Kate S'jet
The Scope Gallente Federation
4
|
Posted - 2016.02.24 12:42:38 -
[1] - Quote
Good Day CCP,
I was very intrigued and i couldn't but help have a smile from ear to ear when i read your blog (http://community.eveonline.com/news/dev-blogs/tranquility-tech-3/) It made me remember the exiting times we had, back in 2007 when we moved to a blade architecture with a central storage unit all connected by a SAN. I know how cool and exciting this is for the OPS team, i beat they are all walking around high-fiving each other on the great work they have accomplished....So they should! Management, as one manager to another give those guys a pat on the back for the great work they have accomplished. I'm sure you have.
Business end of the stick now; While i'm sure CCP has invested a great amount of money and effort, please make sure your engineers are kept well trained as you move into virtualization. Your DB's will have new issues to contend with and will need all the support the virtual engineers will and can provide. MS SQL is a beast when fine tuned to operate in a VM environment. If the DBA's are not familiar with SQL in a VM environment, performance will not be optimised to its full potential and while everyone will observe performance improvements that can very easily be degrade with contention and storage IOPS challenges. Just cause your running SSD drive tier I've been able to degrade 200,000 IOPS SSD based storage over 120 drives with almost 2 servers. vCOPS doesn't always tell you the absolute truth, your SE's will have to trawl through the bowels of Windows Server 2012R2 to find issues, and 9 times out of 10 they will be human issues. NOT disrespecting anyone but when parts of the environment are only considered the entire environment will work against you. If you have RH servers, either legacy or latest version please, please make sure that THP is not enabled by accident on any high demand IOPS server. Can be web server running linux/unix specific apps such as load balancers (more on that later). The /var directory should be careful reviewed as this can cause IOPS issues. Also if you have DB's on RH i beg you to review those servers. If there on don't even think and turn them off. Your server will double in performance "instantly". But trawl through the web and search RH huge page.... Load Balancers (LB) best and i mean the best LB on the market is BSD LB, yes you know it Berkley LB .... a Unix LB. And to all the managers it's FREEEEEE!!!!!! We have 6 in 3 Data centres geographically dispersed and they do not skip a beat. These things are rock solid. If you bought any thing else....shame on you for throwing good many away. Even the F5's which are meant to be the bees-knees can only do half of the functionality of what these can do. And guess what, they run perfectly as a vm, No ISSUE!
Something to watch out for all those non virtual geeks. Be careful when allocating more then 4 vCPU to a server, cause you are trading speed for size. The more vCPU's you allocate to a vm from a host the longer each and every command has to wait to have that number of physical CPU's free/idol to be allocated. Don't believe me? Go do your VCP, VCA, VCD and if they don't talk about this then walk out of the training course.
Last but not least, i must admit the latency upgrade did seem a bit over the top. Why am i being so harsh, i do apologies but latency is directly correlated to distance, not necessary hops only. So me in Australia will experience 300ms of latency ALWAYS, unless you bring the server to Australia at least, then i will experience 20ms anywhere within the country. Oh boy can i but only dream!
To management and OPS team, close your eyes. CISCO is no longer King, it was in 80's, 90's, 2k, 2008 at max. But since then routing and switching has been taken over by Juniper, and Palo alto. Why cause they have changed there approach. First, from a direct first hand experience. We are about to upgrade all our 180 WAN sites with Juniper as Cisco has had to drop it's pricing by 90% from RRP to compete. I AM NOT JOKING. But go ahead, compare the market and see. I am right! Both the new contenders out perform in pure IOPS because they did not build a device that is meant to do everything as the Cisco iOS, it is antiquated and the coding is poor. If your Cisco engineer says I will quiet before i install a Juniper. Here is my response before i would show them the door. 90% of the JUNOS is the same as the Cisco iOS. They have made it similar to cater for the stubborn people. Juniper has out performed every bench mark that has been made against Cisco. Why you ask cause there both dealing with zero's and one's. It's easy, i mentioned it before....coding, coding. Remember as well Cisco iOS is proprietary, JUNOS is "LINUX" well ok i lied.... BSD!...oh s..t that dam Unix OS again. Sorry but management, this time it's not free. If you think your 120Gbps backplane is a serious contender, I'm going to guess you paid $80K+, if you didn't you got a second hand one :p check out the "small" tier Juniper switches EX4200 or the EX4600 there backplane starts at 128Gbps and there only AU$12k. MOVE AWAY FROM CISCO, you are only paying for the name! Happy to chat some more if you like.. Fly safe boyz and girlz...
|