

The answer turned out to be surprisingly sophisticated.

Using a combination of biochemistry and cryo-electron microscopy — an imaging technique that reveals molecular structures at near-atomic resolution — the researchers discovered that the degradation system relies on a dual-RNA recognition process. First, Argonaute must carry a specific microRNA. Second, another RNA molecule called a “trigger RNA” must bind to that microRNA in a particular way.

The degradation machinery activates only when both signals are present.

This dual requirement ensures exquisite specificity. Each cell contains over a hundred thousand Argonaute–microRNA complexes regulating many genes, and destroying them indiscriminately would disrupt essential biological processes.

“The vast majority of Argonaute molecules in the cell are doing useful work regulating gene expression,” says Bartel, who is a professor of biology at MIT and also a Howard Hughes Medical Institute investigator. “You only want to degrade the ones carrying a particular microRNA and bound to the right trigger RNA. Without that specificity, the cell would lose its microRNAs and the essential regulation that they provide.”

The structural images revealed complex molecular interactions. The ZSWIM8 ligase detects multiple structural changes that occur when the two RNAs bind together within the Argonaute protein.

“When we saw the structure, everything clicked,” says Elena Slobodyanyuk, a graduate student in Bartel’s lab and co-first author of the study. “You could see how the pairing of the trigger RNA with the microRNA reshapes the Argonaute complex in a way that the ligase can recognize.”

Beyond explaining how TDMD works, the findings may impact how scientists think about the regulation of RNA molecules more broadly.

“A lot of E3 ligases recognize their targets through simpler signals,” says Jakob Farnung, co-first author and researcher in the Department of Molecular Machines and Signaling at the Max Planck Institute of Biochemistry. “It was like opening a treasure chest where every detail revealed something new and mesmerizing.”

MicroRNAs typically persist in cells for much longer time periods than most messenger RNAs, but some degrade far more quickly, and the TDMD pathway appears to account for many of these unusually short-lived microRNAs.

The researchers are now investigating whether other RNAs can trigger similar degradation pathways and whether additional microRNAs are regulated through variations of the mechanism shown in this study.

“This opens up a whole new way of thinking about how RNA molecules can control protein degradation,” says Brenda Schulman, study co-senior author and director of the Department of Molecular Machines and Signaling at the Max Planck Institute of Biochemistry. “Here, the recognition was far more elaborate than expected. There’s likely much more left to discover.”

Uncovering the details of this intricate regulatory system required interdisciplinary collaboration, combining expertise in RNA biochemistry, structural biology, and ubiquitin enzymology to solve this long-standing molecular puzzle.

“This was a project that required the strengths of two labs working at the forefront of their fields,” says Schulman, who is also an alum of Whitehead Institute. “It was an incredible team effort.”

Chronic wound infections are notoriously difficult to manage because some bacteria can actively interfere with the body’s immune defenses. In wounds, Enterococcus faecalis (E. faecalis) is particularly resilient — it can survive inside tissues, alter the wound environment, and weaken immune signals at the injury site. This disruption creates conditions where other microbes can easily establish themselves, resulting in multi-species infections that are complex and slow to resolve. Such persistent wounds, including diabetic foot ulcers and post-surgical infections, place a heavy burden on patients and health care systems, and sometimes lead to serious complications such as amputations.

Now, researchers have discovered how E. faecalis releases lactic acid to acidify its surroundings and suppresses the immune-cell signal needed to start a proper response to infection. By silencing the body’s defenses, the bacterium can cause persistent and hard-to-treat wound infections. This explains why some wounds struggle to heal, even with treatment, and why infections involving multiple bacteria are especially difficult to eradicate.

The work was led by researchers from the Singapore-MIT Alliance for Research and Technology (SMART) Antimicrobial Resistance (AMR) interdisciplinary research group, alongside collaborators from the Singapore Centre for Environmental Life Sciences Engineering at Nanyang Technological University (NTU Singapore), MIT, and the University of Geneva in Switzerland.

In a paper titled “Enterococcus faecalis-derived lactic acid suppresses macrophage activation to facilitate persistent and polymicrobial wound infections,” recently published in Cell Host & Microbe, the researchers documented how E. faecalis releases large amounts of lactic acid during infection. This acidity suppresses the activation of macrophages — immune cells that normally help to clear infections — and interferes with several important internal processes that help the cell recognize and respond to infection. As a result, the mechanisms that cells rely on to send out “danger” signals are suppressed, leaving the macrophages unable to fully activate.

