0:04
Welcome back to Altitude Addicts, where
0:06
we dive into high-tech engineering and
0:09
strategic infrastructure.
0:11
Today we will examine a significant
0:13
milestone in Russian high performance
0:15
computing at Moscow State University.
0:18
We will break down how a 48 core
0:20
domestic processor, specialized servers,
0:23
and softwaredefined storage were
0:25
combined into an operational supercomput
0:28
network. how the system keeps massive
0:30
scientific data sets safe from hardware
0:32
crashes and what this all means for
0:34
technological self-reliance heading into
0:37
2027. So stick around until the end.
0:41
Russia has officially completed
0:42
deploying a domestically developed data
0:45
storage system for Moscow State
0:47
University's supercomput infrastructure.
0:50
This marks another major step in
0:52
integrating Russian designed processors,
0:55
server hardware and software into high
0:58
performance computing. The SDS TRO
1:01
softwaredefined storage platform
1:03
developed by the Astro Group integrates
1:05
directly with servers manufactured by
1:07
Spintronica and powered by 48 core BICE
1:13
This new installation forms a core part
1:15
of the university's IST hardware
1:17
software complex which has been
1:19
operating since 2026.
1:22
The project was officially announced on
1:24
November 9th, 2026 following the
1:27
completion of development and
1:28
integration work by Astro Group and
1:30
Spintronica backed by financial support
1:33
from Russia's Ministry of Science and
1:38
The system provides critical storage
1:40
infrastructure for heavy scientific
1:42
computing workloads. It also enables
1:45
researchers to access their stored data
1:48
seamlessly via the standard network file
1:52
What makes this project particularly
1:54
significant is that the university
1:57
specifically required compatibility with
1:59
Russian processors and the use of
2:02
domestically developed components.
2:04
Rather than simply taking an existing
2:07
foreign storage platform and trying to
2:09
make it fit, the participating companies
2:11
prioritized building native
2:13
compatibility between the storage
2:15
software and BOL s server hardware from
2:20
To understand how this works in
2:22
practice, we have to look closely at SDS
2:25
truck, a storage platform built entirely
2:27
around Russian hardware.
2:30
Astro Group engineered SDS truck as a
2:32
softwaredefined storage solution. Unlike
2:36
traditional storage systems that rely on
2:38
dedicated proprietary hardware boxes,
2:41
softwaredefined storage uses software to
2:43
manage storage resources distributed
2:45
across multiple standard servers.
2:48
Think of it like pooling a bunch of
2:50
individual hard drives across different
2:52
machines into one giant smart digital
2:55
warehouse that an administrator can
2:57
expand just by plugging in another
3:00
server. That is the basic magic here and
3:03
it gives engineers massive flexibility
3:05
without locking them into rigid hardware
3:09
For the Moscow State University project,
3:12
SDS TRO was deployed on servers designed
3:15
and built by Spintronica using BLE S
3:17
chips. Because the system serves the IST
3:20
computing complex, users access shared
3:23
data through NFS, a standard file
3:25
sharing protocol widely used in Linux
3:29
As a result, researchers can tap into
3:32
the storage infrastructure immediately
3:34
without having to learn or implement a
3:37
brand new data access procedure.
3:40
Financial backing from the Ministry of
3:42
Science and Higher Education was
3:44
essential to getting this off the
3:46
ground. Moscow State University ran a
3:49
competitive program to secure this
3:51
funding, aiming to modernize the
3:54
material and technical infrastructure of
3:56
organizations conducting serious
3:58
scientific research and development.
4:01
Making all these pieces talk to each
4:03
other was a critical engineering hurdle.
4:07
Teams established a dedicated software
4:09
build and assembly pipeline followed by
4:12
two strict phases of testing. The first
4:15
stage validated software compatibility
4:18
using virtual machines while the second
4:20
stage moved to testing physical servers
4:22
equipped with real BOL s processors.
4:26
According to the developers, both
4:28
testing stages demonstrated stable
4:30
operation and met all necessary
4:32
technical specifications.
4:35
Even better, the main SDS truck codebase
4:37
has now been updated to include these
4:39
BEL s compatibility modifications.
