Browse Prior Art Database

Method to improve performance using shared adapter among partition while handling Multicast Packets

IP.com Disclosure Number: IPCOM000012909D
Original Publication Date: 2003-Jun-09
Included in the Prior Art Database: 2003-Jun-09
Document File: 2 page(s) / 42K

Publishing Venue

IBM

Abstract

As logical partitioning becomes popular the trend is to create a high number of partitions. This leads to a situation where there may not be enough I/O slots for each partition and some partition would be required to share the I/O using a hosting partition. Typically the hosting partition will switch packets at the link level between the external network and the other partitions. The performance of the whole degrades when one or more of the partitions joins multicast groups.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 62% of the total text.

Page 1 of 2

  Method to improve performance using shared adapter among partition while handling Multicast Packets

As logical partitioning becomes popular the trend is to create a high number of partitions. This leads to a situation where there may not be enough I/O slots for each partition and some partition would be required to share the I/O using a hosting partition. Typically the hosting partition will switch packets at the link level between the external network and the other partitions. The performance of the whole degrades when one or more of the partitions joins multicast groups.

The present sharing mechanism does the following

The hosting partition puts the interface in promiscuous mode which enables all the packets coming to be handed over to the upper layer. When a multicast packet comes to the interface then all the partitions will receive this packet, only partitions which joined the group will hand over the packet to application and the rest will drop the packet at the IP layer.

The above method causes partitions which didn't join the multicast group to spend their CPU cycles to process a packet which will be dropped by the IP layer. This becomes even more critical in future partitioning technologies which supports CPU slicing and where large number of partitions will be sharing I/O slots. It can lead to a scenario where applications/kernel services will face CPU starvation due to significant loss of performance which will result in end-user satisfaction.

1

Page 2...