04. IaaS

• logical data center in a region

• -> isolated from failures of other availabilty zones

  • -> infrastructural redundancy… as no common points of failures..

• -> inexpensive, low-latency network connectivity to other avaibailbty zones in same AWS region

• geographic cluster of availability zones

• regions isolated from eahc other for fault tolerance and satbility

• users can control where resources are allocated

• -> i.e. meet legal requirements such as in europe

• -> have short latency access for customers…

• yes

• zones in same region -> free

• region to region -> cost

• region level

  • guarantee 99.99% region availabilty

  • region is unavailable when all running instances or tasks, deployed in two or more AZ, concurrently have no external connectivity

• instance-level SLAs

  • guarantee 99.5% reachability of an EC2 instance

  • instance not available when your single EC2 instance has no external connectivity

• vitrual machines running inside amazon cloud

• ephermal storage tied to the virtual machine (node)

• network accessible block storage that persists across time and can be mounted in the VM

• virtual firewall to secure your network in the cloud

• based on xen hypervisor

• bare metal hypervisor

  
  

• One VM is called Domain 0 (DOM0) and runs the host OS.

• It

  • starts first and runs the Xen management software,

  • manages other VMs,

  • has drivers for hardware

  • and provides virtual disks and network access to unprivileged VMs.

• xen has Network I/O, CPU interrupts and storage IO in software

• in nitro, tese are in hardware

  • hardware is faster

  • as Dom0 can be removed

  • no cores reserved for Dom0

• also called VM template

• copy of a server with OS and preinstalled software

• predefined from amazon and third parties; user defined AMIs also possible

• stored in Amazon S3

• difficult to select

• -> could contain trojans or backdoors

• -> amazon provides reviews and ratings…

• persistent block-level storage volumes

  • can be attached to EC2 instances.

• independent of EC2 instances

  • can be detached and reattached to different instances.

• It offers different volume types, each optimized for different performance characteristics and costs.

• survive the termination of the EC2 instance.

• EBS is suitable for applications that require persistent storage, databases, and boot volumes for EC2 instances.

• local, temporary storage that is directly attached to an EC2 instance.

• It provides high-performance, low-latency storage specific to the instance it is attached to.

• The storage is transient and data is lost if the instance is stopped or terminated.

• deal for temporary data, caching, and scratch space where durability or persistence is not required.

• scalable, shared file storage service

  • can be accessed by multiple EC2 instances simultaneously.

• offers file system interface compatible with standard OSs,

  • supporting both read and write operations.

• automatically scales storage capacity as files are added,

  • suitable for workloads that require shared access and dynamic scaling.

• It is highly available and durable, with data replicated across multiple Availability Zones.

• EFS is well-suited for content management, web serving, data sharing, and containerized applications.

• scalable object storage service designed for storing and retrieving large amounts of unstructured data.

• It offers virtually unlimited storage capacity and is accessible over the internet via API calls.

• S3 provides high durability, availability, and data redundancy, with data distributed across multiple facilities within an AWS Region.

• S3 is commonly used for backup and restore, data archiving, static website hosting, content distribution, and big data analytics.

• processor registers

  • very fast, very expensive

  • small size, small capacity

  • power on, immediate term

• processor cache

  • very fast, very expensive

  • small size, small capacity

  • power on, immediate term

• RAM

  • fast, affordable

  • medium size, medium capacity

  • power on, very short term

• flash / USB memory

  • slower, cheap

  • small size, large capacity

  • power off, short term

• hard drives

  • slow, very cheap

  • large size, very large capacity

  • power off, mid term

• tape backup

  • very slow, affordable

  • large size, very large capacity

  • power off, long term

• Flash or disk drives attached to the computer

• Disks are larger and cheaper but slower and more power hungry

• redundant array of independent (inexpensive) disks

• Raid0

  • split data among disks

  • improve performance (read/write simultaneous)

  • no redundancy

• Raid1

  • duplicates among disks

  • no improved performance

  • redudancy

• Raid5

  • srtip data among disks to improve performance

  • include parity information to revocer in case a drive fails

  • balance between performance, capacity and fault tolerance

• high positive voltage between source and control gate

• injects electrons to floating gate

• apply voltage to control gate

• measure current between source and drain

• presence / absence of charge in floating gate influences conductivity of channel

  
  

• high negative voltage to control gate

• -> block-wise, not cell wise…

• single level cells

  • store one bit

• multi level cells

  • store multiple bits in single cell

  • sense strengh of current, not only presence…

  • -> more precise measurement required

  • -> amount of charge determiens states -> precice control of the charge deposit required

  • higher density, lower cost

  • larger bit-error ration

  • lower write speed, lower number of program-erase cycles, higher power consumption

• pressure on storage

• -> shift importance from €/GB to €/IOPS (in out operations per second)

• thus nowadays storage solutions all flash or hybrid flash array

• all flash:

  • flash or SSD

  • combination with RAID

  • higher IOPS, lower latency, higher bandwidth

  • more expensive

• direct attached storage

  • Storage devices are attached to the individual computer

  • Leads to over-provisioning

• storage area network (SAN)

  • NETWORK providing access to block level data storage.

