Many modern datacenters are scaling up dramatically, making network management a crucial yet challenging task. A datacenter network (DCN) that supports 100,000 servers would typically require upwards of thousands of switches and routers, referred to here as network equipment (NE).
At such scale, organizations achieve operational agility through division of labor, with
multiple network engineering groups managing different network aspects
(such as physical wiring and routingdesign).
Furthermore, as performance requirements — such as oversubscription
and fairshare bandwidth — change at each DCN layer, NE from multiple vendors (or multiple NE models from the same vendor) are deployed to achieve an
optimal price-to-performance ratio.
Achieving operational agility and an optimal price-to-performance ratio entail tradeoffs, however.
With each group managing only one aspect of the network, querying the network state incurs a communication overhead spanning multiple engineering groups.
Furthermore, device diversity leads to a network environment rife with
syntax/semantic disparities, and network engineers must grapple with managing the DCN using multiple configuration languages.
To put these two issues in perspective, consider the task of configuring peer-to-peer IP addresses between two routers (X and Y) with different configuration syntax.
To do this, network engineers must know the wiring map, to find all the interface pairs connecting X and Y and the IP subnet assignment, to know which addresses to configure onto each interface.
Because different groups have this information, engineers must make two.
Many modern datacenters are scaling up dramatically, making network management a crucial yet challenging task. A datacenter network (DCN) that supports 100,000 servers would typically require upwards of thousands of switches and routers, referred to here as network equipment (NE). At such scale, organizations achieve operational agility through division of labor, withmultiple network engineering groups managing different network aspects(such as physical wiring and routingdesign).Furthermore, as performance requirements — such as oversubscription and fairshare bandwidth — change at each DCN layer, NE from multiple vendors (or multiple NE models from the same vendor) are deployed to achieve anoptimal price-to-performance ratio. Achieving operational agility and an optimal price-to-performance ratio entail tradeoffs, however. With each group managing only one aspect of the network, querying the network state incurs a communication overhead spanning multiple engineering groups. Furthermore, device diversity leads to a network environment rife with syntax/semantic disparities, and network engineers must grapple with managing the DCN using multiple configuration languages. To put these two issues in perspective, consider the task of configuring peer-to-peer IP addresses between two routers (X and Y) with different configuration syntax. To do this, network engineers must know the wiring map, to find all the interface pairs connecting X and Y and the IP subnet assignment, to know which addresses to configure onto each interface. Because different groups have this information, engineers must make two.
การแปล กรุณารอสักครู่..