4:43
This means the engineering work is not
4:45
just a one-off custom job for Moscow
4:48
State University because those exact
4:50
enhancements are now available to any
4:52
customer deploying the platform on
4:54
Russian processor architectures. Of
4:58
course, software is only half the
5:00
equation, which brings us directly to
5:02
the BLE S, the 48 core processor
5:05
powering the entire system. Known
5:08
technically as the BES1000,
5:10
this chip is the foundational hardware
5:13
component inside the storage servers.
5:16
Created by Bol Electronics, it was
5:18
specifically designed for storage
5:20
infrastructure, high performance
5:22
computation systems, and enterprise
5:26
Under the hood, the processor implements
5:28
the ARM Va architecture, packing 48 ARM
5:31
Cortex A75 cores running at a frequency
5:38
Its memory subsystem supports up to 768
5:41
GB per processor across six DDR43200
5:46
channels with error correcting code
5:49
to handle high-speed storage controllers
5:52
and peripherals. It also provides up to
5:54
80 PCI Express 4.0 lanes and 32 MGB of
6:01
All of this runs at a specified thermal
6:03
design power of roughly 120 watts.
6:07
Throwing 48 cores and 80 high-speed
6:10
lanes onto a single chip at 120 watts is
6:13
no small feat. And it is the kind of raw
6:16
connectivity that makes or breaks a
6:18
modern data center. These specific
6:21
attributes make the BL S particularly
6:23
well suited for infrastructure projects
6:26
demanding heavy memory bandwidth, wide
6:28
connectivity, and serious parallel
6:32
Still, getting the processor to boot is
6:35
only one piece of the engineering puzzle
6:37
for a storage platform.
6:39
The hardware must also cleanly support
6:41
the operating system, storage management
6:44
software, network interfaces, and data
6:47
replication mechanisms.
6:49
Consequently, the Moscow State
6:51
University deployment proves that the
6:53
processor can function reliably as part
6:55
of a complex living infrastructure
6:58
rather than just sitting on a test bench
7:00
as a standalone server. Before we dive
7:03
into how the system protects
7:05
irreplaceable research data from sudden
7:07
hardware failures and look at where else
7:10
this chip is showing up, let us take a
7:13
quick breather. If you enjoy deep dives
7:16
into complex engineering and strategic
7:18
technology like we cover here on
7:20
Altitude Addicts, make sure you are
7:23
subscribed so you do not miss our
7:27
Coming up next, we will examine the
7:29
automated failover tech keeping
7:31
supercomput calculations alive and
7:33
explore the wider ecosystem adopting
7:38
Protecting valuable scientific data is
7:40
where DRBD replication steps in to save
7:44
In fact, the data protection strategy
7:47
built into the SDS truck installation is
7:50
easily one of its most critical
7:53
Scientific computing systems frequently
7:55
run massive calculations for days or
7:58
weeks on end, and losing stored results
8:00
to a sudden server crash can waste
8:02
immense computing time and force
8:04
researchers to start over from scratch.
8:08
SDS truck solves this vulnerability by
8:11
using distributed replicated block
8:13
device technology or DRBD to handle
8:16
synchronous replication.
8:18
Under this setup, a write operation is
8:20
only acknowledged as complete after the
8:23
data has been safely written to the
8:25
required number of separate nodes. If
8:28
one server suddenly dies, the exact
8:30
replicated data remains immediately
8:33
accessible on the surviving nodes.
8:36
The system also automates the recovery
8:38
process when things go wrong. As soon as
8:41
a failed server is repaired and brought
8:43
back online, the storage cluster
8:45
triggers something called delta
8:49
Instead of clogging the network by
8:51
retransmitting the entire multi-ter data
8:54
set, it copies over only the specific
8:57
data blocks that changed during the
9:00
This smart approach drastically cuts
9:03
down network traffic and lets the
9:05
recovered node rejoin the cluster's
9:07
fault tolerant state much faster.
9:10
Another smart design choice is how the
9:13
system divides responsibilities across
9:15
different types of server nodes.
9:18
Dedicated controller nodes handle the
9:20
overall configuration and coordination
9:22
of the storage environment while worker
9:25
nodes focus purely on managing the DRBD
9:30
Separating control from heavy data
9:32
lifting allows the system to scale
9:34
outwards smoothly while keeping
9:36
administration centralized.