  • Typically specialized network separated from LAN.

  • No file abstraction, only block-level operations

  • Shared pool of spare resources

• Network attached storage (NAS)

  • Storage devices connected to a file server.

  • Access available at a file-level for other computers.

• Fibre Channels to access the SAN

• iSCSI (Internet small computing systems interface)

• ATA (advanced technology attachment) over Ethernet (AoE) (also cable standard but over ethernet…)

• Storage device visible to client as disk

• can be mounted after formatting with a file system…

=> AS OWN NETWORK -> NEED ACCESS TO HARDWARE OVER ETHERNET…

• SCSI -> standard for computer system interface (i.e. for peripherals such as drives)

• -> i SCSI -> SCSI commands over TCP/IP -> thus able to run over ethernet as basis…

• shared network bandwidth

• shared performance of storage devices

• security concerns due to transfer of data through network…

• flexible distribution of devices between clients

  • -> no adaption of cabling

• easy replacement of fauilty servers with new servers booting from the same unit

• easier disaster protection

• SAN -> hardware access over some network connection (block level…)

• NAS -> dedicated file server (with file system) attached to the network…

  • -> so not usually own dedicated network…

• file server attached to the network

  • -> has file system and provides access to files for clients

  • access to files using network file sharing protocols, e.g. NFS

• frequently using internally a RAID

• computer based

• embedded systems based

• ASIC based

• provide ability to distribute files and meta-data across multiple NAS servers

• => i.e. cluster of NAS as single…

• block virtualization

• file virtualization

• location transparency

• => ability to access data without knowiung the actual locaiton…

• some meta data determine mapping of virtual disk and block number (local in client system)

• to physical disc and block number

• IO redirection based on the metadata…

=> mapping local blocks (virtual) using some meta data to physical blocks in the e.g. SAN

• flexible mapping

  • only use parts of physical disk

• thin provisioning

  • provision e.g. 2 GB, in reality -> reserve 1 and only 2 if actually needed…

• disk expansion and shrinking

• non-disruptive data migration (between disks)

  • -> migrate and simply switch mapping function…

• improved utilization

• host based

• storatge device based

• network based

• host runs virtualization software

• it maps logical to physical units

• disk array, e.g. RAID provides virtualization level

• nes disk arrays provide storage controllers that allow the attachment of other storage controllers

  • primary controller provides pooling and meta-data management

  • and might provide replicatoin and migration services

• virtualizatoin device is in the LAN and connected to a SAN

• -> provides virtualization services

• most frequent implementaiton…

• in-band vs out-band

• file system

  • e.g. NTFS, FAT32, UFS -> stores files on block based storage

• clustered NAS

  • -> combines NAS from the same vendor

• distributed file system

  • allow files located on multiple NAS to appear as if on single NAS (e.g. windows DFS, Linux DFS)

  • distributed file systems can combine NAS from different vendors

• a block storage volume

• can be mounted

• can be formatted appropriate

• multiple can be combined into a virtual RAID

• snapshots of block storage volume stored in S3 -> for backup or replicatoin

• disks attached to physical host -> i.e. directly to machine thus (aside from RAM) lowest latency…

• some uses NVMe or SATA-based SSD -> high random I/O performance

• any data on instance store volumes lost

• 
  

• scalable file storage

• can be created and mounted into instances

• files can be shared among instances

• file system has to be explicitly created and destroyed

• Simple Storage Services

• reliable and inexpensive data storage infrastructure

• -> supports objects from 1 to 5 TB

• two-level namespace

• slow compared to local discs or EBS

• high durability but low availability

• -> most users use S3 for short-term or long-term backup

• buckets:

  • flat colleciton of buckets,, namespace shared across all amazon customers

• objects:

  • file in the buckets

• in EC3

• from the web

• instance -> running VM which is based on an AMI (amazon machine image)

• instance type -> VM with different compute and memory capabilities

• storage:

  • boot device volume

    • elastic block storage

    • instance storage

  • instance store volumes: local discs of the server

  • !both lost when instnace is terminated

  • for persistency: use EFS, EBS (-> elasitc block storage, elastice file storage)

• elastic compute cloud

• elastic IP address

• static ip address required if:

  • use instance that must always be accessible by the same IP address

• pay for address indepenedent of usage

• account has limit for number of VMs of a certain type…

• pending

• running

• stopping

• stopped

• shutting down

• terminated

billed:

• running

• stopping (if perparing to hibernate)

not billed:

• pending

• stopping (if preparing to stop)

• stopped

• shitting-down

• terminated

• instance preparing to enter running

• -> pending when launched for the first time or when restarting after being stopped

