=========================================== #opnfv-meeting: Weekly Technical Discussion =========================================== Meeting started by bh526r at 14:00:02 UTC. The full logs are available at http://ircbot.wl.linuxfoundation.org/meetings/opnfv-meeting/2018/opnfv-meeting.2018-12-13-14.00.log.html . Meeting summary --------------- * Roll Call (bh526r, 14:00:11) * Brandon Wick (bh526r, 14:00:18) * Trevor Cooper (bh526r, 14:00:25) * Trevor Bramwell (bramwelt, 14:00:29) * Cristina Pauna (bh526r, 14:00:54) * Cristina Pauna (CristinaPauna, 14:01:40) * David McBride (dmcbride_, 14:01:50) * Al Morton (bh526r, 14:05:18) * Infrastructure Evolution (bh526r, 14:05:42) * Al Morton (acm, 14:06:01) * Trevor summarized what has been discussed in email thread (bh526r, 14:06:01) * Trevor reviewed the background of looking into packet.net (bh526r, 14:06:32) * David further added that the reason of adding packet.net was because the Cisco server was not supported in LF lab in Portland (bh526r, 14:07:42) * Comments from email indicated that current Pharos lab (community lab) was under-utilized (bh526r, 14:08:23) * Rex Lee (mj_rex, 14:09:21) * If we roll Intel lab into LaaS, we will get all benefits to Intel lab and community (bh526r, 14:09:35) * Suggestion is to add Pharos lab in LaaS (bh526r, 14:10:10) * At beginning, although LaaS was not visioned to support the use case of spanning over multiple Pharos labs (bh526r, 14:11:13) * Trevor C understands that packet.net will replace all LF-hosted services (bh526r, 14:13:00) * Does it include all servers that Intel donated? (bh526r, 14:13:22) * David M and Trevor B clarified that packet.net will replace those 16 (or 13) Cisco UCS servers, because no one in LF lab was trained to support those hardware. We used to have support contract, but not any more because of budget. Thus those 16 (or 13) Cisco UCS server has not support (though Frank arranged some kind of ad-hoc support as a favor from some Cisco engineer). (bh526r, 14:15:15) * What will be remaining in LF lab? (bh526r, 14:16:14) * Answer is Intel servers (3 or 6), used for XCI (bh526r, 14:16:30) * Packet.net is VM based. The only option of baremetal is AWS, but it needs some kind of configuration of networking which may or may not work for OPNFV (bh526r, 14:19:19) * Trevor B and David M further indicated that we may have a good deal with packet.net that intends to support LFN, and replacing hardware could be free of charge. (bh526r, 14:20:17) * David M indicated that we can have a trial of packet.net for a couple of months, like test drive, for free. (bh526r, 14:20:47) * Trevor C thinks that if those 16 Cisco UCS server was replaced, we should compare those 16 UCS server with packet.net. (bh526r, 14:21:54) * David M indicated we had this comparison a couple of months ago (bh526r, 14:22:38) * Trevor C indicated that we should compared the support of UCS v.s. support of packet.net (bh526r, 14:23:17) * Current support of 6 Intel is very low. If we replace those 16 UCS server with 16 vanilla servers, the support will still be low. (bh526r, 14:24:19) * The cost of replacing 16 server is $300k, and support of packet.net is $300k for 18 months. So the break-even point is 18 months. And the cost of $19k per server is very expensive. Usual cost of each server is about $4k-5k (bh526r, 14:26:52) * David M clarified it is based on cost model in UNH-IOL lab (bh526r, 14:27:16) * Trevor C mentioned that many of UNH-IOL lab hardware was donated by Intel (bh526r, 14:27:43) * Trevor B agrees that we should have a clearer picture. (bh526r, 14:31:51) * David M, Trevor B and Trevor C work together to find out more accurate estimate of the cost v.s. support, and the break-even point. (bh526r, 14:35:19) * Once we have that information, it is easier to agree to a way to move forward (bh526r, 14:35:43) * The information should also cover the information such as capacity, networking options, etc., as much as possible, for evaluation purpose. (bh526r, 14:37:10) * David M mentioned that perhaps the best way of evaluation is to design a trial (bh526r, 14:39:46) * Jack Morgan (bh526r, 14:51:10) * Trevor C asked the question is if we use packet.net, is it feasible to do dynamically-allocated pods? (bh526r, 14:52:21) * Trevor B said it actually makes it easier to do dynamic pods (bh526r, 14:53:05) * AOB (bh526r, 15:03:59) * Meeting adjourned (bh526r, 15:04:08) Meeting ended at 15:04:12 UTC. People present (lines said) --------------------------- * bh526r (41) * dmcbride_ (3) * collabot` (3) * acm (1) * bramwelt (1) * CristinaPauna (1) * mj_rex (1) Generated by `MeetBot`_ 0.1.4