9:39
Simply put, if a drive drops dead at
9:41
3:00 in the morning, nobody has to
9:43
sprint into the server room in their
9:45
pajamas to manually rebuild arrays.
9:48
According to the developers, researchers
9:50
accessing NFS volumes should barely
9:53
notice a hiccup while their previously
9:55
saved calculation results remain
10:00
Beyond Moscow State University, the BCL
10:03
S is steadily expanding its footprint
10:05
across Russian digital infrastructure.
10:08
This university installation is actually
10:10
just one piece of a much broader push to
10:13
establish byol s as a foundational
10:15
platform for domestic servers.
10:18
Several other Russian hardware and
10:20
software developers have already jumped
10:22
on board with its 48 core architecture.
10:26
For example, the company Aquarius has
10:28
delivered its S2-1 storage
10:30
infrastructure built on its AQ2BS1000
10:34
server and powered by BLE S chips. that
10:37
particular system is tailored for
10:39
protected IT environments including
10:42
government agencies and critical
10:44
national information infrastructure.
10:47
Another solid example is Graviton's
10:51
server which packs two 48 core BCOL s
10:54
processors into a single chassis.
10:57
Graviton engineered that dual socket
10:59
machine specifically for heavy
11:01
higherformance computing and artificial
11:04
intelligence workloads.
11:06
Astra Group has also integrated BYOL s
11:09
across other corners of its own product
11:11
lineup. They commercially launched Astra
11:14
cloud on this exact processor
11:16
architecture and they also adapted their
11:19
Tantor X data 2B database machine to run
11:22
their internal 1C enterprise
11:26
With the hardware and storage foundation
11:28
firmly in place, the IST complex is now
11:31
moving rapidly toward full-scale
11:33
scientific computing. Operating at
11:36
Moscow State University since 2026,
11:39
ISTOK is actively used by internal
11:41
university research teams alongside more
11:44
than 10 organizations affiliated with
11:46
institutes of the Russian Academy of
11:50
The complex is built to tackle
11:52
computationally brutal fields including
11:54
environmental studies, biomedical
11:56
research, cognitive science, and
11:58
advanced climate modeling. Throughout
12:01
every simulation, these applications
12:04
generate staggering volumes of data that
12:06
must be continuously stored, rapidly
12:09
accessed, and rigorously protected.
12:12
We are entering the final stretch of our
12:14
analysis where we will look at the real
12:16
world speed gains researchers are
12:18
already seeing and test whether this
12:20
domestic ecosystem can truly stand on
12:23
its own. Stick with us right to the
12:26
finish as we wrap up what this means for
12:31
Preliminary performance evaluations from
12:33
the university suggest that climate
12:36
modeling tasks are already finishing
12:38
significantly faster than they did on
12:40
older computing resources of comparable
12:43
size. Looking ahead, the university
12:46
plans to transition IST into full-scale
12:49
scientific calculations and major
12:51
research projects beginning in 2027.
12:55
As those workloads ramp up, the storage
12:58
system is going to shoulder an even
13:00
heavier burden. Massive scientific
13:03
simulations demand far more than just
13:05
fast processor clock speeds. They
13:08
require the capacity to preserve massive
13:10
intermediate and final data sets,
13:12
rocksolid storage reliability, and wide
13:15
openen network bandwidth. Because a
13:17
bottleneck or failure in any single link
13:20
can stall the entire computing pipeline.
13:23
Ultimately, this entire deployment
13:25
serves as a highstakes test case for
13:28
Russia's high performance computing
13:32
The final outcome of the SDS truck
13:34
rollout represents a major milestone in
13:36
the country's drive to build a
13:38
comprehensive self-reliant domestic
13:43
in a real world live scientific
13:45
environment. The project successfully
13:47
merges Astro Group's storage software,
13:49
Spintronica server hardware and BL
13:52
processors under one roof. Thanks for
13:55
watching Altitude Addicts and we will
13:57
see you in the next deep dive.
14:02
We thank the people who have shared,
14:04
subscribed and hyped the videos.