Posts

int main(void) { printf("hello, world (again)\n"); return 0; /* 8 years later */ } Second attempt at blogging. Let’s see if this one sticks.

My notes on where AI/ML networking stands in October 2026. This area moves fast and there is a lot of active research going on, so parts of this will age quickly. Naming is NVIDIA’s (NVLink, NCCL, PXN), AMD and Intel have their own terms for similar things. Traffic For comparison, a normal DC Clos carries lots of small TCP flows. ECMP spreads them fine, TCP recovers from drops, 3:1 oversubscription is ok. ...

Two really great books that will help you get into the basics of SAN : Introduction to Storage Area Networks Fibre Channel over Ethernet (FCoE) Data Center Bridging (DCB) Concepts and Protocols

Just finished the TCP course from **INE ** : CCIE R&S: Understanding Transmission Control Protocol (TCP) Really recommend this to anyone who is interested in a bit deeper dive of how TCP works, great quality content, **Keith Bogart **is doing amazing job as instructor ! Some TCP facts which I didn’t mentioned in my previous post : Before TCP connection is established TCP goes through the following steps: Closed Initial State Active Open (APP notifies the CPU and requests for service time) - client side Transmission Control Block setup process - CPU creates TCB to keep track of each new TCP connection - for each TCP session we create a new TCB with unique identifier (socket - SRC IP; SRC TCP Port; DST IP; DST TCP Port) Initial sequence number will be random to prevent attacks on TCP When TCB is ready we are sending SYN(x random Seq# number) - this is called a Syn Sent State Receiving Syn(y random seq# number from server) + ACK (seq# x + 1) Sending ACK(seq# y+1) for that TCP Client side state changes to Established This is Active Open - TCB creation process, initiated only per need of client. ...

Old known stuff, probably for everybody who is somehow related to networking, but anyway, putting it here too. IEEE 802.3x - Wikipedia link If QoS is enabled and you like to prioritize the traffic, flow control needs to be disabled,as it doesn’t care about any higher level prioritization, just when ingress traffic is coming in faster than receiver can accept it, flow control will kick in and send pause frames until the ingress-egress rate will be equalized or ingress rate is lower than egress of that interface. ...

Installing – Let’s use Debian based manager : Apt-get install open-iscsi Apt-get install iperf Apt-get install nload Scanning for targets : Iscsiadmin -m discovery -t st -p target_ip Iscsiadm -m node -u Iscsiadm -m node –login Check if you see the object : Lsblk New disk should appear in dev - like sdx \-sdx1 How to partition (if wasn’t partitioned yet) Fdisk /dev/sdc N P Enter w Mkfs.ext4 /dev/sdx1 Connecting the drive to system ...

End to end delay is the time from when we sent the first bit of the first line until last bit arrives to destination. Basically everything can be determined except queuing delay - it’s an unpredictable variable.

Packetization delay is basically how close together we can pack a bits. So it’s the time from when the first to the last bit of a packet is transmitted. Formula would be : Packetization delay = P(size) / R(transmit rate)

Propagation delay - l is the amount of time it takes for a single bit to travel over a link at propagation speed c. Formula would be PropagationDelay=l(cable lenght)/c(speed)

As we know TCP is called reliable - connection-oriented protocol. But why ? Basically because it keeps the data in its buffer before and after sending it, it makes sure that data was delivered in sequence and receiver send the confirmation about received data (ack). Otherwise data will be re-transmitted there are different ways to re-send the data, let’s explore some of them. So what is happening under the hood of this massive (big overhead comparing to UDP protocol). ...