From mboxrd@z Thu Jan 1 00:00:00 1970 From: Loic Dachary Subject: Re: EC backend benchmark Date: Tue, 12 May 2015 00:02:27 +0200 Message-ID: <555126F3.1050607@dachary.org> References: <755F6B91B3BE364F9BCA11EA3F9E0C6F2CD8A41B@SACMBXIP01.sdcorp.global.sandisk.com> Mime-Version: 1.0 Content-Type: multipart/signed; micalg=pgp-sha1; protocol="application/pgp-signature"; boundary="ITAbtoFL2s4uu3Owue2QeUn9JxFTAE8Gv" Return-path: Received: from mail2.dachary.org ([91.121.57.175]:33243 "EHLO smtp.dmail.dachary.org" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1751693AbbEKWCa (ORCPT ); Mon, 11 May 2015 18:02:30 -0400 In-Reply-To: <755F6B91B3BE364F9BCA11EA3F9E0C6F2CD8A41B@SACMBXIP01.sdcorp.global.sandisk.com> Sender: ceph-devel-owner@vger.kernel.org List-ID: To: Somnath Roy Cc: "ceph-users@lists.ceph.com" , Ceph Development This is an OpenPGP/MIME signed message (RFC 4880 and 3156) --ITAbtoFL2s4uu3Owue2QeUn9JxFTAE8Gv Content-Type: text/plain; charset=windows-1252 Content-Transfer-Encoding: quoted-printable Hi, [Sorry I missed the body of your questions, here is my answer ;-] On 11/05/2015 23:13, Somnath Roy wrote:> Summary : >=20 > ------------- >=20 > =20 >=20 > 1. It is doing pretty good in Reads and 4 Rados Bench clients are satur= ating 40 GB network. With more physical server, it is scaling almost line= arly and saturating 40 GbE on both the host. >=20 > =20 >=20 > 2. As suspected with Ceph, problem is again with writes. Throughput wis= e it is beating replicated pools in significant numbers. But, it is not s= caling with multiple clients and not saturating anything. >=20 > =20 >=20 > =20 >=20 > So, my question is the following. >=20 > =20 >=20 > 1. Probably, nothing to do with EC backend, we are suffering because of= filestore inefficiencies. Do you think any tunable like EC stipe size (o= r anything else) will help here ? I think Mark Nelson would be in a better position that me to answer as he= has conducted many experiments with erasure coded pools. > 2. I couldn=92t make fault domain as =91host=92, because of HW limitati= on. Do you think will that play a role in performance for bigger k values= ? I don't see a reason why there would be a direct relationship between the= failure domain and the values of k. Do you have a specific example in mi= nd ? > 3. Even though it is not saturating 40 GbE for writes, do you think sep= arating out public/private network will help in terms of performance ? I don't think so. What is the bottleneck ? CPU or disk I/O ? Cheers --=20 Lo=EFc Dachary, Artisan Logiciel Libre --ITAbtoFL2s4uu3Owue2QeUn9JxFTAE8Gv Content-Type: application/pgp-signature; name="signature.asc" Content-Description: OpenPGP digital signature Content-Disposition: attachment; filename="signature.asc" -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.22 (GNU/Linux) iEYEARECAAYFAlVRJvMACgkQ8dLMyEl6F23kPQCgr+0HHZxdUTQtTIj7WqUV8lZ1 2WsAoJJukyfmoG0JMaJIj112Pn39AQ9T =ZHYq -----END PGP SIGNATURE----- --ITAbtoFL2s4uu3Owue2QeUn9JxFTAE8Gv--