• instance running and ready for use

• instnace preparing to be stopped or stop-hibernated

• instance shut down and cannot be used

• -> can be restarted at any time

• instance preparing to be terminated

• instance has been permanently deleted and cannot be restarted

EBS:

• usually < 1 min

ISB:

• usually < 5 min

EBS:

• 16 TiB

ISB:

• 10 GiB

EBS:

• amazon EBS volume

IBS:

• instance store volume

EBS:

• by default root volume deleted upon termination; on any other EBS volume persists after instance termination by default

ISB:

• data on any instance store only persists during life of the instance

EBS:

• instance type, kernel, RAM, upse data can be changed while instance is stopped

ISB:

• instance attributes fixed for lifetime of instance

EBS:

• charged for:

  • instance usage

  • EBS volume usage

  • storing AMI as amazon EBS snapshot

ISB:

• charged for:

  • instance usage

  • storing AMI in amazon S3

EBS:

• using single command/call

ISB:

• requires installation and use of AMI tools

EBS:

• can be placed in stopped state where instance is not running, but root volume persisted in amazon EBS

ISB:

• cannot be in stopped state!

• -> instances running or terminated

reboot:

• instance stays on same host computer

stop/start (EBS-baced only; no ISB)

• in most cases -> move to new host cmputer; may stay at same if there are no problems with host

hibernate (EBS-only)

• same as stop/start

terminate

• none (as instance will be no more…)

reboot:

• addresses stay the same

stop/start (EBS-baced only; no ISB)

• instance keep private IPv4

• get new public IPv4 (unless has elastic IP which doesnt change during stop/start

hibernate (EBS-only)

• same as stop/start

terminate

• none (as instance will be no more…)

reboot:

• elastic IP remains associated with instance

stop/start (EBS-baced only; no ISB)

• isame as reboot

hibernate (EBS-only)

• same as stop/start

terminate

• none (as instance will be no more…)

reboot:

• address stays the same

stop/start (EBS-baced only; no ISB)

• instance keeps address

hibernate (EBS-only)

• same as stop/start

terminate

• none (as instance will be no more…)

reboot:

• data preserved

stop/start (EBS-baced only; no ISB)

• data erased

hibernate (EBS-only)

• same as stop/start

terminate

• data erased

reboot:

• The volume is preserved

stop/start (EBS-baced only; no ISB)

• volume is preserved

hibernate (EBS-only)

• same as stop/start

terminate

• volume is deleted by default

reboot:

• RAM erased

stop/start (EBS-baced only; no ISB)

• RAM erased

hibernate (EBS-only)

• RAM saved to file on the root volume

terminate

• RAM erased

reboot:

• instance billing hour does not change

stop/start (EBS-baced only; no ISB)

• stop incurring charges as soon as it hits stopping

• each time instance transits from stopeed to running -> start new instance billing period (billing minimum of one minute every time you restart the instance)

hibernate (EBS-only)

• incur charges while instance in stopping state

terminate

• stop incurring charges as soon as it state changes to shutting down

• cluster placement gropu

• partition placement group

• spread placement group

• determine how instances are placed on the untderlying hardware

• => i.e. makes sense if you have several intercorrelated instances (that e.g. communicate with each other…)

• logical grouping of instances

• -> packed closely in availablity zone to increase network performance

• spread instances across partitions such that

  • different partitions do not share the underlying hardware

• each partition gets its own rack

• partitions can be placed in different availability zones

• reduce likelyhood of correlated hardware failures

  • -> and imporve performance in a partition

• spreads instances across distinct underlying hardware

• reduce correlated hardware failures

• accounts have their own virtual private cloud

• -> resources launced in this VPC

• => resembles your network in your own data center…

• Ip address range

• subnets

• route tables

• network gateways

• security settings

• connect instances to the internet

• conect your VPC to your data center

• amazon creates default VPC (for your account)

• additional VPCs can be created by the user

• primary means through web services API

• interactive tools on top of API

  • AWS concole

  • amazon command line tools

• third-party infrastructure tools

  • -> management of while infrastructure with multiple servers, accounts, reports, etc.

• private/public key pair

Allows to model your infrastructure

• inftastructure as code

• specify all resources in textual way as json template

• allows to standardize components accross your instituztion

• automatic deployment of all resources, controlled and predictable

• use code editor and versioning tools

• on-demand pricing

  • -> pay by second for instance you launch

• reserved instances pricing

  • make commitment to consistant configuration, including type and region

  • -> long term (1 or 3 years)

• pricing spot-market

  • request unused (avilablle) instances with lower current price

  • -> provdie max willing price

• internet:

  • in: free

  • out: <$0.09 per GB

• inside availability zone (private IP address)

  • none

• regional transfer (private IP address)

  • between different availability zones in same region

  • $0.01 per GB (in/out)

• public and elastic IP address inside EC2

  • $0.01 per GB (in/out)

• block storages

• snapshot to S3

• elastic iP address