Researchers found that E. faecalis uses a two‑step mechanism to achieve this. Lactic acid enters the macrophages through a lactate transporter called MCT‑1 and also binds to a lactate-sensing receptor, GPR81, on the cell surface. By engaging both pathways, the bacterium effectively shuts down downstream immune signalling and blocks the macrophage’s inflammatory response, allowing E. faecalis to persist in the wound much longer than it should. Specifically, the lactic acid prevents a key immune alarm signal, known as NF-κB, from switching on inside these cells.

This was proven in a mouse wound model, where strains of E. faecalis that could not make lactic acid were cleared much more quickly, and the wounds also showed stronger immune activity. In wounds infected with both E. faecalis and Escherichia coli, the weakened immune response caused by lactic acid also allowed E. coli to grow better. This explains why wound infections often involve multiple species of bacteria and become harder to treat over time, particularly since E. faecalis is among the most common bacteria found in chronic wounds.

“Chronic wound infections often fail not because antibiotics are powerless, but because the immune system has effectively been ‘switched off’ at the infection site. We found that E. faecalis floods the wound with lactic acid, lowering pH and muting the NF‑κB alarm inside macrophages — the very cells that should be calling for help. By pinpointing how acidity rewires immune signalling, we now have clear targets to reactivate the immune response,” says first author Ronni da Silva, research scientist at SMART AMR, former postdoc in the lab of co-author and MIT professor of biology Jianzhu Chen, and SCELSE-NTU visiting researcher.

“This discovery strengthens our understanding of host-pathogen interactions and offers new directions for developing treatments and wound care that target the bacteria’s immunosuppressive strategies. By revealing how the immune response is shut down, this research may help improve infection management and support better recovery outcomes for patients, especially those with chronic wounds or weakened immunity,” says Kimberly Kline, principal investigator at SMART AMR, SCELSE-NTU visiting academic, professor at the University of Geneva, and corresponding author of the paper.

By identifying lactic‑acid‑driven immune suppression as a root cause of persistent wound infections, this work highlights the potential of treatment approaches that support the immune system, rather than rely on antibiotics alone. This could lead to therapies that help wounds heal more reliably and reduce the risk of complications. Potential directions include reducing acidity in the wound or blocking the signals that lactic acid uses to switch off immune cells.

Building on their study, the researchers plan to explore validation in additional pathogens and human wound samples, followed by assessments in advanced preclinical models ahead of any potential clinical trials.

The research was partially supported by the National Research Foundation Singapore under its Campus for Research Excellence and Technological Enterprise program.

U.S. News and World Report has again placed MIT’s graduate program in engineering at the top of its annual rankings, released today. The Institute has held the No. 1 spot since 1990, when the magazine first ranked such programs.

The MIT Sloan School of Management also placed highly, occupying the No. 6 spot for the best graduate business programs.

Among individual engineering disciplines, MIT placed first in six areas: aerospace/aeronautical/astronautical engineering, chemical engineering, computer engineering (tied with the University of California at Berkeley), electrical/electronic/communications engineering (tied with Stanford University and Berkeley), materials engineering, and mechanical engineering. It placed second in nuclear engineering.

In the rankings of individual MBA specialties, MIT placed first in four areas: business analytics, entrepreneurship (with Stanford), production/operations, and supply chain/logistics. It placed second in executive MBA programs (with the University of Chicago).

U.S. News bases its rankings of graduate schools of engineering and business on two types of data: reputational surveys of deans and other academic officials, and statistical indicators that measure the quality of a school’s faculty, research, and students. The magazine’s less-frequent rankings of graduate programs in the sciences, social sciences, and humanities are based solely on reputational surveys.

In the sciences, ranked by U.S. News for the first time in four years, MIT’s doctoral programs placed first in four areas: biology (with Scripps Research Institute), chemistry (with Berkeley and Caltech), computer science (with Carnegie Mellon University and Stanford), and physics (with Caltech, Princeton University, and Stanford). The Institute placed second in mathematics (with Harvard University, Stanford, and Berkeley).

To improve data center efficiency, multiple storage devices are often pooled together over a network so many applications can share them. But even with pooling, significant device capacity remains underutilized due to performance variability across the devices.

MIT researchers have now developed a system that boosts the performance of storage devices by handling three major sources of variability simultaneously. Their approach delivers significant speed improvements over traditional methods that tackle only one source of variability at a time.

The system uses a two-tier architecture, with a central controller that makes big-picture decisions about which tasks each storage device performs, and local controllers for each machine that rapidly reroute data if that device is struggling.

The method, which can adapt in real-time to shifting workloads, does not require specialized hardware. When the researchers tested this system on realistic tasks like AI model training and image compression, it nearly doubled the performance delivered by traditional approaches. By intelligently balancing the workloads of multiple storage devices, the system can increase overall data center efficiency.

