[prev in list] [next in list] [prev in thread] [next in thread]
List: drbd-user
Subject: Re: [DRBD-user] drbd storage size
From: "Cesar Peschiera" <brain () click ! com ! py>
Date: 2014-10-27 13:50:48
Message-ID: FC9CCF4D86334DC8B422ED216A9EF835 () gerencia
[Download RAW message or body]
Hi Ivan
Many thanks for the warnings
I know about of the tuning for the RAID controller, DRBD, but not of cpu
pinning in KVM. I should see this topic.
If you have information for me about this topic, please let me know, i use
proxmox as Virtualization System (based on KVM and Openvz)
Best regards
Cesar Peschiera
----- Original Message -----
From: "Ivan" <ivan@c3i.bg>
To: "Cesar Peschiera" <brain@click.com.py>
Sent: Saturday, October 25, 2014 10:53 AM
Subject: Re: [DRBD-user] drbd storage size
> Hi Cesar,
>
> On 10/25/2014 04:17 PM, Cesar Peschiera wrote:
> > Hi Ivan
> >
> > Thanks for the link, it seem very interesting. In my case, each server
> > have
> > 2 Intel processors of 10 cores and 20 threads, according with this link:
> > http://ark.intel.com/products/75279/Intel-Xeon-Processor-E5-2690-v2-25M-Cache-3_00-GHz
> >
> >
> > But i have a question for do about of DRBD and MS-SQL Server 2008 x64, as
> > i
> > use KVM for virtualize a Windows Server with MS-SQL Server, and i only
> > have
> > this VM on the Server. What will be better for me in terms of
> > performance,
> > enable or not the threads of the processors?
>
> you mean hyperthreading ? I can't really advice on that, but from what
> I've read on forums it's OK to have it enabled (only the first versions of
> cpus with hyperthreading many years ago had problems).
>
> You should also investigate cpu pinning in KVM.
>
> That said I don't think you'll be CPU bound if all you run is a database
> server. The bottleneck will rather be I/O, so you'll have to understand
> kvm's caching strategy to make good use of the hefty setup you have. If
> your raid controllers have a battery backup unit (BBU) you should be safe
> with cache=none.
>
> Also, you'll have to use virtio drivers in windows.
>
> Have a look at that, it should help you.
>
> https://www.ibm.com/developerworks/community/blogs/sagitech/entry/tuning_kvm_guest_for_performance?lang=en
>
> good luck
> ivan
>
>
> >
> > Please, if you can explain me, do it clearly for that i can understand
> > you.
> >
> > Best regards
> > Cesar
> >
> >
> > ----- Original Message ----- From: "Ivan" <ivan@c3i.bg>
> > To: <drbd-user@lists.linbit.com>
> > Sent: Saturday, October 25, 2014 9:12 AM
> > Subject: Re: [DRBD-user] drbd storage size
> >
> >
> > > Hi,
> > >
> > > On 10/25/2014 02:38 PM, Cesar Peschiera wrote:
> > > > Hi Meji
> > > >
> > > > In three weeks i will have two Intel NIC X520-QDA1 of 40 Gb/s,
> > > > according
> > > > to these link:
> > > > http://ark.intel.com/products/68672/Intel-Ethernet-Converged-Network-Adapter-X520-QDA1
> > > >
> > > >
> > > > http://www.intel.com/content/www/us/en/network-adapters/converged-network-adapters/ethernet-x520-qda1-brief.html
> > > >
> > > >
> > >
> > > At those speeds it would be interesting to test the upcoming 3.18 kernel
> > > with the bulk network transmission patch [1] ; that should save you a
> > > bunch of cpu cycles.
> > >
> > > [1] http://lwn.net/Articles/615238
> > >
> > > ivan
> > >
> > >
> > > >
> > > > In my Hardware setup also i have a RAID controller H710p of Dell (LSI
> > > > chipset with 1 MB of cache) and with two groups of 4 HDDs SAS 15K RPM,
> > > > each group is configured in RAID 10, this setup is applied for each
> > > > Server (the HDDs for the OS are in other RAID), obviously i don't have
> > > > much storage compared to yours.
> > > >
> > > > In these Servers i will have running DRBD 8.4.5 version
> > > >
> > > > If you want to know the result of my tests, only let me know.
> > > >
> > > > Best regards
> > > > Cesar
> > > >
> > > > ----- Original Message ----- From: "Meij, Henk" <hmeij@wesleyan.edu>
> > > > To: <drbd-user@lists.linbit.com>
> > > > Sent: Thursday, October 23, 2014 5:10 PM
> > > > Subject: Re: [DRBD-user] drbd storage size
> > > >
> > > >
> > > > > a) it turns out the counter (8847740/11287100)M goes down, not up,
> > > > > deh, never noticed
> > > > >
> > > > > b) ran plain rsync across eth0 (public, with switches/routers) and
> > > > > eth1 (nic to nic)
> > > > > eth0 sent 585260755954 bytes received 10367 bytes 116690412.98
> > > > > bytes/sec
> > > > > eth1 sent 585260755954 bytes received 10367 bytes 122580535.41
> > > > > bytes/sec
> > > > > so my LSI raid card is behaving and DRBD is slowing the initialization
> > > > > down somehow.
> > > > > Found chapter 15 and will try some suggestions but ideas welcome.
> > > > >
> > > > > c) for grins
> > > > > version: 8.4.5 (api:1/proto:86-101)
> > > > > GIT-hash: 1d360bde0e095d495786eaeb2a1ac76888e4db96 build by
> > > > > mockbuild@Build64R6, 2014-08-17 19:26:04
> > > > > 0: cs:SyncTarget ro:Secondary/Secondary ds:Inconsistent/UpToDate C
> > > > > r-----
> > > > > ns:0 nr:3601728 dw:3601408 dr:0 al:0 bm:0 lo:4 pe:11 ua:3 ap:0 ep:1
> > > > > wo:f oos:109374215324
> > > > > [>....................] sync'ed: 0.1% (106810756/106814272)M
> > > > > finish: 731:23:05 speed: 41,532 (31,868) want: 41,000 K/sec
> > > > >
> > > > > 100TB in 731 hours would be 30 days. Can I expect large delta data
> > > > > replication to go equally slow using DRDB?
> > > > >
> > > > > -Henk
> > > > >
> > > > >
> > > > >
> > > > > ________________________________________
> > > > > From: drbd-user-bounces@lists.linbit.com
> > > > > [drbd-user-bounces@lists.linbit.com] on behalf of Meij, Henk
> > > > > [hmeij@wesleyan.edu]
> > > > > Sent: Thursday, October 23, 2014 9:57 AM
> > > > > To: Philipp Reisner; drbd-user@lists.linbit.com
> > > > > Subject: Re: [DRBD-user] drbd storage size
> > > > >
> > > > > Thanks for the write up y'll. I'll have to think about #3 not sure I
> > > > > grasp it fully.
> > > > >
> > > > > Last night I started a 12 TB test and started first initialization for
> > > > > observation (0 is primary).
> > > > > I have node0:eth1 wired directly into node1:eth1 with 10 foot CAT 6
> > > > > cable (MTU=9000)
> > > > > Data from node1 to node0
> > > > > PING 10.10.52.232 (10.10.52.232) 8970(8998) bytes of data.
> > > > > 8978 bytes from 10.10.52.232: icmp_seq=1 ttl=64 time=0.316 ms
> > > > >
> > > > > This morning's progress report from node1:(drbd v8.4.5)
> > > > >
> > > > > [===>................] sync'ed: 21.7% (8847740/11287100)M
> > > > > finish: 62:50:50 speed: 40,032 (39,008) want: 68,840 K/sec
> > > > >
> > > > > which confuses me: 8.8M out of 11.3M is 77.8% synced, not? I will let
> > > > > this test finish before I do a dd attempt.
> > > > >
> > > > > iostat reveals %idle cpu 99%+ and little to no %iowait (near 0%),
> > > > > iotop confirms very little IO (<5 K/s), typical data
> > > > > Device: rrqm/s wrqm/s r/s w/s rsec/s wsec/s
> > > > > avgrq-sz avgqu-sz await svctm %util
> > > > > sdb1 0.00 231.00 0.00 156.33 0.00 79194.67
> > > > > 506.58 0.46 2.94 1.49 23.37
> > > > >
> > > > > Something is throttling this IO as 40M/s is about half of what I was
> > > > > hoping for. Will dig some more.
> > > > >
> > > > > -Henk
> > > > >
> > > > > ________________________________________
> > > > > From: drbd-user-bounces@lists.linbit.com
> > > > > [drbd-user-bounces@lists.linbit.com] on behalf of Philipp Reisner
> > > > > [philipp.reisner@linbit.com]
> > > > > Sent: Thursday, October 23, 2014 9:17 AM
> > > > > To: drbd-user@lists.linbit.com
> > > > > Subject: Re: [DRBD-user] drbd storage size
> > > > >
> > > > > Am Donnerstag, 23. Oktober 2014, 08:55:03 schrieb Digimer:
> > > > > > On 23/10/14 04:00 AM, Philipp Reisner wrote:
> > > > > > > 2a) Initialize both backend devices to a known state.
> > > > > > >
> > > > > > > I.e. dd if=/dev/zero of=/dev/sdb1 bs=$((1024*1024))
> > > > > > oflag=direct
> > > > > >
> > > > > > Question;
> > > > > >
> > > > > > What I've done in the past to speed up initial sync is to create
> > > > > > the
> > > > > > DRBD device, pause-sync, then do your 'dd if=/dev/zero ...' trick to
> > > > > > /dev/drbd0. This effectively drives the resync speed to the max
> > > > > > possible
> > > > > > and ensures full sync across both nodes. Is this a sane approach?
> > > > > >
> > > > >
> > > > > Yes, sure that is a way to do it. (I have the impression that is
> > > > > something
> > > > > form the drbd-8.3 world.)
> > > > >
> > > > > I do not know from the top of the head if that will be faster than the
> > > > > built-in background resync in drbd-8.4.
> > > > >
> > > > > Best,
> > > > > Phil
> > > > >
> > > > > _______________________________________________
> > > > > drbd-user mailing list
> > > > > drbd-user@lists.linbit.com
> > > > > http://lists.linbit.com/mailman/listinfo/drbd-user
> > > > > _______________________________________________
> > > > > drbd-user mailing list
> > > > > drbd-user@lists.linbit.com
> > > > > http://lists.linbit.com/mailman/listinfo/drbd-user
> > > > > _______________________________________________
> > > > > drbd-user mailing list
> > > > > drbd-user@lists.linbit.com
> > > > > http://lists.linbit.com/mailman/listinfo/drbd-user
> > > > >
> > > >
> > > > _______________________________________________
> > > > drbd-user mailing list
> > > > drbd-user@lists.linbit.com
> > > > http://lists.linbit.com/mailman/listinfo/drbd-user
> > > _______________________________________________
> > > drbd-user mailing list
> > > drbd-user@lists.linbit.com
> > > http://lists.linbit.com/mailman/listinfo/drbd-user
> > >
> >
>
_______________________________________________
drbd-user mailing list
drbd-user@lists.linbit.com
http://lists.linbit.com/mailman/listinfo/drbd-user
[prev in list] [next in list] [prev in thread] [next in thread]
Configure |
About |
News |
Add a list |
Sponsored by KoreLogic