“There is a tendency to want to throw more resources at a problem to solve it, but that is not sustainable in many ways. We want to be able to maximize the longevity of these very expensive and carbon-intensive resources,” says Gohar Chaudhry, an electrical engineering and computer science (EECS) graduate student and lead author of a paper on this technique. “With our adaptive software solution, you can still squeeze a lot of performance out of your existing devices before you need to throw them away and buy new ones.”

Chaudhry is joined on the paper by Ankit Bhardwaj, an assistant professor at Tufts University; Zhenyuan Ruan PhD ’24; and senior author Adam Belay, an associate professor of EECS and a member of the MIT Computer Science and Artificial Intelligence Laboratory. The research will be presented at the USENIX Symposium on Networked Systems Design and Implementation.

Leveraging untapped performance

Solid-state drives (SSDs) are high-performance digital storage devices that allow applications to read and write data. For instance, an SSD can store vast datasets and rapidly send data to a processor for machine-learning model training.   

Pooling multiple SSDs together so many applications can share them improves efficiency, since not every application needs to use the entire capacity of an SSD at a given time. But not all SSDs perform equally, and the slowest device can limit the overall performance of the pool.

These inefficiencies arise from variability in SSD hardware and the tasks they perform.

To utilize this untapped SSD performance, the researchers developed Sandook, a software-based system that tackles three major forms of performance-hampering variability simultaneously. “Sandook” is an Urdu word that means “box,” to signify “storage.”

One type of variability is caused by differences in the age, amount of wear, and capacity of SSDs that may have been purchased at different times from multiple vendors.

The second type of variability is due to the mismatch between read and write operations occurring on the same SSD. To write new data to the device, the SSD must erase some existing data. This process can slow down data reads, or retrievals, happening at the same time.

The third source of variability is garbage collection, a process of gathering and removing outdated data to free up space. This process, which slows SSD operations, is triggered at random intervals that a data center operator cannot control.

“I can’t assume all SSDs will behave identically through my entire deployment cycle. Even if I give them all the same workload, some of them will be stragglers, which hurts the net throughput I can achieve,” Chaudhry explains.

Plan globally, react locally

To handle all three sources of variability, Sandook utilizes a two-tier structure. A global schedular optimizes the distribution of tasks for the overall pool, while faster schedulers on each SSD react to urgent events and shift operations away from congested devices.

The system overcomes delays from read-write interference by rotating which SSDs an application can use for reads and writes. This reduces the chance reads and writes happen simultaneously on the same machine.

Sandook also profiles the typical performance of each SSD. It uses this information to detect when garbage collection is likely slowing operations down. Once detected, Sandook reduces the workload on that SSD by diverting some tasks until garbage collection is finished.

“If that SSD is doing garbage collection and can’t handle the same workload anymore, I want to give it a smaller workload and slowly ramp things back up. We want to find the sweet spot where it is still doing some work, and tap into that performance,” Chaudhry says.

The SSD profiles also allow Sandook’s global controller to assign workloads in a weighted fashion that considers the characteristics and capacity of each device.

Because the global controller sees the overall picture and the local controllers react on the fly, Sandook can simultaneously manage forms of variability that happen over different time scales. For instance, delays from garbage collection occur suddenly, while latency caused by wear and tear builds up over many months.

The researchers tested Sandook on a pool of 10 SSDs and evaluated the system on four tasks: running a database, training a machine-learning model, compressing images, and storing user data. Sandook boosted the throughput of each application between 12 and 94 percent when compared to static methods, and improved the overall utilization of SSD capacity by 23 percent.

The system enabled SSDs to achieve 95 percent of their theoretical maximum performance, without the need for specialized hardware or application-specific updates.

“Our dynamic solution can unlock more performance for all the SSDs and really push them to the limit. Every bit of capacity you can save really counts at this scale,” Chaudhry says.

In the future, the researchers want to incorporate new protocols available on the latest SSDs that give operators more control over data placement. They also want to leverage the predictability in AI workloads to increase the efficiency of SSD operations.

“Flash storage is a powerful technology that underpins modern datacenter applications, but sharing this resource across workloads with widely varying performance demands remains an outstanding challenge. This work moves the needle meaningfully forward with an elegant and practical solution ready for deployment, bringing flash storage closer to its full potential in production clouds,” says Josh Fried, a software engineer at Google and incoming assistant professor at the University of Pennsylvania, who was not involved with this work.

This research was funded, in part, by the National Science Foundation, the U.S. Defense Advanced Research Projects Agency, and the Semiconductor Research Corporation.